Add readme and files

README.md

@@ -0,0 +1,3 @@
+# Lexicap: Lex Fridman Podcast Whisper captions
+
+These are transcripts for Lex Fridman episodes, generated using OpenAI Whisper. This dataset was created by Dr. Andrej Karpathy.

vtt/episode_002_small.vtt

@@ -0,0 +1,1910 @@
+WEBVTT
+
+00:00.000 --> 00:06.000
+ As part of MIT course 6S099 on artificial general intelligence, I got a chance to sit down with
+
+00:06.000 --> 00:13.680
+ Kristof Koch, who is one of the seminal figures in neurobiology, neuroscience, and generally in
+
+00:13.680 --> 00:20.320
+ the study of consciousness. He is the president, the chief scientific officer of the Allen Institute
+
+00:20.320 --> 00:27.520
+ for Brain Science in Seattle. From 1986 to 2013, he was the professor at Caltech. Before that,
+
+00:27.520 --> 00:34.400
+ he was at MIT. He is extremely well cited, over a hundred thousand citations. His research,
+
+00:34.400 --> 00:40.080
+ his writing, his ideas have had big impact on the scientific community and the general public
+
+00:40.080 --> 00:44.560
+ in the way we think about consciousness, in the way we see ourselves as human beings.
+
+00:44.560 --> 00:49.120
+ He's the author of several books, The Quest for Consciousness and Your Biological Approach,
+
+00:49.120 --> 00:54.320
+ and a more recent book, Consciousness, Confessions of a Romantic Reductionist.
+
+00:54.320 --> 01:00.000
+ If you enjoy this conversation, this course, subscribe, click the little bell icon to make
+
+01:00.000 --> 01:05.360
+ sure you never miss a video. And in the comments, leave suggestions for any people you'd like to
+
+01:05.360 --> 01:10.480
+ see be part of the course or any ideas that you would like us to explore. Thanks very much,
+
+01:10.480 --> 01:16.320
+ and I hope you enjoy. Okay, before we delve into the beautiful mysteries of consciousness,
+
+01:16.320 --> 01:22.240
+ let's zoom out a little bit. And let me ask, do you think there's intelligent life out there in
+
+01:22.240 --> 01:27.920
+ the universe? Yes, I do believe so. We have no evidence of it, but I think the probabilities
+
+01:27.920 --> 01:33.440
+ are overwhelming in favor of it. Give me a universe where we have 10 to the 11 galaxies,
+
+01:33.440 --> 01:39.280
+ and each galaxy has between 10 to the 11, 10 to the 12 stars, and we know more stars have one or
+
+01:39.280 --> 01:47.360
+ more planets. So how does that make you feel? It still makes me feel special, because I have
+
+01:47.360 --> 01:54.480
+ experiences. I feel the world, I experience the world. And independent of whether there are the
+
+01:54.480 --> 02:00.480
+ creatures out there, I still feel the world and I have access to this world in this very strange,
+
+02:00.480 --> 02:07.840
+ compelling way. And that's the core of human existence. Now you said human, do you think
+
+02:07.840 --> 02:13.360
+ if those intelligent creatures are out there, do you think they experience their world?
+
+02:13.360 --> 02:18.080
+ Yes, they are evolved. If they are a product of natural evolution, as they would have to be,
+
+02:18.080 --> 02:22.400
+ they will also experience their own world. So consciousness isn't just a human, you're right,
+
+02:22.400 --> 02:28.320
+ it's much wider. It's probably, it may be spread across all of biology. We have, the only thing
+
+02:28.320 --> 02:33.040
+ that we have special is we can talk about it. Of course, not all people can talk about it.
+
+02:33.040 --> 02:38.320
+ Babies and little children can talk about it. Patients who have a stroke in the left
+
+02:38.320 --> 02:43.040
+ inferior frontal gyrus can talk about it. But most normal adult people can talk about it.
+
+02:43.040 --> 02:48.000
+ And so we think that makes us special compared to little monkeys or dogs or cats or mice or all
+
+02:48.000 --> 02:52.400
+ the other creatures that we share the planet with. But all the evidence seems to suggest
+
+02:52.400 --> 02:56.960
+ that they too experience the world. And so it's overwhelmingly likely that other aliens,
+
+02:56.960 --> 03:00.640
+ that aliens would also experience their world. Of course, differently, because they have a
+
+03:00.640 --> 03:04.000
+ different sensorium, they have different sensors, they have a very different environment.
+
+03:04.640 --> 03:11.360
+ But the fact that I would strongly suppose that they also have experiences. They feel pain and
+
+03:11.360 --> 03:17.600
+ pleasure and see in some sort of spectrum and hear and have all the other sensors.
+
+03:17.600 --> 03:21.040
+ Of course, their language, if they have one would be different. So we might not be able to
+
+03:21.040 --> 03:24.320
+ understand their poetry about the experiences that they have.
+
+03:24.320 --> 03:32.240
+ That's correct. Right. So in a talk, in a video, I've heard you mention Siputzo, a Daxhound that
+
+03:32.240 --> 03:37.360
+ you came up with, that you grew up with was part of your family when you were young. First of all,
+
+03:37.360 --> 03:46.320
+ you're technically a Midwestern boy. Technically. But after that, you traveled there on a bit,
+
+03:46.320 --> 03:50.800
+ hence a little bit of the accent. You talked about Siputzo, the Daxhound, having
+
+03:51.680 --> 03:57.440
+ these elements of humanness, of consciousness that you discovered. So I just wanted to ask,
+
+03:58.000 --> 04:03.520
+ can you look back in your childhood and remember when was the first time you realized you yourself,
+
+04:03.520 --> 04:09.600
+ sort of from a third person perspective, our conscious being, this idea of,
+
+04:11.280 --> 04:16.320
+ you know, stepping outside yourself and seeing there's something special going on here in my brain?
+
+04:17.840 --> 04:22.480
+ I can't really actually. It's a good question. I'm not sure I recall a discrete moment. I mean,
+
+04:22.480 --> 04:27.280
+ you take it for granted because that's the only world you know, right? The only world I know,
+
+04:27.280 --> 04:33.520
+ you know, is the world of seeing and hearing voices and touching and all the other things.
+
+04:33.520 --> 04:40.000
+ So it's only much later at early in my undergraduate days when I became, when I enrolled in physics
+
+04:40.000 --> 04:43.520
+ and in philosophy that I really thought about it and thought, well, this is really fundamentally
+
+04:43.520 --> 04:48.800
+ very, very mysterious. And there's nothing really in physics right now that explains this transition
+
+04:48.800 --> 04:54.560
+ from the physics of the brain to feelings. Where do the feelings come in? So you can look at the
+
+04:54.560 --> 04:58.800
+ foundational equation of quantum mechanics, general relativity, you can look at the period table of
+
+04:58.800 --> 05:05.360
+ the elements, you can look at the endless ATG seed chat in our genes and no way is consciousness.
+
+05:05.360 --> 05:11.120
+ Yet I wake up every morning to a world where I have experiences. And so that's the heart of the
+
+05:11.120 --> 05:19.040
+ ancient mind body problem. How do experiences get into the world? So what is consciousness?
+
+05:19.040 --> 05:25.680
+ Experience. Consciousness is any, any, any experience. Some people call it subjective
+
+05:25.680 --> 05:30.560
+ feelings, some people call it phenomenon, phenomenology, some people call it qualia,
+
+05:30.560 --> 05:34.480
+ their philosophy, but they all denote the same thing. It feels like something in the
+
+05:34.480 --> 05:40.640
+ famous word of the philosopher Thomas Nagel, it feels like something to be a bad or to be, you know,
+
+05:40.640 --> 05:47.440
+ an American or to be angry or to be sad or to be in love or to have pain.
+
+05:49.040 --> 05:54.480
+ And that is what experience is, any possible experience could be as mundane as just sitting
+
+05:54.480 --> 05:59.760
+ here in a chair could be as exalted as, you know, having a mystical moment, you know,
+
+05:59.760 --> 06:03.200
+ in deep meditation, those are just different forms of experiences.
+
+06:03.200 --> 06:10.080
+ Experience. So if you were to sit down with maybe the next skip a couple generations of
+
+06:10.080 --> 06:16.480
+ IBM Watson, something that won jeopardy, what is the gap? I guess the question is between Watson,
+
+06:18.000 --> 06:26.000
+ that might be much smarter than you than us than all any human alive, but may not have experience.
+
+06:26.000 --> 06:31.920
+ What is the gap? Well, so that's a big, big question that's occupied people for the last,
+
+06:32.640 --> 06:38.800
+ certainly last 50 years since we, you know, since the advent of birth of, of computers.
+
+06:38.800 --> 06:42.720
+ That's a question on Turing tried to answer. And of course, he did it in this indirect way
+
+06:42.720 --> 06:48.480
+ by proposing a test, an operational test. So, but that's not really that's, you know,
+
+06:48.480 --> 06:52.480
+ he tried to get it. What does it mean for a person to think? And then he had this test,
+
+06:52.480 --> 06:56.400
+ right? You lock him away, and then you have a communication with them. And then you try to,
+
+06:56.400 --> 07:00.400
+ to guess after a while whether that is a person or whether it's a computer system.
+
+07:00.400 --> 07:05.680
+ There's no question that now or very soon, you know, Alexa or Siri or, you know, Google now
+
+07:05.680 --> 07:10.720
+ will pass this test, right? And you can game it. But, you know, ultimately, certainly in your
+
+07:10.720 --> 07:15.440
+ generation, there will be machines that will speak with complete points that will remember
+
+07:15.440 --> 07:20.320
+ everything you ever said. They'll remember every email you ever had, like, like Samantha remember
+
+07:20.320 --> 07:25.520
+ in the movie, her snow question is going to happen. But of course, the key question is,
+
+07:25.520 --> 07:31.120
+ does it feel like anything to be Samantha in the movie? Does it feel like anything to be Watson?
+
+07:31.120 --> 07:38.240
+ And there one has to very, very strongly think there are two different concepts here that we
+
+07:38.240 --> 07:44.400
+ co mingle. There is a concept of intelligence, natural or artificial, and there is a concept of
+
+07:44.400 --> 07:48.880
+ consciousness of experience, natural or artificial. Those are very, very different things.
+
+07:49.520 --> 07:55.520
+ Now, historically, we associate consciousness with intelligence. Why? Because we live in a world
+
+07:55.520 --> 08:01.520
+ living aside computers of natural selection, where we're surrounded by creatures, either our own kin
+
+08:01.520 --> 08:06.880
+ that are less or more intelligent, or we go across species, some some are more adapted to
+
+08:06.880 --> 08:12.000
+ particular environment, others are less adapted, whether it's a whale or dog, or you go talk about
+
+08:12.000 --> 08:17.680
+ a permitium or a little worm, all right. And we see the complexity of the nervous system goes from
+
+08:17.680 --> 08:24.400
+ one cell to to a specialized cells to a worm that has three net that has 30% of its cells are nerve
+
+08:24.400 --> 08:29.920
+ cells, to creature like us or like a blue whale that has 100 billion even more nerve cells.
+
+08:29.920 --> 08:35.280
+ And so based on behavioral evidence and based on the underlying neuroscience, we believe that
+
+08:35.840 --> 08:41.440
+ as these creatures become more complex, they are better adapted to to their particular ecological
+
+08:41.440 --> 08:46.960
+ niche. And they become more conscious, partly because their brain grows. And we believe
+
+08:46.960 --> 08:52.080
+ consciousness unlike the ancient ancient people thought most almost every culture thought that
+
+08:52.080 --> 08:56.800
+ consciousness with intelligence has to do with your heart. And you still to see that today,
+
+08:56.800 --> 09:00.800
+ you see honey, I love you with all my heart. Yes. But what you should actually say is they
+
+09:00.800 --> 09:05.520
+ know honey, I love you with all my lateral hypothalamus. And for Valentine's Day, you should
+
+09:05.520 --> 09:11.040
+ give your sweetheart, you know, hypothalamus in piece of chocolate, not a heart shaped chocolate,
+
+09:11.040 --> 09:15.280
+ right. And so we still have this language, but now we believe it's a brain. And so we see brains
+
+09:15.280 --> 09:19.680
+ of different complexity. And we think, well, they have different levels of consciousness,
+
+09:19.680 --> 09:28.160
+ they're capable of different experiences. But now we confront a world where we know where we're
+
+09:28.160 --> 09:34.960
+ beginning to engineer intelligence. And it's radical unclear whether the intelligence we're
+
+09:34.960 --> 09:40.160
+ engineering has anything to do with consciousness and whether it can experience anything. Because
+
+09:40.160 --> 09:45.520
+ fundamentally, what's the difference? Intelligence is about function. Intelligence, no matter exactly
+
+09:45.520 --> 09:50.400
+ how you define it sort of adaptation to new environments, being able to learn and quickly
+
+09:50.400 --> 09:54.400
+ understand, you know, the setup of this and what's going on and who are the actors and what's
+
+09:54.400 --> 10:00.800
+ going to happen next. That's all about function. Consciousness is not about function. Consciousness
+
+10:00.800 --> 10:08.080
+ is about being. It's in some sense much fundamental. You can see folks, you can see this in several
+
+10:08.080 --> 10:13.600
+ cases, you can see it, for instance, in the case of the clinic, when you're dealing with patients
+
+10:13.600 --> 10:19.200
+ who are, let's say, had a stroke or had wear and traffic accident, etc. They're pretty much
+
+10:19.200 --> 10:24.960
+ immobile. Terri Schievo, you may have heard historically, she was a person here in the
+
+10:24.960 --> 10:29.680
+ 90s in Florida. Her heart stood still. She was reanimated. Then for the next 14 years,
+
+10:29.680 --> 10:33.600
+ she was what's called in a vegetative state. So there's thousands of people in the vegetative
+
+10:33.600 --> 10:37.840
+ state. So they're, you know, they're, you know, they're like this. Occasionally, they open their
+
+10:37.840 --> 10:42.160
+ eyes for two, three, four, five, six, eight hours and then close their eyes. They have sleep wake cycle.
+
+10:42.160 --> 10:48.720
+ Occasionally, they have behaviors. They do like, you know, they, but there's no way that you can
+
+10:48.720 --> 10:53.200
+ establish a lawful relationship between what you say, or the doctor says, or the mom says,
+
+10:53.200 --> 11:00.000
+ and what the patient does. Right. So, so the, so the, there isn't any behavior yet in some of
+
+11:00.000 --> 11:06.480
+ these people, there is still experience. You can, you can design and build brain machine interfaces
+
+11:06.480 --> 11:10.400
+ where you can see there's, they still explain something. And of course, that these cases
+
+11:10.400 --> 11:15.040
+ are blocked in state. There's this famous book called the, the, the diving bell in the butterfly
+
+11:15.040 --> 11:20.160
+ where you had an editor, a French editor, he had a stroke in the, in the brainstem, unable to move
+
+11:20.160 --> 11:25.920
+ except his vertical eyes, eye movement. He could just move his eyes up and down. You need dictated
+
+11:25.920 --> 11:31.760
+ in an entire book. And some people even lose this at the end. And all the evidence seems to suggest
+
+11:31.760 --> 11:37.120
+ that they're still in there. In this case, you have no behavior. You have consciousness.
+
+11:37.120 --> 11:42.080
+ Second case is tonight, like all of us, you're going to go to sleep, close your eyes, you go to
+
+11:42.080 --> 11:46.720
+ sleep. You will wake up inside your sleeping body and you will have conscious experiences.
+
+11:47.440 --> 11:52.000
+ They're different from everyday experience. You might fly. You might not be surprised that you're
+
+11:52.000 --> 11:57.600
+ flying. You might meet a long dead pet, childhood dog, and you're not surprised that you're meeting
+
+11:57.600 --> 12:01.280
+ them, you know, but you have conscious experience of love, of hate, you know, they can be very
+
+12:01.280 --> 12:07.680
+ emotional. Your body doing this state, typically to them, state sends an active signal to your
+
+12:07.680 --> 12:13.120
+ motor neurons to paralyze you. It's called atonia, right? Because if you don't have that, like some
+
+12:13.120 --> 12:17.200
+ patients, what do you do? You act out your dreams, you get for example, rem behavioral disorder,
+
+12:17.200 --> 12:23.440
+ which is the bad, which is bad juju to get. Okay. Third case is pure experience. So I recently had
+
+12:23.440 --> 12:29.600
+ this, what some people call a mystical experience, I went to Singapore and went into a flotation
+
+12:29.600 --> 12:36.960
+ tank. Yeah. All right. So this is a big tub filled with water, that's body temperature and absent
+
+12:36.960 --> 12:42.400
+ salt. You strip completely naked, you lie inside of it, you close the, the, the darkness, complete
+
+12:42.400 --> 12:48.480
+ darkness, soundproof. So very quickly, you become bodyless because you're floating and you're naked.
+
+12:48.480 --> 12:53.600
+ You have no rings, no watch, no nothing. You don't feel your body anymore. It's no sound,
+
+12:53.600 --> 13:00.320
+ soundless. There's no photon, a sightless, timeless, because after a while, early on,
+
+13:00.320 --> 13:05.360
+ you actually hear your heart, but then that you, you sort of adapt to that and then sort of the
+
+13:05.360 --> 13:11.360
+ passage of time ceases. And if you train yourself like in a, in a meditation, not to think early
+
+13:11.360 --> 13:15.200
+ on, you think a lot, it's a little bit spooky. You feel somewhat uncomfortable or you think,
+
+13:15.200 --> 13:20.320
+ well, I'm going to get bored. But if you try to not to think actively, you become mindless.
+
+13:20.320 --> 13:27.520
+ There you are, bodyless, timeless, soundless, sightless, mindless, but you're in a conscious
+
+13:27.520 --> 13:32.560
+ experience. You're not asleep. You're not asleep. You are, you are being of pure,
+
+13:32.560 --> 13:36.400
+ you're pure being. There isn't any function. You aren't doing any computation. You're not
+
+13:36.400 --> 13:40.400
+ remembering, you're not projecting, you're not planning, yet you are fully conscious.
+
+13:40.400 --> 13:44.400
+ You're fully conscious. There's something going on there. It could be just a side effect. So what
+
+13:44.400 --> 13:52.480
+ is the, the, um, you mean epiphenomena? So what's the select, meaning why, uh, why, what, what,
+
+13:52.480 --> 13:59.200
+ what is the function of you being able to lay in this, uh, sense sensory free deprivation tank
+
+13:59.200 --> 14:04.640
+ and still have a conscious experience? Evolutionary. Obviously we didn't evolve with floatation
+
+14:04.640 --> 14:10.480
+ tanks in our, in our environment. I mean, so biology is not totally bad at asking why question to
+
+14:10.480 --> 14:14.560
+ leonormical question. Why do we have two eyes? Why don't we have four eyes like some teachers or three
+
+14:14.560 --> 14:19.360
+ eyes or something? Well, no, there's probably, there is a function to that, but it's, we're not
+
+14:19.360 --> 14:23.200
+ very good at answering those questions. We can speculate. And Leslie, where biology is very,
+
+14:23.200 --> 14:26.800
+ or science is very good about mechanistic question. Why is there charge in the universe,
+
+14:26.800 --> 14:31.040
+ right? We find a certain universe where there are positive and negative charges. Why? Why does
+
+14:31.040 --> 14:36.320
+ quantum mechanics hold? You know, why, why, why doesn't some other theory hold quantum mechanics
+
+14:36.320 --> 14:41.040
+ hold in our universe? It's very unclear why. So telenomical question, why questions are difficult
+
+14:41.040 --> 14:46.480
+ to answer. Clearly there's some relationship between complexity, brain processing power and
+
+14:46.480 --> 14:53.280
+ consciousness. But however, in these cases, in the three examples I gave, one is an everyday
+
+14:53.280 --> 14:57.920
+ experience at night. The other one is a trauma and third one is in principle, you can, everybody
+
+14:57.920 --> 15:04.080
+ can have these sort of mystical experiences. You have a dissociation of function from, of
+
+15:04.080 --> 15:12.320
+ intelligence from, from conscious consciousness. You caught me asking a white question. Let me
+
+15:12.320 --> 15:17.120
+ ask a question that's not a white question. You're giving a talk later today on the touring test
+
+15:17.760 --> 15:23.200
+ for intelligence and consciousness drawn lines between the two. So is there a scientific way
+
+15:23.200 --> 15:30.560
+ to say there's consciousness present in this entity or not? And to anticipate your answer,
+
+15:30.560 --> 15:35.840
+ because you, you will also, there's a neurobiological answer. So we can test a human brain. But if you
+
+15:35.840 --> 15:42.240
+ take a machine brain that you don't know tests for yet, how would you even begin to approach
+
+15:42.960 --> 15:47.760
+ a test if there's consciousness present in this thing? Okay, that's a really good question. So
+
+15:47.760 --> 15:54.000
+ let me take in two steps. So as you point out for, for, for humans, let's just stick with humans,
+
+15:54.000 --> 15:58.640
+ there's now a test called a zap and zip is a procedure where you ping the brain using
+
+15:58.640 --> 16:04.160
+ transcranial magnetic stimulation. You look at the electrical reverberations essentially using EG
+
+16:04.960 --> 16:08.800
+ and then you can measure the complexity of this brain response. And you can do this in awake
+
+16:08.800 --> 16:13.680
+ people in asleep, normal people, you can do it in awake people and then anesthetize them. You
+
+16:13.680 --> 16:20.240
+ can do it in patients and it has 100% accuracy that in all those cases, when you're clear,
+
+16:20.240 --> 16:24.080
+ the patient or the person is either conscious or unconscious, the complexity is either high or
+
+16:24.080 --> 16:28.320
+ low. And then you can adopt these techniques to similar creatures like monkeys and dogs and,
+
+16:28.320 --> 16:34.400
+ and mice that have very similar brains. Now, of course, you point out that may not help you
+
+16:34.400 --> 16:39.040
+ because we don't have a cortex, you know, and if I send a magnetic pulse into my iPhone or my
+
+16:39.040 --> 16:43.600
+ computer, it's probably going to break something. So we don't have that. So what we need ultimately,
+
+16:45.120 --> 16:50.080
+ we need a theory of consciousness. We can't just rely on our intuition. Our intuition is, well,
+
+16:50.080 --> 16:55.040
+ yeah, if somebody talks, they're conscious. However, then they're all these page children,
+
+16:55.040 --> 17:00.560
+ babies don't talk, right? But we believe that that the babies also have conscious experiences,
+
+17:00.560 --> 17:05.360
+ right? And then they're all these patients I mentioned. And they don't talk when you dream,
+
+17:05.360 --> 17:10.000
+ you can't talk because you're paralyzed. So, so what we ultimately need, we can't just rely
+
+17:10.000 --> 17:15.280
+ on our intuition, we need a theory of consciousness that tells us what is it about a piece of matter,
+
+17:15.280 --> 17:19.600
+ what is it about a piece of highly excitable matter like the brain or like a computer that
+
+17:19.600 --> 17:24.160
+ gives rise to conscious experience. We all believe none of us believe anymore in the old story,
+
+17:24.160 --> 17:28.160
+ it's a soul, right? That used to be the most common explanation that most people accept that
+
+17:28.160 --> 17:33.440
+ it's still a lot of people today believe, well, there's God in doubt, only us with a special
+
+17:33.440 --> 17:38.240
+ thing that animals don't have, René Descartes famously said, a dog, if you hit it with your
+
+17:38.240 --> 17:42.800
+ carriage, may yell, may cry, but it doesn't have this special thing. It doesn't have the magic,
+
+17:42.800 --> 17:47.920
+ the magic sauce. So yeah, it doesn't have red corketons, the soul. Now we believe that isn't
+
+17:47.920 --> 17:54.000
+ the case anymore. So what is the difference between brains and, and these guys, silicon.
+
+17:55.200 --> 18:01.120
+ And in particular, once their behavior matches. So if you have Siri of Alexa and 20 years from now
+
+18:01.120 --> 18:06.400
+ that she can talk just as good as any possible human, what grounds do you have to say she's not
+
+18:06.400 --> 18:11.280
+ conscious? In particular, if she says, it's of course she will. Well, of course I'm conscious.
+
+18:11.280 --> 18:15.600
+ You ask, how are you doing? And she'll say, well, you know, they'll generate some way to,
+
+18:15.600 --> 18:21.840
+ yeah, yeah, exactly. She'll behave like a, like a person. Now there's several differences. One is,
+
+18:23.280 --> 18:29.120
+ so this relates to the problem, the very heart. Why is consciousness a heart problem? It's because
+
+18:29.120 --> 18:35.280
+ it's subjective, right? Only I have it, for only I know, I have direct experience of my own
+
+18:35.280 --> 18:40.240
+ consciousness. I don't have experience, your consciousness. Now I assume as a sort of a
+
+18:40.240 --> 18:43.680
+ Bayesian person who believes in probability theory and all of that, you know, I can do,
+
+18:43.680 --> 18:48.160
+ I can do an abduction to the, to the best available facts. I deduce your brain is very
+
+18:48.160 --> 18:51.840
+ similar to mine. If I put you in a scanner, your brain is roughly going to behave the same with
+
+18:51.840 --> 18:56.240
+ I do. If, if, if, you know, if I give you this music and ask you, how does it taste?
+
+18:56.240 --> 19:00.240
+ Do you tell me things that, you know, that, that I would also say more or less, right?
+
+19:00.240 --> 19:03.920
+ So I infer based on all of that, that you're conscious. Now with Siri, I can't do that. So
+
+19:03.920 --> 19:09.680
+ there I really need a theory that tells me what is it about, about any system this or this that
+
+19:09.680 --> 19:14.880
+ makes it conscious. We have such a theory. Yes. So the, the integrated information theory.
+
+19:15.520 --> 19:19.360
+ But let me first, maybe it's introduction for people who are not familiar to car.
+
+19:20.640 --> 19:28.400
+ Can you, you talk a lot about panpsychism. Can you describe what physicalism versus dualism
+
+19:29.040 --> 19:35.280
+ this you mentioned the soul? What, what is the history of that idea? What is the idea of panpsychism?
+
+19:35.280 --> 19:44.560
+ Well, no, the debate really out of which panpsychism can emerge of, of, of dualism versus
+
+19:45.200 --> 19:49.200
+ physicalism. Or do you not see panpsychism as fitting into that?
+
+19:49.760 --> 19:53.840
+ No, you can argue there's some, well, okay, so let's step back. So panpsychism is a very
+
+19:53.840 --> 19:58.560
+ ancient belief that's been around. I mean, Plato and us totally talks about it.
+
+19:59.200 --> 20:05.040
+ Modern philosophers talk about it. Of course, in Buddhism, the idea is very prevalent that
+
+20:05.040 --> 20:08.960
+ I mean, there are different versions of it. One version says everything is in salt,
+
+20:08.960 --> 20:13.680
+ everything rocks and stones and dogs and people and forest and iPhones all have a soul.
+
+20:14.240 --> 20:20.240
+ All matter is in soul. That's sort of one version. Another version is that all biology,
+
+20:20.240 --> 20:26.080
+ all creatures, smaller, large from a single cell to a giant sequoia tree feel like something. That's
+
+20:26.080 --> 20:30.800
+ one I think is somewhat more realistic. So the different versions of what do you mean by feel
+
+20:30.800 --> 20:36.240
+ like something? Well, have, have feeling, have some kind of experience. It may well be possible
+
+20:36.240 --> 20:41.120
+ that it feels like something to be a paramedium. I think it's pretty likely it feels like something
+
+20:41.120 --> 20:48.320
+ to be a bee or a mouse or a dog. Sure. So, okay. So, so that you can say that's also,
+
+20:48.320 --> 20:53.920
+ so panpsychism is very broad, right? And you can, so some people, for example,
+
+20:53.920 --> 21:00.720
+ Bertrand Russell, try to advocate this, this idea is called Rasellian monism that, that
+
+21:00.720 --> 21:06.640
+ panpsychism is really physics viewed from the inside. So the idea is that physics is very
+
+21:06.640 --> 21:13.120
+ good at describing relationship among objects like charges or like gravity, right? You know,
+
+21:13.120 --> 21:17.360
+ describe the relationship between curvature and mass distribution. Okay. That's the relationship
+
+21:17.360 --> 21:21.440
+ among things. Physics doesn't really describe the ultimate reality itself. It's just
+
+21:21.440 --> 21:27.280
+ relationship among, you know, quarks or all these other stuff from like a third person observer.
+
+21:27.280 --> 21:33.440
+ Yes. Yes. And consciousness is what physics feels from the inside to my conscious experience.
+
+21:33.440 --> 21:37.760
+ It's the way the physics of my brain, particular my cortex feels from the inside.
+
+21:38.400 --> 21:42.400
+ And so if you are a paramedium, you got to remember, you say paramedium, well, that's a
+
+21:42.400 --> 21:48.960
+ pretty dumb creature. It is, but it has already a billion different molecules, probably, you know,
+
+21:48.960 --> 21:54.160
+ 5,000 different proteins assembled in a highly, highly complex system that no single person,
+
+21:54.160 --> 21:58.880
+ no computer system so far on this planet has ever managed to accurately simulate.
+
+21:58.880 --> 22:04.240
+ It's complexity vastly escapes us. Yes. And it may well be that that little thing feels like a
+
+22:04.240 --> 22:08.160
+ tiny bit. Now, it doesn't have a voice in the head like me. It doesn't have expectations.
+
+22:08.160 --> 22:12.400
+ You know, it doesn't have all that complex things, but it may well feel like something.
+
+22:12.400 --> 22:17.520
+ Yeah. So this is really interesting. Can we draw some lines and maybe try to understand
+
+22:17.520 --> 22:24.560
+ the difference between life, intelligence and consciousness? How do you see all of those?
+
+22:25.280 --> 22:31.120
+ If you have to define what is a living thing, what is a conscious thing and what is an intelligent
+
+22:31.120 --> 22:35.920
+ thing? Do those intermix for you or are they totally separate? Okay. So A, that's a question
+
+22:35.920 --> 22:40.000
+ that we don't have a full answer. Right. A lot of the stuff we're talking about today
+
+22:40.000 --> 22:44.240
+ is full of mysteries and fascinating ones, right? Well, you can go to Aristotle,
+
+22:44.240 --> 22:48.480
+ who's probably the most important scientist and philosopher who's ever lived in certainly
+
+22:48.480 --> 22:53.280
+ in Western culture. He had this idea. It's called hylomorphism. It's quite popular these days
+
+22:53.280 --> 22:58.080
+ that there are different forms of soul. The soul is really the form of something. He says,
+
+22:58.080 --> 23:02.480
+ all biological creatures have a vegetative soul. That's life principle. Today we think we
+
+23:02.480 --> 23:07.680
+ understand something more than this biochemistry in nonlinear thermodynamics. Right. Then he says
+
+23:07.680 --> 23:13.840
+ they have a sensitive soul. Only animals and humans have also a sensitive soul or a
+
+23:13.840 --> 23:18.960
+ petitive soul. They can see, they can smell and they have drives. They want to reproduce,
+
+23:18.960 --> 23:24.080
+ they want to eat, etc. Then only humans have what he called a rational soul.
+
+23:25.040 --> 23:29.840
+ Okay. Right. And that idea that made it into Christendom and then the rational soul is the one
+
+23:29.840 --> 23:34.240
+ that lives forever. He was very unclear. He wasn't really... I mean, different readings of Aristotle
+
+23:34.240 --> 23:39.040
+ give different... Did he believe that rational soul was immortal or not? I probably think he
+
+23:39.040 --> 23:43.040
+ didn't. But then, of course, that made it through play to into Christianity and then this soul
+
+23:43.040 --> 23:49.840
+ became immortal and then became the connection to God. Now, so you asked me, essentially,
+
+23:49.840 --> 23:56.240
+ what is our modern conception of these three... Aristotle would have called them different forms.
+
+23:56.240 --> 24:00.160
+ Life, we think we know something about it, at least life on this planet. Right. Although,
+
+24:00.160 --> 24:05.120
+ we don't understand how to originate it, but it's been difficult to rigorously pin down.
+
+24:05.120 --> 24:10.320
+ You see this in modern definitions of death. It's in fact, right now there's a conference
+
+24:10.320 --> 24:16.000
+ ongoing, again, that tries to define legally and medically what is death. It used to be very
+
+24:16.000 --> 24:19.920
+ simple. Death is, you stop breathing, your heart stops beating, you're dead. Right?
+
+24:19.920 --> 24:24.160
+ Totally unconventional. If you're unsure, you wait another 10 minutes. If the patient doesn't
+
+24:24.160 --> 24:28.640
+ breathe, you know, he's dead. Well, now we have ventilators, we have pacemakers. So,
+
+24:28.640 --> 24:33.360
+ it's much more difficult to define what death is. Typically, death is defined as the end of life
+
+24:33.360 --> 24:38.160
+ and life is defined before death. Thank you for that. Okay. So, we don't have really very good
+
+24:38.160 --> 24:42.640
+ definitions. Intelligence, we don't have a rigorous definition. We know something how to
+
+24:42.640 --> 24:49.920
+ measure. It's called IQ or G factors. Right. And we're beginning to build it in a narrow sense.
+
+24:49.920 --> 24:56.560
+ Right. Like, go AlphaGo and Watson and, you know, Google cars and Uber cars and all of that.
+
+24:56.560 --> 25:00.960
+ That's still narrow AI. And some people are thinking about artificial general intelligence.
+
+25:00.960 --> 25:05.120
+ But roughly, as we said before, it's something to do with the ability to learn and to adapt
+
+25:05.120 --> 25:10.160
+ to new environments. But that is, as I said, also its radical difference from experience.
+
+25:10.800 --> 25:16.480
+ And it's very unclear if you build a machine that has AGI, it's not at all a priori. It's
+
+25:16.480 --> 25:20.560
+ not at all clear that this machine will have consciousness. It may or may not.
+
+25:20.560 --> 25:25.280
+ So, let's ask it the other way. Do you think if you were to try to build an artificial general
+
+25:25.280 --> 25:31.040
+ intelligence system, do you think figuring out how to build artificial consciousness
+
+25:31.040 --> 25:39.360
+ would help you get to an AGI? Or put another way, do you think intelligent requires consciousness?
+
+25:40.320 --> 25:46.240
+ In human, it goes hand in hand. In human, or I think in biology, consciousness, intelligence
+
+25:46.240 --> 25:52.640
+ goes hand in hand. Quay is illusion because the brain evolved to be highly complex, complexity
+
+25:52.640 --> 25:58.160
+ via the theory integrated information theory is sort of ultimately is what is closely tied to
+
+25:58.160 --> 26:04.000
+ consciousness. Ultimately, it's causal power upon itself. And so, in evolved systems, they go
+
+26:04.000 --> 26:09.120
+ together. In artificial systems, particularly in digital machines, they do not go together.
+
+26:09.120 --> 26:16.800
+ And if you ask me point blank, is Alexa 20.0 in the year 2040, once she can easily pass every
+
+26:16.800 --> 26:21.600
+ Turing test that she conscious? No. Even if she claims she's conscious. In fact, you could even
+
+26:21.600 --> 26:25.840
+ do a more radical version of this thought experiment. We can build a computer simulation
+
+26:25.840 --> 26:30.320
+ of the human brain. You know what Henry Markham in the blue brain project or the human brain
+
+26:30.320 --> 26:34.240
+ project in Switzerland is trying to do. Let's grant them all the success. So in 10 years,
+
+26:34.240 --> 26:38.560
+ we have this perfect simulation of the human brain, every neuron is simulated. And it has
+
+26:38.560 --> 26:43.520
+ alarmics, and it has motor neurons, it has a blockers area. And of course, they'll talk and
+
+26:43.520 --> 26:48.480
+ they'll say, Hi, I just woken up. I feel great. Okay, even that computer simulation that can in
+
+26:48.480 --> 26:53.840
+ principle map onto your brain will not be conscious. Why? Because it simulates, it's a difference
+
+26:53.840 --> 26:59.280
+ between the simulated and the real. So it simulates the behavior associated with consciousness. It
+
+26:59.280 --> 27:03.920
+ might be, it will, if it's done properly, will have all the intelligence that that particular
+
+27:03.920 --> 27:09.760
+ person that simulating has. But simulating intelligence is not the same as having conscious
+
+27:09.760 --> 27:14.240
+ experiences. And I give you a really nice metaphor that engineers and physicists typically get.
+
+27:15.120 --> 27:20.000
+ I can write down Einstein's field equation nine or 10 equations that describe the link in general
+
+27:20.000 --> 27:27.600
+ relativity between curvature and mass. I can do that. I can run this on my laptop to predict that
+
+27:27.600 --> 27:34.960
+ the center, the black hole at the center of our galaxy will be so massive that it will twist space
+
+27:34.960 --> 27:40.240
+ time around it so no light can escape. It's a black hole, right? But funny, have you ever wondered
+
+27:40.240 --> 27:47.040
+ why doesn't this computer simulation suck me in? Right? It simulates gravity, but it doesn't have
+
+27:47.040 --> 27:52.720
+ the causal power of gravity. That's a huge difference. So it's a difference between the real
+
+27:52.720 --> 27:57.360
+ and the simulated, just like it doesn't get wet inside a computer when the computer runs code
+
+27:57.360 --> 28:03.360
+ that simulates a weather storm. And so in order to have to have artificial consciousness, you have
+
+28:03.360 --> 28:08.720
+ to give it the same causal power as the human brain. You have to build so called a neuromorphic
+
+28:08.720 --> 28:15.040
+ machine that has hardware that is very similar to the human brain, not a digital clock for
+
+28:15.040 --> 28:22.480
+ Neumann computer. So that's just to clarify, though, you think that consciousness is not required
+
+28:22.480 --> 28:29.360
+ to create human level intelligence. It seems to accompany in the human brain, but for machine
+
+28:29.360 --> 28:37.360
+ not. That's correct. So maybe just because this is AGI, let's dig in a little bit about what we
+
+28:37.360 --> 28:44.000
+ mean by intelligence. So one thing is the g factor of these kind of IQ tests of intelligence.
+
+28:44.000 --> 28:51.520
+ But I think if you maybe another way to say so in 2040, 2050, people will have Siri that is
+
+28:52.320 --> 28:57.600
+ just really impressive. Do you think people will say Siri is intelligent?
+
+28:57.600 --> 29:04.080
+ Yes. Intelligence is this amorphous thing. So to be intelligent, it seems like you have to have
+
+29:04.080 --> 29:09.520
+ some kind of connections with other human beings in the sense that you have to impress them with
+
+29:09.520 --> 29:17.200
+ your intelligence. And there feels you have to somehow operate in this world full of humans.
+
+29:17.200 --> 29:22.240
+ And for that, there feels like there has to be something like consciousness. So you think you
+
+29:22.240 --> 29:28.080
+ can have just the world's best natural NLP system, natural language, understanding a generation,
+
+29:28.080 --> 29:33.840
+ and that will be that will get us happy and say, you know what, we've created an AGI.
+
+29:33.840 --> 29:41.120
+ I don't know, happy. Yes, I do believe we can get what we call high level functional intelligence,
+
+29:41.120 --> 29:47.200
+ particularly sort of the G, you know, this fluid like intelligence that we charge,
+
+29:47.200 --> 29:52.640
+ particularly the place like MIT, right? In machines, I see a priori no reasons,
+
+29:52.640 --> 29:58.160
+ and I see a lot of reason to believe it's going to happen very over the next 50 years or 30 years.
+
+29:58.160 --> 30:04.160
+ So for beneficial AI, for creating an AI system that's, so you mentioned ethics,
+
+30:04.880 --> 30:10.880
+ that is exceptionally intelligent, but also does not do, does, you know, aligns its values
+
+30:10.880 --> 30:14.800
+ with our values of humanity. Do you think then it needs consciousness?
+
+30:14.800 --> 30:19.920
+ Yes, I think that that is a very good argument that if we're concerned about AI and the threat
+
+30:19.920 --> 30:26.160
+ of AI, like Nick Bostrom, existentialist threat, I think having an intelligence that has empathy,
+
+30:26.160 --> 30:32.480
+ right? Why do we find abusing a dog? Why do most of us find that abhorrent or abusing any animal?
+
+30:32.480 --> 30:37.600
+ Right? Why do we find that abhorrent? Because we have this thing called empathy, which if you
+
+30:37.600 --> 30:42.960
+ look at the Greek really means feeling with. I feel a pathos empathy. I have feeling with you.
+
+30:42.960 --> 30:48.400
+ I see somebody else suffer. That isn't even my conspecific. It's not a person. It's not a love.
+
+30:48.400 --> 30:53.280
+ It's not my wife or my kids. It's a dog. But I feel naturally, most of us, not all of us,
+
+30:53.280 --> 31:00.560
+ most of us will feel emphatic. And so it may well be in the long term interest of survival
+
+31:00.560 --> 31:05.840
+ of Homo sapiens sapiens, that if we do build AGI, and it's really becomes very powerful,
+
+31:05.840 --> 31:10.880
+ that it has an emphatic response and doesn't just exterminate humanity.
+
+31:11.760 --> 31:17.920
+ So as part of the full conscious experience to create a consciousness, artificial, or in our
+
+31:17.920 --> 31:24.320
+ human consciousness, do you think fear, maybe we're going to get into your earlier days with
+
+31:24.320 --> 31:30.560
+ Nietzsche and so on, but do you think fear and suffering are essential to have consciousness?
+
+31:30.560 --> 31:34.800
+ Do you have to have the full range of experience to have a system that has experience?
+
+31:36.480 --> 31:41.600
+ Or can you have a system that only has very particular kinds of very positive experiences?
+
+31:41.600 --> 31:46.880
+ Look, you can have, in principle, people have done this in a rat where you implant an electrode
+
+31:46.880 --> 31:51.680
+ in the hypothalamus, the pleasure center of the rat, and the rat stimulates itself above and
+
+31:51.680 --> 31:57.200
+ beyond anything else. It doesn't care about food or natural sex or drink anymore, it just stimulates
+
+31:57.200 --> 32:02.960
+ itself because it's such a pleasurable feeling. I guess it's like an orgasm, just you have all
+
+32:02.960 --> 32:11.040
+ day long. And so a priori, I see no reason why you need different, why you need a great variety.
+
+32:11.040 --> 32:16.800
+ Now, clearly to survive, that wouldn't work. But if I engineered artificially, I don't think
+
+32:18.720 --> 32:24.960
+ you need a great variety of conscious experience. You could have just pleasure or just fear.
+
+32:25.680 --> 32:30.240
+ It might be a terrible existence, but I think that's possible, at least on conceptual logical
+
+32:30.240 --> 32:34.640
+ ground. Because any real creature, whether artificial or engineered, you want to give
+
+32:34.640 --> 32:40.080
+ it fear, the fear of extinction that we all have. And you also want to give it a positive,
+
+32:40.080 --> 32:45.600
+ repetitive states, states that you want the machine encouraged to do because they give
+
+32:45.600 --> 32:52.000
+ the machine positive feedback. So you mentioned panpsychism to jump back a little bit.
+
+32:53.440 --> 33:00.400
+ Everything having some kind of mental property. How do you go from there to something like human
+
+33:01.120 --> 33:05.760
+ consciousness? So everything having some elements of consciousness. Is there something
+
+33:05.760 --> 33:12.320
+ special about human consciousness? Well, so just it's not everything like a spoon. There's no
+
+33:13.600 --> 33:18.160
+ the form of panpsychism, I think about doesn't ascribe consciousness to anything like this,
+
+33:18.160 --> 33:25.440
+ the spoon on my liver. However, it is the theory of integrated information theory does say that
+
+33:25.440 --> 33:29.680
+ system even ones that look from the outside relatively simple, at least if they have this
+
+33:29.680 --> 33:36.800
+ internal causal power, they are they does feel like something. The theory doesn't say anything
+
+33:36.800 --> 33:41.920
+ what's special about human. Biologically, we know what the one thing that's special about
+
+33:41.920 --> 33:48.800
+ human is we speak. And we have an overblown sense of our own importance. Right. We believe we're
+
+33:48.800 --> 33:54.800
+ exceptional. And we're just God's gift to to into the universe. But the but behaviorally,
+
+33:54.800 --> 33:58.720
+ the main thing that we have, we can plan, we can plan over the long term, we have language,
+
+33:58.720 --> 34:04.080
+ and that gives us enormous amount of power. And that's why we are the the condominant species
+
+34:04.080 --> 34:11.440
+ on the planet. So you mentioned God, you grew up a devout Roman Catholic, you know, Roman Catholic
+
+34:11.440 --> 34:19.360
+ family. So, you know, with consciousness, you're sort of exploring some really deeply fundamental
+
+34:19.360 --> 34:24.320
+ human things that religion also touches on. So where does, where does religion fit into your
+
+34:24.320 --> 34:29.760
+ thinking about consciousness? And you've you've grown throughout your life and changed your views
+
+34:29.760 --> 34:35.520
+ on religion, as far as I understand. Yeah, I mean, I'm not much closer to so I'm not a Roman
+
+34:35.520 --> 34:40.960
+ Catholic anymore. I don't believe there's sort of this God, the God I was, I was educated to
+
+34:40.960 --> 34:46.560
+ believe in, you know, sit somewhere in the fullness of time, I'll be united in some sort of everlasting
+
+34:46.560 --> 34:52.160
+ bliss. I just don't see any evidence for that. Look, the world, the night is large and full of
+
+34:52.160 --> 34:57.360
+ wonders, right? There are many things that I don't understand. I think many things that we as a
+
+34:57.360 --> 35:01.520
+ cult, you look, we don't even understand more than 4% of all the the universe, right? Dark
+
+35:01.520 --> 35:05.760
+ matter, dark energy. We have no idea what it is. Maybe it's lost socks. What do I know? So,
+
+35:06.480 --> 35:12.960
+ so all I can tell you is it's a sort of my current religious or spiritual sentiment is much closer
+
+35:12.960 --> 35:19.600
+ to some form of Buddhism. Can you just without the reincarnation? Unfortunately, there's no evidence
+
+35:19.600 --> 35:25.280
+ for reincarnation. So, can you describe the way Buddhism sees the world a little bit?
+
+35:25.280 --> 35:31.920
+ Well, so, you know, they talk about so when I spent several meetings with the Dalai Lama and what
+
+35:31.920 --> 35:36.720
+ always impressed me about him, he really unlike, for example, let's say the Pope or some Cardinal,
+
+35:36.720 --> 35:42.560
+ he always emphasized minimizing the suffering of all creatures. So, they have this from the early
+
+35:42.560 --> 35:47.520
+ beginning, they look at suffering in all creatures, not just in people, but in everybody, this
+
+35:47.520 --> 35:52.880
+ universal. And of course, by degrees, right in the animal, general will have less is less capable
+
+35:52.880 --> 35:59.520
+ of suffering than a well developed, normally developed human. And they think consciousness
+
+35:59.520 --> 36:05.440
+ pervades in this universe. And they have these techniques, you know, you can think of them
+
+36:05.440 --> 36:10.880
+ like mindfulness, etc. in meditation that tries to access sort of what they claim of this more
+
+36:10.880 --> 36:15.840
+ fundamental aspect of reality. I'm not sure it's more fundamentalist. I think about it. There's
+
+36:15.840 --> 36:20.240
+ a physical and then there's this inside view consciousness. And those are the two aspects
+
+36:20.240 --> 36:24.880
+ that the only thing I have access to in my life. And you got to remember my conscious
+
+36:24.880 --> 36:28.640
+ experience and your conscious experience comes prior to anything you know about physics,
+
+36:28.640 --> 36:33.680
+ comes prior to knowledge about the universe and atoms and super strings and molecules and all of
+
+36:33.680 --> 36:39.040
+ that. The only thing you directly are acquainted with is this world that's populated with things
+
+36:39.040 --> 36:42.720
+ and images and sounds in your head and touches and all of that.
+
+36:42.720 --> 36:49.120
+ I actually have a question. So it sounds like you kind of have a rich life. You talk about
+
+36:49.120 --> 36:55.040
+ rock climbing and it seems like you really love literature and consciousness is all about
+
+36:55.040 --> 37:00.160
+ experiencing things. So do you think that has helped your research on this topic?
+
+37:00.640 --> 37:05.440
+ Yes, particularly if you think about it, the various states of what I'm going to do rock
+
+37:05.440 --> 37:11.600
+ climbing. And now I do a rowing crew rowing and a bike every day, you can get into this thing called
+
+37:11.600 --> 37:16.320
+ the zone. And I've always I want to I wanted about a particular with respect to consciousness
+
+37:16.320 --> 37:21.120
+ because it's a strangely addictive state. You want to you want to appear. I mean, once people
+
+37:21.120 --> 37:25.360
+ have it once, they want to keep on going back to it. And you wonder why what is it so addicting
+
+37:25.360 --> 37:31.840
+ about it. And I think it's the experience of almost close to pure experience. Because in this
+
+37:31.840 --> 37:35.840
+ in this zone, you're not conscious or inner voice anymore. There's always an inner voice
+
+37:35.840 --> 37:38.960
+ nagging you, right? You have to do this, you have to do that, you have to pay your taxes,
+
+37:38.960 --> 37:42.960
+ you had this fight with your ex and all of those things are always there. But when you're in the
+
+37:42.960 --> 37:47.280
+ zone, all of that is gone. And you're just this in this wonderful state where you're fully out in
+
+37:47.280 --> 37:53.440
+ the world, right? You're climbing or you're rowing or biking or doing soccer, whatever you're doing.
+
+37:53.440 --> 37:59.280
+ And sort of consciousness sort of is this your all action, or in this case of pure experience,
+
+37:59.280 --> 38:06.160
+ you're not action at all. But in both cases, you experience some aspect of you touch some basic
+
+38:06.160 --> 38:13.600
+ part of conscious existence that is so basic and so deeply satisfying. You I think you touch the
+
+38:13.600 --> 38:18.320
+ root of being. That's really what you're touching there, you're getting close to the root of being.
+
+38:18.320 --> 38:24.400
+ And that's very different from intelligence. So what do you think about the simulation hypothesis,
+
+38:24.400 --> 38:28.320
+ simulation theory, the idea that we all live in a computer simulation? Have you even ever?
+
+38:28.320 --> 38:35.760
+ Rapture for nerds. Rapture for nerds. I think it's as likely as the hypothesis that
+
+38:35.760 --> 38:41.440
+ engaged hundreds of scholars for many centuries, are we all just existing in the mind of God?
+
+38:42.000 --> 38:46.320
+ Right. And this is just a modern version of it. It's it's it's equally plausible.
+
+38:47.280 --> 38:51.280
+ People love talking about these sorts of things. I know their book written about the simulation
+
+38:51.280 --> 38:56.480
+ hypothesis. If that's what people want to do, that's fine. It seems rather esoteric. It's never
+
+38:56.480 --> 39:02.240
+ testable. But it's not useful for you to think of in those terms. So maybe connecting to the
+
+39:02.240 --> 39:07.200
+ questions of free will, which you've talked about. I think I vaguely remember you saying
+
+39:07.200 --> 39:11.760
+ that the idea that there's no free will, it makes you very uncomfortable.
+
+39:13.280 --> 39:17.840
+ So what do you think about free will? And from the from a physics perspective,
+
+39:17.840 --> 39:22.400
+ from a consciousness perspective, what does it all fit? Okay, so from the physics perspective,
+
+39:22.400 --> 39:26.960
+ leaving aside quantum mechanics, we believe we live in a fully deterministic world, right?
+
+39:26.960 --> 39:30.880
+ But then comes, of course, quantum mechanics. So now we know that certain things are in principle
+
+39:30.880 --> 39:37.040
+ not predictable, which I, as you said, I prefer because the idea that at the initial condition
+
+39:37.040 --> 39:40.800
+ of the universe, and then everything else, we're just acting out the initial condition of the
+
+39:40.800 --> 39:47.120
+ universe that doesn't that doesn't make it's not a romantic notion. Certainly not. Right. Now,
+
+39:47.120 --> 39:52.400
+ when it comes to consciousness, I think we do have certain freedom. We are much more constrained by
+
+39:52.400 --> 39:57.040
+ physics, of course, and by our past and by our own conscious desires and what our parents told us
+
+39:57.040 --> 40:01.200
+ and what our environment tells us, we all know that, right? There's hundreds of experiments
+
+40:01.200 --> 40:06.480
+ that show how we can be influenced. But finally, in the in the final analysis, when you make a
+
+40:06.480 --> 40:10.560
+ life and I'm talking really about critical decision, what you really think, should I marry,
+
+40:10.560 --> 40:14.800
+ should I go to this school of that good, should I take this job with that job,
+
+40:14.800 --> 40:20.240
+ should I cheat on my taxes or not? These sort of these are things where you really deliberate.
+
+40:20.240 --> 40:25.200
+ And I think on those conditions, you are as free as you can be. When you when you bring your
+
+40:25.200 --> 40:33.360
+ entire being your entire conscious being to that question and try to analyze it on all the
+
+40:33.360 --> 40:37.920
+ the various conditions, then you take you make a decision you are as free as you can ever be.
+
+40:38.560 --> 40:44.160
+ That is I think what what free will is it's not a will that's totally free to do anything it wants.
+
+40:44.160 --> 40:50.880
+ That's not possible. Right. So as Jack mentioned, yet you actually read a blog about books you've
+
+40:50.880 --> 41:01.680
+ read amazing books from Russian from Bulgakov. Yeah, Neil Gaiman, Carl Sagan, Murakami. So what
+
+41:01.680 --> 41:07.360
+ is a book that early in your life transformed the way you saw the world, something that changed your
+
+41:07.360 --> 41:14.160
+ life? Nietzsche, I guess, did. That's broke out Trista, because he talks about some of these problems.
+
+41:14.160 --> 41:18.720
+ You know, he was one of the first discoverer of the unconscious. This is, you know, a little bit
+
+41:18.720 --> 41:26.240
+ before Freud when it was in the air. And you know, he makes all these claims that people sort of
+
+41:26.240 --> 41:33.360
+ under the guise of under the mass of charity actually are very non charitable. So he is sort
+
+41:33.360 --> 41:40.480
+ of really the first discoverer of the great land of the of the unconscious. And that that really
+
+41:40.480 --> 41:44.000
+ struck me. And what do you think? What do you think about the unconscious? What do you think
+
+41:44.000 --> 41:49.760
+ about Freud? What do you think about these ideas? What's what's just like dark matter in the universe?
+
+41:49.760 --> 41:54.560
+ What's over there in that unconscious? A lot. I mean, much more than we think, right? This is what
+
+41:54.560 --> 42:00.080
+ a lot of last 100 years of research has shown. So I think he was a genius, misguided towards the
+
+42:00.080 --> 42:04.800
+ end. But he was all he started out as a neuroscientist, right? He contributed. He did the studies on the
+
+42:06.000 --> 42:11.600
+ on the lamp. He contributed himself to the neuron hypothesis, the idea that they're discrete units
+
+42:11.600 --> 42:17.920
+ that we call nerve cells now. And then he started then he he wrote, you know, about the unconscious.
+
+42:17.920 --> 42:22.720
+ And I think it's true. There's lots of stuff happening. You feel this particular when you're
+
+42:22.720 --> 42:27.520
+ in a relationship and it breaks a thunder, right? And then you have this terrible, you can have
+
+42:27.520 --> 42:33.200
+ love and hate and lust and anger and all of it is mixed in. And when you try to analyze yourself,
+
+42:33.200 --> 42:40.080
+ why am I so upset? It's very, very difficult to penetrate to those basements, those caverns in
+
+42:40.080 --> 42:45.440
+ your mind, because the prying eyes of conscience doesn't have access to those. But they're there
+
+42:45.440 --> 42:50.240
+ in the amygdala or, you know, in lots of other places, they make you upset or angry or sad or
+
+42:50.240 --> 42:55.360
+ depressed. And it's very difficult to try to actually uncover the reason you can go to a shrink,
+
+42:55.360 --> 42:59.760
+ you can talk with your friend endlessly, you construct finally a story why this happened,
+
+42:59.760 --> 43:03.760
+ why you love her or don't love her or whatever. But you don't really know whether that's actually
+
+43:04.640 --> 43:08.400
+ that actually happened, because you simply don't have access to those parts of the brain. And
+
+43:08.400 --> 43:13.120
+ they're very powerful. Do you think that's a feature or a bug of our brain? The fact that we
+
+43:13.120 --> 43:19.120
+ have this deep, difficult to dive into subconscious? I think it's a feature, because otherwise,
+
+43:19.120 --> 43:28.480
+ look, we are like any other brain or nervous system or computer, we are severely band limited.
+
+43:28.480 --> 43:34.640
+ If we, if everything I do, every emotion I feel, every eye movements I make, if all of that had
+
+43:34.640 --> 43:41.520
+ to be under the control of consciousness, I couldn't, I couldn't, I wouldn't be here. Right.
+
+43:41.520 --> 43:46.880
+ So, so what you do early on your brain, you have to be conscious when you learn things like typing
+
+43:46.880 --> 43:52.640
+ or like riding on a bike. But then you what you do, you train up route, I think that involved
+
+43:52.640 --> 43:57.760
+ basal ganglia and stratum, you train up different parts of your brain. And then once you do it
+
+43:57.760 --> 44:01.600
+ automatically like typing, you can show you do it much faster without even thinking about it,
+
+44:01.600 --> 44:05.280
+ because you've got these highly specialized what Franz Krik and I called zombie agents
+
+44:06.160 --> 44:09.680
+ that I sort of they're taking care of that while your consciousness can sort of worry
+
+44:09.680 --> 44:14.640
+ about the abstract sense of the text you want to write. I think that's true for many, many things.
+
+44:14.640 --> 44:20.400
+ But for the things like all the fights you had with an ex girlfriend, things that
+
+44:20.960 --> 44:25.120
+ you would think are not useful to still linger somewhere in the subconscious.
+
+44:25.120 --> 44:29.600
+ So that seems like a bug that it would stick there. You think it would be better if you can
+
+44:29.600 --> 44:34.320
+ analyze and then get it out of the system or just forget it ever happened. You know,
+
+44:34.320 --> 44:39.520
+ that that seems a very buggy kind of. Well, yeah, in general, we don't have and that's
+
+44:39.520 --> 44:43.840
+ probably functional. We don't have an ability unless it's extreme, the outcases clinical
+
+44:43.840 --> 44:48.880
+ dissociations, right? When people are heavily abused when they completely repress them,
+
+44:48.880 --> 44:53.280
+ they the memory, but that doesn't happen in, in, in, you know, in normal people,
+
+44:53.280 --> 44:58.800
+ we don't have an ability to remove traumatic memories. And of course, we suffer from that.
+
+44:58.800 --> 45:03.680
+ On the other hand, probably if you have the ability to constantly wipe your memory,
+
+45:03.680 --> 45:10.160
+ you probably do it to an extent that isn't useful to you. So yeah, it's a good question.
+
+45:10.160 --> 45:15.760
+ It's a balance. So on the books, as Jack mentioned, correct me if I'm wrong, but
+
+45:16.640 --> 45:21.280
+ broadly speaking, academia and different scientific disciplines, certainly in engineering,
+
+45:21.920 --> 45:27.680
+ reading literature seems to be a rare pursuit. Perhaps I'm wrong in this, but that's in my
+
+45:27.680 --> 45:34.880
+ experience, most people read much more technical texts and do not sort of escape or seek truth
+
+45:34.880 --> 45:40.400
+ in literature. It seems like you do. So what do you think is the value? What do you think
+
+45:40.400 --> 45:46.000
+ literature adds to the pursuit of scientific truth? Do you think it's good? It's useful for
+
+45:47.120 --> 45:51.520
+ give you access to much wider array of human experiences?
+
+45:52.320 --> 45:54.000
+ How valuable do you think it is?
+
+45:54.000 --> 45:58.720
+ Well, if you want to understand human nature and nature in general, then I think you have to
+
+45:58.720 --> 46:04.800
+ better understand a wide variety of experiences, not just sitting in a lab staring at a screen and
+
+46:04.800 --> 46:09.200
+ having a face flashed onto you for 100 milliseconds and pushing a button. That's what I used to do.
+
+46:09.200 --> 46:13.280
+ That's what most psychologists do. There's nothing wrong with that, but you need to consider
+
+46:13.280 --> 46:18.800
+ lots of other strange states. And literature is a shortcut for this.
+
+46:18.800 --> 46:22.880
+ Well, yeah, because literature, that's what literature is all about. All sorts of interesting
+
+46:22.880 --> 46:28.640
+ experiences that people have, the contingency of it, the fact that women experience a world
+
+46:28.640 --> 46:33.840
+ different, black people experience a world different. One way to experience that is reading
+
+46:33.840 --> 46:38.320
+ all these different literature and try to find out. You see everything so relative. You read
+
+46:38.320 --> 46:42.320
+ the books 100 years ago, they thought about certain problems very, very differently than
+
+46:42.320 --> 46:47.120
+ us today. We today, like any culture, think we know it all. That's common to every culture.
+
+46:47.120 --> 46:50.960
+ Every culture believes that it's heyday, they know it all. And then you realize, well, there's
+
+46:50.960 --> 46:55.360
+ other ways of viewing the universe. And some of them may have lots of things in their favor.
+
+46:56.560 --> 47:03.200
+ So this is a question I wanted to ask about timescale or scale in general. When you, with
+
+47:03.200 --> 47:07.520
+ IIT or in general, try to think about consciousness, try to think about these ideas,
+
+47:08.880 --> 47:17.520
+ kind of naturally thinking human timescales. Do you or and also entities that are sized
+
+47:17.520 --> 47:21.680
+ close to humans? Do you think of things that are much larger, much smaller as containing
+
+47:21.680 --> 47:31.200
+ consciousness? And do you think of things that take, you know, well, ages, eons to operate
+
+47:31.200 --> 47:36.560
+ in their conscious cause effect? It's a very good question. So I think a lot of what small
+
+47:36.560 --> 47:42.400
+ creatures, because experimentally, a lot of people work on flies and bees. So most people
+
+47:42.400 --> 47:46.320
+ just think they're automata. They're just bugs, for heaven's sake. But if you look at their behavior,
+
+47:46.320 --> 47:50.080
+ like bees, they can recognize individual humans. They have this very complicated
+
+47:50.960 --> 47:54.880
+ way to communicate. If you've ever been involved, or you know, your parents, when they bought a
+
+47:54.880 --> 47:59.760
+ house, what sort of agonizing decision that is. And bees have to do that once a year, right?
+
+47:59.760 --> 48:03.200
+ When they swarm in the spring, and then they have this very elaborate way, they have three
+
+48:03.200 --> 48:07.360
+ nuts scouts, they go to the individual sites, they come back, they have this power, this dance,
+
+48:07.360 --> 48:11.520
+ literally where they dance for several days, they try to recruit other these very complicated
+
+48:11.520 --> 48:16.480
+ decision weight. When they finally want to make a decision, the entire swarm, the scouts warm up
+
+48:16.480 --> 48:20.320
+ the entire swarm, then go to one location, they don't go to 50 equation, they go to one location
+
+48:20.320 --> 48:24.800
+ that the scouts have agreed upon by themselves. That's awesome. If we look at the circuit complexity,
+
+48:24.800 --> 48:28.400
+ it's 10 times more denser than anything we have in our brain. Now they only have a million
+
+48:28.400 --> 48:32.640
+ neurons, but the neurons are amazingly complex, complex behavior, very complicated circuitry.
+
+48:32.640 --> 48:37.120
+ So there's no question, they experience something, their life is very different, they're tiny,
+
+48:37.120 --> 48:44.240
+ they only live for, well, workers live maybe for two months. So I think an IIT tells you this,
+
+48:44.240 --> 48:49.680
+ in principle, the substrate of consciousness is the substrate that maximizes the cause effect
+
+48:49.680 --> 48:54.640
+ power over all possible spatial temple grains. So when I think about, for example, do you know
+
+48:54.640 --> 48:59.840
+ the science fiction story, The Black Cloud? Okay, it's a classic by Fred Hoel, the astronomer.
+
+48:59.840 --> 49:07.120
+ He has this cloud intervening between the earth and the sun and leading to some sort of global
+
+49:07.120 --> 49:13.200
+ cooling that's written in the 50s. It turns out you can, using the radio dish, they communicate
+
+49:13.200 --> 49:18.240
+ with actually an entity, it's actually an intelligent entity. And they sort of, they
+
+49:18.240 --> 49:23.600
+ convince it to move away. So here you have a radical different entity. And in principle,
+
+49:23.600 --> 49:28.640
+ IIT says, well, you can measure the integrated information in principle at least. And yes,
+
+49:28.640 --> 49:34.640
+ if that, if the maximum of that occurs at a time scale of month, rather than an acid sort of fraction
+
+49:34.640 --> 49:40.720
+ of a second, yes, and they would experience life where each moment is a month rather than, or
+
+49:40.720 --> 49:47.600
+ microsecond, right, rather than a fraction of a second in the human case. And so there may be
+
+49:47.600 --> 49:51.760
+ forms of consciousness that we simply don't recognize for what they are, because they are so
+
+49:51.760 --> 49:56.960
+ radical different from anything you and I are used to. Again, that's why it's good to read
+
+49:56.960 --> 50:03.120
+ or to watch science fiction movie or to think about this. Like this is, do you know Stanislaw
+
+50:03.120 --> 50:07.520
+ Lem, this Polish science fiction writer, he wrote Solaris, it was turned into a Hollywood movie?
+
+50:07.520 --> 50:13.280
+ Yes. His best novel was in the 60s, a very, very ingenious, an ingenious background. His most
+
+50:13.280 --> 50:19.680
+ interesting novel is called The Victorious, where human civilization, they have this mission to
+
+50:19.680 --> 50:25.280
+ this planet and everything is destroyed and they discover machines, humans got killed and then
+
+50:25.280 --> 50:30.480
+ these machines took over and there was a machine evolution, a Darwinian evolution, he talks about
+
+50:30.480 --> 50:37.200
+ this very vividly. And finally, the dominant, the dominant machine intelligence organism that
+
+50:37.200 --> 50:42.960
+ survived are gigantic clouds of little hexagonal universal cell automata. This is written in the
+
+50:42.960 --> 50:48.320
+ 60s. So typically they're all lying on the ground individual by themselves, but in times of crisis
+
+50:48.320 --> 50:54.080
+ they can communicate, they assemble into gigantic nets, into clouds of trillions of these particles
+
+50:54.080 --> 50:59.760
+ and then they become hyper intelligent and they can beat anything that humans can throw at it.
+
+50:59.760 --> 51:05.040
+ It's a very beautiful and compelling way you have an intelligence where finally the humans
+
+51:05.040 --> 51:09.600
+ leave the planet, they simply unable to understand and comprehend this creature and they can say,
+
+51:09.600 --> 51:14.000
+ well, either we can nuke the entire planet and destroy it or we just have to leave because
+
+51:14.000 --> 51:19.520
+ fundamentally it's an alien, it's so alien from us and our ideas that we cannot communicate with them.
+
+51:19.520 --> 51:25.440
+ Yeah, actually in conversation, cellular automata, Stephen Wolf from
+
+51:25.440 --> 51:31.760
+ brought up is that there could be some ideas that you already have these artificial general
+
+51:31.760 --> 51:36.000
+ intelligence like super smart or maybe conscious beings in the cellular automata, we just don't
+
+51:36.000 --> 51:40.320
+ know how to talk to them. So it's the link with the communication, but you don't know what to do
+
+51:40.320 --> 51:47.920
+ with it. So that's one sort of view is consciousness is only something you can measure. So it's not
+
+51:47.920 --> 51:52.880
+ conscious if you can't measure it. But so you're making an ontological and an epistemic statement.
+
+51:52.880 --> 51:58.640
+ One is there, it's just like seeing their multiverses, that might be true, but I can't
+
+51:58.640 --> 52:03.600
+ communicate with them. I don't have any knowledge of them. That's an epistemic argument. Those are
+
+52:03.600 --> 52:07.520
+ two different things. So it may well be possible. Look, another case that's happening right now,
+
+52:07.520 --> 52:12.240
+ people are building these mini organoids. Do you know about this? So you can take stem cells from
+
+52:12.240 --> 52:15.920
+ under your arm, put it in a dish, add for transcription factors and then you can induce
+
+52:15.920 --> 52:20.800
+ them to grow into large, well large, they're a few millimeter, they're like a half a million
+
+52:20.800 --> 52:25.840
+ neurons that look like nerve cells in a dish called mini organoids. At Harvard, at Stanford,
+
+52:25.840 --> 52:29.760
+ everywhere they're building them. It may be well be possible that they're beginning to feel like
+
+52:29.760 --> 52:34.560
+ something. But we can't really communicate with them right now. So people are beginning to think
+
+52:34.560 --> 52:40.800
+ about the ethics of this. So yes, he may be perfectly right. But it's one question, are
+
+52:40.800 --> 52:44.480
+ they conscious or not? It's totally separate question. How would I know? Those are two different
+
+52:44.480 --> 52:52.480
+ things. Right. If you could give advice to a young researcher sort of dreaming of understanding or
+
+52:52.480 --> 52:58.560
+ creating human level intelligence or consciousness, what would you say?
+
+52:59.840 --> 53:07.680
+ Follow your dreams. Read widely. No, I mean, I suppose with discipline, what is the pursuit
+
+53:07.680 --> 53:11.760
+ that they should take on? Is it neuroscience? Is it competition, cognitive science? Is it
+
+53:11.760 --> 53:19.600
+ philosophy? Is it computer science or robotics? No, in a sense that, okay, so the only known
+
+53:20.960 --> 53:25.520
+ system that have high level of intelligence is homo sapiens. So if you wanted to build it,
+
+53:25.520 --> 53:30.080
+ it's probably good to continue to study closely what humans do. So cognitive neuroscience,
+
+53:30.080 --> 53:34.400
+ you know, somewhere between cognitive neuroscience on the one hand, and some philosophy of mind,
+
+53:34.400 --> 53:40.160
+ and then AI computer science. You can look at all the original ideas, neural networks,
+
+53:40.160 --> 53:45.440
+ they all came from neuroscience, right? And reinforce whether it's snarky, Minsky building
+
+53:45.440 --> 53:49.040
+ is snarky, or whether it's, you know, the early Schubel and Wiesel experiments at Harvard that
+
+53:49.040 --> 53:54.880
+ then gave rise to networks and then multi layer networks. So it may well be possible. In fact,
+
+53:54.880 --> 53:59.760
+ some people argue that to make the next big step in AI once we realize the limits of deep
+
+53:59.760 --> 54:03.680
+ convolutional networks, they can do certain things, but they can't really understand.
+
+54:03.680 --> 54:09.440
+ But they don't, they don't, they can't really, I can't really show them one image. I can show you a
+
+54:09.440 --> 54:15.600
+ single image of somebody a pickpocket who steals a wallet from a purse, you immediately know that's
+
+54:15.600 --> 54:20.800
+ a pickpocket. Now computer system would just say, well, it's a man, it's a woman, it's a purse, right?
+
+54:20.800 --> 54:25.200
+ Unless you train this machine on showing it 100,000 pickpockets, right? So it doesn't,
+
+54:25.200 --> 54:31.040
+ it doesn't have this easy understanding that you have, right? So, so some people make the argument
+
+54:31.040 --> 54:34.640
+ in order to go to the next step where you really want to build machines that understand in a way
+
+54:34.640 --> 54:39.280
+ you and I, we have to go to psychology. We need to understand how we do it and how our brains
+
+54:39.280 --> 54:44.480
+ enable us to do it. And so therefore being on the cusp, it's also so exciting to try to understand
+
+54:44.480 --> 54:49.040
+ better our nature and then to build to take some of those insight and build them. So I think the
+
+54:49.040 --> 54:53.680
+ most exciting thing is somewhere in the interface between cognitive science, neuroscience, AI,
+
+54:53.680 --> 54:55.520
+ computer science and philosophy of mind.
+
+54:55.520 --> 55:00.160
+ Beautiful. Yeah, I'd say if there is, from the machine learning from, from the computer science,
+
+55:00.160 --> 55:05.760
+ computer vision perspective, many of the research is kind of ignore the way the human brain works.
+
+55:05.760 --> 55:12.160
+ Ignore even psychology or literature or studying the brain. I would hope Josh Tannenbaum talks
+
+55:12.160 --> 55:18.640
+ about bringing that in more and more. And that's, yeah. So you worked on some amazing
+
+55:18.640 --> 55:25.520
+ stuff throughout your life. What's the thing that you're really excited about? What's the mystery
+
+55:25.520 --> 55:31.440
+ that you would love to uncover in the near term beyond, beyond all the mysteries that you already
+
+55:31.440 --> 55:37.200
+ surrounded by? Well, so there's this structure called the claustrum. Okay, this is a structure.
+
+55:37.200 --> 55:42.480
+ It's underneath our cortex. It's a big, you have one on the left and a right underneath this
+
+55:42.480 --> 55:46.720
+ underneath the insular. It's very thin. It's like one millimeter. It's embedded in, in wiring,
+
+55:46.720 --> 55:53.760
+ in white matters. It's very difficult to image. And it has, it has connection to every cortical
+
+55:53.760 --> 55:58.640
+ region. And Francis Crick, the last paper he ever wrote, he dictated corrections the day he died
+
+55:58.640 --> 56:05.360
+ in hospital on this paper. He now we hypothesize, well, because it has this unique anatomy, it gets
+
+56:05.360 --> 56:11.280
+ input from every cortical area and projects back to every, every cortical area, that the function
+
+56:11.280 --> 56:18.560
+ of this structure is similar. It's just a metaphor to the role of a conductor in a symphony orchestra.
+
+56:18.560 --> 56:22.720
+ You have all the different cortical players. You have some that do motion, some that do theory of
+
+56:22.720 --> 56:26.880
+ mind, some that infer social interaction and color and hearing and all the different modules and
+
+56:26.880 --> 56:31.280
+ cortex. But of course, what consciousness is, consciousness puts it all together into one
+
+56:31.280 --> 56:35.840
+ package, right? The binding problem, all of that. And this is really the function because it has
+
+56:35.840 --> 56:41.280
+ relatively few neurons compared to cortex, but it, it talks, it sort of receives input from all of
+
+56:41.280 --> 56:45.520
+ them, and it projects back to all of them. And so we're testing that right now. We've got this
+
+56:45.520 --> 56:51.120
+ beautiful neuronal reconstruction in the mouse called crown of thorn, crown of thorn neurons that
+
+56:51.120 --> 56:56.160
+ are in the classroom that have the most widespread connection of any nerve neuron I've ever seen.
+
+56:56.160 --> 57:00.320
+ They're very deep. You have individual neurons that sit in the clouds from tiny, but then they have
+
+57:00.320 --> 57:06.000
+ this single neuron that have this huge axonal tree that cover both ipsy and contralateral cortex
+
+57:06.880 --> 57:11.280
+ and, and trying to turn using, you know, fancy tools like optogenetics, trying to turn those
+
+57:11.280 --> 57:14.720
+ neurons on or off and study it. What happens in the, in the mouse?
+
+57:14.720 --> 57:18.720
+ So this thing is perhaps where the parts become the whole.
+
+57:19.920 --> 57:25.440
+ Perhaps it's one of the structures. It's a very good way of putting it where the, the individual
+
+57:25.440 --> 57:31.760
+ parts turn into the whole of, the whole of the conscious experience. Well, with that,
+
+57:32.640 --> 57:36.000
+ thank you very much for being here today. Thank you very much.
+
+57:36.000 --> 57:56.560
+ I'll be back in a minute. Thanks Jack. Thank you very much.
+

vtt/episode_003_small.vtt

@@ -0,0 +1,1118 @@
+WEBVTT
+
+00:00.000 --> 00:06.400
+ You've studied the human mind, cognition, language, vision, evolution, psychology, from child to adult,
+
+00:07.360 --> 00:11.120
+ from the level of individual to the level of our entire civilization,
+
+00:11.680 --> 00:14.880
+ so I feel like I can start with a simple multiple choice question.
+
+00:16.240 --> 00:21.840
+ What is the meaning of life? Is it A, to attain knowledge, as Plato said,
+
+00:22.400 --> 00:28.000
+ B, to attain power, as Nietzsche said, C, to escape death, as Ernest Becker said,
+
+00:28.000 --> 00:35.040
+ D, to propagate our genes, as Darwin and others have said, E, there is no meaning,
+
+00:35.040 --> 00:41.440
+ as the nihilists have said, F, knowing the meaning of life is beyond our cognitive capabilities,
+
+00:41.440 --> 00:47.360
+ as Stephen Pinker said, based on my interpretation 20 years ago, and G, none of the above.
+
+00:48.160 --> 00:54.720
+ I'd say A comes closest, but I would amend that to attaining not only knowledge, but fulfillment
+
+00:54.720 --> 01:06.000
+ more generally. That is, life, health, stimulation, access to the living cultural and social world.
+
+01:06.000 --> 01:10.720
+ Now, this is our meaning of life. It's not the meaning of life, if you were to ask our genes.
+
+01:12.160 --> 01:17.600
+ Their meaning is to propagate copies of themselves, but that is distinct from the
+
+01:17.600 --> 01:26.640
+ meaning that the brain that they lead to sets for itself. So, to you, knowledge is a small subset
+
+01:26.640 --> 01:33.280
+ or a large subset? It's a large subset, but it's not the entirety of human striving, because we
+
+01:33.280 --> 01:39.120
+ also want to interact with people. We want to experience beauty. We want to experience the
+
+01:39.120 --> 01:47.840
+ richness of the natural world, but understanding what makes the universe tick is way up there.
+
+01:47.840 --> 01:54.000
+ For some of us more than others, certainly for me, that's one of the top five.
+
+01:54.560 --> 02:00.080
+ So, is that a fundamental aspect? Are you just describing your own preference, or is this a
+
+02:00.080 --> 02:05.920
+ fundamental aspect of human nature, is to seek knowledge? In your latest book, you talk about
+
+02:05.920 --> 02:11.760
+ the power, the usefulness of rationality and reason and so on. Is that a fundamental
+
+02:11.760 --> 02:16.160
+ nature of human beings, or is it something we should just strive for?
+
+02:16.960 --> 02:21.840
+ Both. We're capable of striving for it, because it is one of the things that
+
+02:22.640 --> 02:31.360
+ make us what we are, homo sapiens, wise men. We are unusual among our animals in the degree to
+
+02:31.360 --> 02:39.760
+ which we acquire knowledge and use it to survive. We make tools. We strike agreements via language.
+
+02:39.760 --> 02:47.760
+ We extract poisons. We predict the behavior of animals. We try to get at the workings of plants.
+
+02:47.760 --> 02:52.640
+ And when I say we, I don't just mean we in the modern west, but we as a species everywhere,
+
+02:52.640 --> 02:58.160
+ which is how we've managed to occupy every niche on the planet, how we've managed to drive other
+
+02:58.160 --> 03:06.480
+ animals to extinction. And the refinement of reason in pursuit of human well being, of health,
+
+03:06.480 --> 03:13.680
+ happiness, social richness, cultural richness, is our main challenge in the present. That is,
+
+03:14.480 --> 03:19.280
+ using our intellect, using our knowledge to figure out how the world works, how we work,
+
+03:19.280 --> 03:25.200
+ in order to make discoveries and strike agreements that make us all better off in the long run.
+
+03:25.200 --> 03:31.920
+ Right. And you do that almost undeniably in a data driven way in your recent book,
+
+03:31.920 --> 03:36.480
+ but I'd like to focus on the artificial intelligence aspect of things, and not just
+
+03:36.480 --> 03:41.840
+ artificial intelligence, but natural intelligence too. So 20 years ago in the book, you've written
+
+03:41.840 --> 03:49.600
+ on how the mind works, you conjecture, again, am I right to interpret things? You can correct me
+
+03:49.600 --> 03:55.200
+ if I'm wrong, but you conjecture that human thought in the brain may be a result of a network, a massive
+
+03:55.200 --> 04:00.560
+ network of highly interconnected neurons. So from this interconnectivity emerges thought,
+
+04:01.280 --> 04:05.600
+ compared to artificial neural networks, which we use for machine learning today,
+
+04:06.160 --> 04:12.640
+ is there something fundamentally more complex, mysterious, even magical about the biological
+
+04:12.640 --> 04:19.440
+ neural networks versus the ones we've been starting to use over the past 60 years and
+
+04:19.440 --> 04:24.960
+ become to success in the past 10? There is something a little bit mysterious about
+
+04:25.840 --> 04:31.760
+ the human neural networks, which is that each one of us who is a neural network knows that we
+
+04:31.760 --> 04:36.960
+ ourselves are conscious, conscious not in the sense of registering our surroundings or even
+
+04:36.960 --> 04:42.720
+ registering our internal state, but in having subjective first person, present tense experience.
+
+04:42.720 --> 04:49.840
+ That is, when I see red, it's not just different from green, but there's a redness to it that I
+
+04:49.840 --> 04:54.720
+ feel. Whether an artificial system would experience that or not, I don't know and I don't think I
+
+04:54.720 --> 05:00.480
+ can know. That's why it's mysterious. If we had a perfectly lifelike robot that was behaviorally
+
+05:00.480 --> 05:06.800
+ indistinguishable from a human, would we attribute consciousness to it or ought we to attribute
+
+05:06.800 --> 05:12.160
+ consciousness to it? And that's something that it's very hard to know. But putting that aside,
+
+05:12.160 --> 05:19.040
+ putting aside that largely philosophical question, the question is, is there some difference between
+
+05:19.040 --> 05:23.920
+ the human neural network and the ones that we're building in artificial intelligence will mean that
+
+05:23.920 --> 05:30.400
+ we're on the current trajectory not going to reach the point where we've got a lifelike robot
+
+05:30.400 --> 05:35.120
+ indistinguishable from a human because the way their so called neural networks are organized
+
+05:35.120 --> 05:40.560
+ are different from the way ours are organized. I think there's overlap, but I think there are some
+
+05:40.560 --> 05:48.720
+ big differences that their current neural networks, current so called deep learning systems are in
+
+05:48.720 --> 05:53.840
+ reality not all that deep. That is, they are very good at extracting high order statistical
+
+05:53.840 --> 06:00.640
+ regularities. But most of the systems don't have a semantic level, a level of actual understanding
+
+06:00.640 --> 06:06.400
+ of who did what to whom, why, where, how things work, what causes, what else.
+
+06:06.400 --> 06:10.960
+ Do you think that kind of thing can emerge as it does so artificial neural networks are much
+
+06:10.960 --> 06:16.480
+ smaller the number of connections and so on than the current human biological networks? But do you
+
+06:16.480 --> 06:22.640
+ think sort of go to consciousness or to go to this higher level semantic reasoning about things?
+
+06:22.640 --> 06:28.960
+ Do you think that can emerge with just a larger network with a more richly, weirdly interconnected
+
+06:28.960 --> 06:33.280
+ network? Separate it in consciousness because consciousness is even a matter of complexity.
+
+06:33.280 --> 06:37.920
+ A really weird one. Yeah, you could sensibly ask the question of whether shrimp are conscious,
+
+06:37.920 --> 06:43.200
+ for example. They're not terribly complex, but maybe they feel pain. So let's just put that
+
+06:43.200 --> 06:50.000
+ part of it aside. But I think sheer size of a neural network is not enough to give it
+
+06:50.960 --> 06:57.360
+ structure and knowledge. But if it's suitably engineered, then why not? That is, we're neural
+
+06:57.360 --> 07:03.680
+ networks. Natural selection did a kind of equivalent of engineering of our brains. So I don't think
+
+07:03.680 --> 07:10.880
+ there's anything mysterious in the sense that no systemated of silicon could ever do what a human
+
+07:10.880 --> 07:16.080
+ brain can do. I think it's possible in principle. Whether it'll ever happen depends not only on
+
+07:16.080 --> 07:21.040
+ how clever we are in engineering these systems, but whether even we even want to, whether that's
+
+07:21.040 --> 07:27.440
+ even a sensible goal. That is, you can ask the question, is there any locomotion system that is
+
+07:28.320 --> 07:32.960
+ as good as a human? Well, we kind of want to do better than a human ultimately in terms of
+
+07:32.960 --> 07:39.360
+ legged locomotion. There's no reason that humans should be our benchmark. They're tools that might
+
+07:39.360 --> 07:49.280
+ be better in some ways. It may be that we can't duplicate a natural system because at some point,
+
+07:49.280 --> 07:53.840
+ it's so much cheaper to use a natural system that we're not going to invest more brain power
+
+07:53.840 --> 08:00.000
+ and resources. So for example, we don't really have an exact substitute for wood. We still build
+
+08:00.000 --> 08:04.400
+ houses out of wood. We still build furniture out of wood. We like the look. We like the feel. It's
+
+08:04.400 --> 08:09.280
+ wood has certain properties that synthetics don't. There's not that there's anything magical or
+
+08:09.280 --> 08:16.400
+ mysterious about wood. It's just that the extra steps of duplicating everything about wood is
+
+08:16.400 --> 08:20.480
+ something we just haven't bothered because we have wood. Likewise, cotton. I'm wearing cotton
+
+08:20.480 --> 08:26.880
+ clothing now. It feels much better than polyester. It's not that cotton has something magic in it,
+
+08:27.600 --> 08:33.120
+ and it's not that we couldn't ever synthesize something exactly like cotton,
+
+08:33.120 --> 08:37.760
+ but at some point, it's just not worth it. We've got cotton. Likewise, in the case of human
+
+08:37.760 --> 08:43.520
+ intelligence, the goal of making an artificial system that is exactly like the human brain
+
+08:43.520 --> 08:49.440
+ is a goal that we probably know is going to pursue to the bitter end, I suspect, because
+
+08:50.080 --> 08:53.600
+ if you want tools that do things better than humans, you're not going to care whether it
+
+08:53.600 --> 08:58.720
+ does something like humans. So for example, diagnosing cancer or predicting the weather,
+
+08:58.720 --> 09:07.360
+ why set humans as your benchmark? But in general, I suspect you also believe that even if the human
+
+09:07.360 --> 09:11.440
+ should not be a benchmark and we don't want to imitate humans in their system, there's a lot
+
+09:11.440 --> 09:16.800
+ to be learned about how to create an artificial intelligence system by studying the humans.
+
+09:16.800 --> 09:23.440
+ Yeah, I think that's right. In the same way that to build flying machines, we want to understand
+
+09:23.440 --> 09:28.880
+ the laws of aerodynamics, including birds, but not mimic the birds, but they're the same laws.
+
+09:30.480 --> 09:38.400
+ You have a view on AI, artificial intelligence and safety, that from my perspective,
+
+09:38.400 --> 09:49.360
+ is refreshingly rational, or perhaps more importantly, has elements of positivity to it,
+
+09:49.360 --> 09:55.440
+ which I think can be inspiring and empowering as opposed to paralyzing. For many people,
+
+09:55.440 --> 10:02.320
+ including AI researchers, the eventual existential threat of AI is obvious, not only possible but
+
+10:02.320 --> 10:08.640
+ obvious. And for many others, including AI researchers, the threat is not obvious. So
+
+10:09.520 --> 10:16.480
+ Elon Musk is famously in the highly concerned about AI camp, saying things like AI is far
+
+10:16.480 --> 10:22.240
+ more dangerous than nuclear weapons, and that AI will likely destroy human civilization.
+
+10:22.960 --> 10:30.400
+ So in February, you said that if Elon was really serious about AI, the threat of AI,
+
+10:30.400 --> 10:34.960
+ he would stop building self driving cars that he's doing very successfully as part of Tesla.
+
+10:35.840 --> 10:40.880
+ Then he said, wow, if even Pinker doesn't understand the difference between narrow AI
+
+10:40.880 --> 10:47.280
+ like a car and general AI, when the latter literally has a million times more compute power
+
+10:47.280 --> 10:54.240
+ and an open ended utility function, humanity is in deep trouble. So first, what did you mean by
+
+10:54.240 --> 10:59.200
+ the statement about Elon Musk should stop building self driving cars if he's deeply concerned?
+
+10:59.200 --> 11:03.520
+ Well, not the last time that Elon Musk has fired off an intemperate tweet.
+
+11:04.320 --> 11:07.600
+ Well, we live in a world where Twitter has power.
+
+11:07.600 --> 11:16.640
+ Yes. Yeah, I think there are two kinds of existential threat that have been discussed
+
+11:16.640 --> 11:19.760
+ in connection with artificial intelligence, and I think that they're both incoherent.
+
+11:20.480 --> 11:28.800
+ One of them is a vague fear of AI takeover, that just as we subjugated animals and less
+
+11:28.800 --> 11:33.360
+ technologically advanced peoples, so if we build something that's more advanced than us,
+
+11:33.360 --> 11:39.200
+ it will inevitably turn us into pets or slaves or domesticated animal equivalents.
+
+11:40.240 --> 11:46.720
+ I think this confuses intelligence with a will to power that it so happens that in the
+
+11:46.720 --> 11:52.240
+ intelligence system we are most familiar with, namely Homo sapiens, we are products of natural
+
+11:52.240 --> 11:56.800
+ selection, which is a competitive process. And so bundled together with our problem solving
+
+11:56.800 --> 12:05.200
+ capacity are a number of nasty traits like dominance and exploitation and maximization of
+
+12:05.200 --> 12:11.040
+ power and glory and resources and influence. There's no reason to think that sheer problem
+
+12:11.040 --> 12:16.640
+ solving capability will set that as one of its goals. Its goals will be whatever we set its goals
+
+12:16.640 --> 12:21.760
+ as, and as long as someone isn't building a megalomaniacal artificial intelligence,
+
+12:22.560 --> 12:25.360
+ then there's no reason to think that it would naturally evolve in that direction.
+
+12:25.360 --> 12:31.600
+ Now you might say, well, what if we gave it the goal of maximizing its own power source?
+
+12:31.600 --> 12:35.280
+ That's a pretty stupid goal to give an autonomous system. You don't give it that goal.
+
+12:36.000 --> 12:41.360
+ I mean, that's just self evidently idiotic. So if you look at the history of the world,
+
+12:41.360 --> 12:45.680
+ there's been a lot of opportunities where engineers could instill in a system destructive
+
+12:45.680 --> 12:49.520
+ power and they choose not to because that's the natural process of engineering.
+
+12:49.520 --> 12:52.880
+ Well, except for weapons. I mean, if you're building a weapon, its goal is to destroy
+
+12:52.880 --> 12:58.400
+ people. And so I think there are good reasons to not build certain kinds of weapons. I think
+
+12:58.400 --> 13:06.240
+ building nuclear weapons was a massive mistake. You do. So maybe pause on that because that is
+
+13:06.240 --> 13:12.880
+ one of the serious threats. Do you think that it was a mistake in a sense that it should have been
+
+13:12.880 --> 13:19.200
+ stopped early on? Or do you think it's just an unfortunate event of invention that this was
+
+13:19.200 --> 13:22.800
+ invented? Do you think it's possible to stop, I guess, is the question on that? Yeah, it's hard to
+
+13:22.800 --> 13:27.440
+ rewind the clock because, of course, it was invented in the context of World War II and the
+
+13:27.440 --> 13:33.120
+ fear that the Nazis might develop one first. Then once it was initiated for that reason,
+
+13:33.120 --> 13:40.800
+ it was hard to turn off, especially since winning the war against the Japanese and the Nazis was
+
+13:40.800 --> 13:46.160
+ such an overwhelming goal of every responsible person that they were just nothing that people
+
+13:46.160 --> 13:51.440
+ wouldn't have done then to ensure victory. It's quite possible if World War II hadn't happened
+
+13:51.440 --> 13:56.560
+ that nuclear weapons wouldn't have been invented. We can't know. But I don't think it was, by any
+
+13:56.560 --> 14:01.760
+ means, a necessity any more than some of the other weapon systems that were envisioned but never
+
+14:01.760 --> 14:09.040
+ implemented, like planes that would disperse poison gas over cities like crop dusters or systems to
+
+14:09.040 --> 14:16.080
+ try to create earthquakes and tsunamis in enemy countries, to weaponize the weather,
+
+14:16.080 --> 14:21.520
+ weaponize solar flares, all kinds of crazy schemes that we thought the better of. I think
+
+14:21.520 --> 14:26.800
+ analogies between nuclear weapons and artificial intelligence are fundamentally misguided because
+
+14:26.800 --> 14:31.520
+ the whole point of nuclear weapons is to destroy things. The point of artificial intelligence
+
+14:31.520 --> 14:37.360
+ is not to destroy things. So the analogy is misleading. So there's two artificial
+
+14:37.360 --> 14:42.320
+ intelligence you mentioned. The first one was the highly intelligent or power hungry. Yeah,
+
+14:42.320 --> 14:47.040
+ an assistant that we design ourselves where we give it the goals. Goals are external to the
+
+14:48.320 --> 14:55.840
+ means to attain the goals. If we don't design an artificially intelligent system to maximize
+
+14:56.560 --> 15:02.400
+ dominance, then it won't maximize dominance. It's just that we're so familiar with homo sapiens
+
+15:02.400 --> 15:08.800
+ where these two traits come bundled together, particularly in men, that we are apt to confuse
+
+15:08.800 --> 15:16.720
+ high intelligence with a will to power. But that's just an error. The other fear is that
+
+15:16.720 --> 15:23.040
+ we'll be collateral damage that will give artificial intelligence a goal like make paper clips
+
+15:23.040 --> 15:28.320
+ and it will pursue that goal so brilliantly that before we can stop it, it turns us into paper
+
+15:28.320 --> 15:34.400
+ clips. We'll give it the goal of curing cancer and it will turn us into guinea pigs for lethal
+
+15:34.400 --> 15:40.000
+ experiments or give it the goal of world peace and its conception of world peace is no people,
+
+15:40.000 --> 15:44.480
+ therefore no fighting and so it will kill us all. Now, I think these are utterly fanciful. In fact,
+
+15:44.480 --> 15:49.600
+ I think they're actually self defeating. They first of all assume that we're going to be so
+
+15:49.600 --> 15:54.880
+ brilliant that we can design an artificial intelligence that can cure cancer. But so stupid
+
+15:54.880 --> 16:00.160
+ that we don't specify what we mean by curing cancer in enough detail that it won't kill us in the
+
+16:00.160 --> 16:06.720
+ process. And it assumes that the system will be so smart that it can cure cancer. But so
+
+16:06.720 --> 16:11.520
+ idiotic that it doesn't can't figure out that what we mean by curing cancer is not killing
+
+16:11.520 --> 16:17.920
+ everyone. So I think that the collateral damage scenario, the value alignment problem is also
+
+16:17.920 --> 16:23.200
+ based on a misconception. So one of the challenges, of course, we don't know how to build either system
+
+16:23.200 --> 16:27.440
+ currently, or are we even close to knowing? Of course, those things can change overnight,
+
+16:27.440 --> 16:33.840
+ but at this time, theorizing about is very challenging in either direction. So that's
+
+16:33.840 --> 16:39.600
+ probably at the core of the problem is without that ability to reason about the real engineering
+
+16:39.600 --> 16:45.200
+ things here at hand is your imagination runs away with things. Exactly. But let me sort of ask,
+
+16:45.920 --> 16:52.320
+ what do you think was the motivation, the thought process of Elon Musk? I build autonomous vehicles,
+
+16:52.320 --> 16:58.000
+ I study autonomous vehicles, I study Tesla autopilot. I think it is one of the greatest
+
+16:58.000 --> 17:02.880
+ currently application, large scale application of artificial intelligence in the world.
+
+17:02.880 --> 17:09.120
+ It has a potentially very positive impact on society. So how does a person who's creating this
+
+17:09.120 --> 17:17.680
+ very good, quote unquote, narrow AI system also seem to be so concerned about this other
+
+17:17.680 --> 17:21.120
+ general AI? What do you think is the motivation there? What do you think is the thing?
+
+17:21.120 --> 17:30.640
+ Well, you probably have to ask him, but there and he is notoriously flamboyant, impulsive to the,
+
+17:30.640 --> 17:35.120
+ as we have just seen, to the detriment of his own goals of the health of a company.
+
+17:36.000 --> 17:41.600
+ So I don't know what's going on in his mind. You probably have to ask him. But I don't think the,
+
+17:41.600 --> 17:48.160
+ and I don't think the distinction between special purpose AI and so called general AI is relevant
+
+17:48.160 --> 17:54.400
+ that in the same way that special purpose AI is not going to do anything conceivable in order to
+
+17:54.400 --> 18:00.560
+ attain a goal, all engineering systems have to are designed to trade off across multiple goals.
+
+18:00.560 --> 18:05.920
+ When we build cars in the first place, we didn't forget to install brakes because the goal of a
+
+18:05.920 --> 18:12.320
+ car is to go fast. It occurred to people, yes, you want to go fast, but not always. So you build
+
+18:12.320 --> 18:18.960
+ and brakes too. Likewise, if a car is going to be autonomous, that doesn't program it to take the
+
+18:18.960 --> 18:23.440
+ shortest route to the airport. It's not going to take the diagonal and mow down people and trees
+
+18:23.440 --> 18:28.000
+ and fences because that's the shortest route. That's not what we mean by the shortest route when we
+
+18:28.000 --> 18:34.720
+ program it. And that's just what an intelligence system is by definition. It takes into account
+
+18:34.720 --> 18:40.640
+ multiple constraints. The same is true, in fact, even more true of so called general intelligence.
+
+18:40.640 --> 18:47.040
+ That is, if it's genuinely intelligent, it's not going to pursue some goal single mindedly,
+
+18:47.040 --> 18:53.280
+ omitting every other consideration and collateral effect. That's not artificial and
+
+18:53.280 --> 18:58.560
+ general intelligence. That's artificial stupidity. I agree with you, by the way,
+
+18:58.560 --> 19:03.280
+ on the promise of autonomous vehicles for improving human welfare. I think it's spectacular.
+
+19:03.280 --> 19:08.080
+ And I'm surprised at how little press coverage notes that in the United States alone,
+
+19:08.080 --> 19:13.200
+ something like 40,000 people die every year on the highways, vastly more than are killed by
+
+19:13.200 --> 19:19.440
+ terrorists. And we spend a trillion dollars on a war to combat deaths by terrorism,
+
+19:19.440 --> 19:24.080
+ about half a dozen a year, whereas every year and year out, 40,000 people are
+
+19:24.080 --> 19:27.600
+ massacred on the highways, which could be brought down to very close to zero.
+
+19:28.560 --> 19:31.840
+ So I'm with you on the humanitarian benefit.
+
+19:31.840 --> 19:36.240
+ Let me just mention that as a person who's building these cars, it is a little bit offensive to me
+
+19:36.240 --> 19:41.680
+ to say that engineers would be clueless enough not to engineer safety into systems. I often
+
+19:41.680 --> 19:46.400
+ stay up at night thinking about those 40,000 people that are dying. And everything I try to
+
+19:46.400 --> 19:52.000
+ engineer is to save those people's lives. So every new invention that I'm super excited about,
+
+19:52.000 --> 19:59.680
+ every new, all the deep learning literature and CVPR conferences and NIPS, everything I'm super
+
+19:59.680 --> 20:08.320
+ excited about is all grounded in making it safe and help people. So I just don't see how that
+
+20:08.320 --> 20:13.200
+ trajectory can all of a sudden slip into a situation where intelligence will be highly
+
+20:13.200 --> 20:17.840
+ negative. You and I certainly agree on that. And I think that's only the beginning of the
+
+20:17.840 --> 20:23.760
+ potential humanitarian benefits of artificial intelligence. There's been enormous attention
+
+20:23.760 --> 20:27.680
+ to what are we going to do with the people whose jobs are made obsolete by artificial
+
+20:27.680 --> 20:31.600
+ intelligence. But very little attention given to the fact that the jobs that are going to be
+
+20:31.600 --> 20:37.600
+ made obsolete are horrible jobs. The fact that people aren't going to be picking crops and making
+
+20:37.600 --> 20:43.760
+ beds and driving trucks and mining coal, these are soul deadening jobs. And we have a whole
+
+20:43.760 --> 20:51.280
+ literature sympathizing with the people stuck in these menial, mind deadening, dangerous jobs.
+
+20:52.080 --> 20:56.160
+ If we can eliminate them, this is a fantastic boon to humanity. Now, granted,
+
+20:56.160 --> 21:02.160
+ we, you solve one problem and there's another one, namely, how do we get these people a decent
+
+21:02.160 --> 21:08.320
+ income? But if we're smart enough to invent machines that can make beds and put away dishes and
+
+21:09.520 --> 21:14.080
+ handle hospital patients, I think we're smart enough to figure out how to redistribute income
+
+21:14.080 --> 21:20.960
+ to a portion, some of the vast economic savings to the human beings who will no longer be needed to
+
+21:20.960 --> 21:28.400
+ make beds. Okay. Sam Harris says that it's obvious that eventually AI will be an existential risk.
+
+21:29.280 --> 21:36.240
+ He's one of the people who says it's obvious. We don't know when the claim goes, but eventually
+
+21:36.240 --> 21:41.760
+ it's obvious. And because we don't know when we should worry about it now. It's a very interesting
+
+21:41.760 --> 21:49.120
+ argument in my eyes. So how do we think about timescale? How do we think about existential
+
+21:49.120 --> 21:55.040
+ threats when we don't really, we know so little about the threat, unlike nuclear weapons, perhaps,
+
+21:55.040 --> 22:02.400
+ about this particular threat, that it could happen tomorrow, right? So, but very likely it won't.
+
+22:03.120 --> 22:08.320
+ Very likely it'd be 100 years away. So how do, do we ignore it? Do, how do we talk about it?
+
+22:08.880 --> 22:13.040
+ Do we worry about it? What, how do we think about those? What is it?
+
+22:13.040 --> 22:19.600
+ A threat that we can imagine, it's within the limits of our imagination, but not within our
+
+22:19.600 --> 22:25.760
+ limits of understanding to sufficient, to accurately predict it. But what, what is, what is the it
+
+22:25.760 --> 22:31.280
+ that we're referring to? Oh, AI, sorry, AI, AI being the existential threat. AI can always...
+
+22:31.280 --> 22:34.400
+ How? But like enslaving us or turning us into paperclips?
+
+22:35.120 --> 22:38.800
+ I think the most compelling from the Sam Harris perspective would be the paperclip situation.
+
+22:38.800 --> 22:44.000
+ Yeah. I mean, I just think it's totally fanciful. I mean, that is, don't build a system. Don't give a,
+
+22:44.000 --> 22:50.400
+ don't... First of all, the code of engineering is you don't implement a system with massive
+
+22:50.400 --> 22:55.040
+ control before testing it. Now, perhaps the culture of engineering will radically change,
+
+22:55.040 --> 23:00.320
+ then I would worry, but I don't see any signs that engineers will suddenly do idiotic things,
+
+23:00.320 --> 23:05.440
+ like put a, an electrical power plant in control of a system that they haven't tested
+
+23:05.440 --> 23:14.720
+ first. Or all of these scenarios not only imagine a almost a magically powered intelligence,
+
+23:15.360 --> 23:20.000
+ you know, including things like cure cancer, which is probably an incoherent goal because
+
+23:20.000 --> 23:25.440
+ there's so many different kinds of cancer or bring about world peace. I mean, how do you even specify
+
+23:25.440 --> 23:31.360
+ that as a goal? But the scenarios also imagine some degree of control of every molecule in the
+
+23:31.360 --> 23:38.480
+ universe, which not only is itself unlikely, but we would not start to connect these systems to
+
+23:39.200 --> 23:45.840
+ infrastructure without, without testing as we would any kind of engineering system. Now,
+
+23:45.840 --> 23:53.920
+ maybe some engineers will be irresponsible and we need legal and regulatory and legal
+
+23:53.920 --> 23:59.440
+ responsibility implemented so that engineers don't do things that are stupid by their own standards.
+
+23:59.440 --> 24:08.560
+ But the, I've never seen enough of a plausible scenario of existential threat to devote large
+
+24:08.560 --> 24:14.720
+ amounts of brain power to, to forestall it. So you believe in the sort of the power en masse of
+
+24:14.720 --> 24:19.520
+ the engineering of reason as you argue in your latest book of reason and science to sort of
+
+24:20.400 --> 24:26.160
+ be the very thing that guides the development of new technology so it's safe and also keeps us
+
+24:26.160 --> 24:32.480
+ safe. Yeah, the same, you know, granted the same culture of safety that currently is part of the
+
+24:32.480 --> 24:38.960
+ engineering mindset for airplanes, for example. So yeah, I don't think that, that that should
+
+24:38.960 --> 24:44.800
+ be thrown out the window and that untested, all powerful systems should be suddenly implemented.
+
+24:44.800 --> 24:47.360
+ But there's no reason to think they are. And in fact, if you look at the
+
+24:48.160 --> 24:51.760
+ progress of artificial intelligence, it's been, you know, it's been impressive, especially in
+
+24:51.760 --> 24:56.960
+ the last 10 years or so. But the idea that suddenly there'll be a step function that all of a sudden
+
+24:56.960 --> 25:02.160
+ before we know it, it will be all powerful, that there'll be some kind of recursive self
+
+25:02.160 --> 25:11.200
+ improvement, some kind of fume is also fanciful. Certainly by the technology that we that we're
+
+25:11.200 --> 25:16.720
+ now impresses us, such as deep learning, where you train something on hundreds of thousands or
+
+25:16.720 --> 25:22.720
+ millions of examples, they're not hundreds of thousands of problems of which curing cancer is
+
+25:24.320 --> 25:30.560
+ typical example. And so the kind of techniques that have allowed AI to increase in the last
+
+25:30.560 --> 25:37.600
+ five years are not the kind that are going to lead to this fantasy of exponential sudden
+
+25:37.600 --> 25:43.680
+ self improvement. So I think it's kind of a magical thinking. It's not based on our understanding
+
+25:43.680 --> 25:49.200
+ of how AI actually works. Now, give me a chance here. So you said fanciful, magical thinking.
+
+25:50.240 --> 25:55.280
+ In his TED Talk, Sam Harris says that thinking about AI killing all human civilization is somehow
+
+25:55.280 --> 26:00.400
+ fun intellectually. Now, I have to say as a scientist engineer, I don't find it fun.
+
+26:01.200 --> 26:08.560
+ But when I'm having beer with my non AI friends, there is indeed something fun and appealing about
+
+26:08.560 --> 26:14.720
+ it. Like talking about an episode of Black Mirror, considering if a large meteor is headed towards
+
+26:14.720 --> 26:20.640
+ Earth, we were just told a large meteor is headed towards Earth, something like this. And can you
+
+26:20.640 --> 26:25.840
+ relate to this sense of fun? And do you understand the psychology of it? Yes, great. Good question.
+
+26:26.880 --> 26:33.440
+ I personally don't find it fun. I find it kind of actually a waste of time, because there are
+
+26:33.440 --> 26:39.840
+ genuine threats that we ought to be thinking about, like pandemics, like cybersecurity
+
+26:39.840 --> 26:47.040
+ vulnerabilities, like the possibility of nuclear war and certainly climate change. This is enough
+
+26:47.040 --> 26:55.280
+ to fill many conversations without. And I think Sam did put his finger on something, namely that
+
+26:55.280 --> 27:03.120
+ there is a community, sometimes called the rationality community, that delights in using its
+
+27:03.120 --> 27:10.160
+ brain power to come up with scenarios that would not occur to mere mortals, to less cerebral people.
+
+27:10.160 --> 27:15.360
+ So there is a kind of intellectual thrill in finding new things to worry about that no one
+
+27:15.360 --> 27:21.200
+ has worried about yet. I actually think, though, that it's not only is it a kind of fun that doesn't
+
+27:21.200 --> 27:27.280
+ give me particular pleasure. But I think there can be a pernicious side to it, namely that you
+
+27:27.280 --> 27:35.280
+ overcome people with such dread, such fatalism, that there's so many ways to die to annihilate
+
+27:35.280 --> 27:40.160
+ our civilization that we may as well enjoy life while we can. There's nothing we can do about it.
+
+27:40.160 --> 27:46.560
+ If climate change doesn't do us in, then runaway robots will. So let's enjoy ourselves now. We
+
+27:46.560 --> 27:55.200
+ got to prioritize. We have to look at threats that are close to certainty, such as climate change,
+
+27:55.200 --> 28:00.320
+ and distinguish those from ones that are merely imaginable, but with infinitesimal probabilities.
+
+28:01.360 --> 28:07.120
+ And we have to take into account people's worry budget. You can't worry about everything. And
+
+28:07.120 --> 28:13.920
+ if you sow dread and fear and terror and and fatalism, it can lead to a kind of numbness. Well,
+
+28:13.920 --> 28:18.240
+ they're just these problems are overwhelming and the engineers are just going to kill us all.
+
+28:19.040 --> 28:25.760
+ So let's either destroy the entire infrastructure of science, technology,
+
+28:26.640 --> 28:32.080
+ or let's just enjoy life while we can. So there's a certain line of worry, which I'm
+
+28:32.080 --> 28:36.160
+ worried about a lot of things engineering. There's a certain line of worry when you cross,
+
+28:36.160 --> 28:42.800
+ you allow it to cross, that it becomes paralyzing fear as opposed to productive fear. And that's
+
+28:42.800 --> 28:49.760
+ kind of what you're highlighting. Exactly right. And we've seen some, we know that human effort is
+
+28:49.760 --> 28:58.080
+ not well calibrated against risk in that because a basic tenet of cognitive psychology is that
+
+28:59.440 --> 29:05.120
+ perception of risk and hence perception of fear is driven by imaginability, not by data.
+
+29:05.920 --> 29:11.200
+ And so we misallocate vast amounts of resources to avoiding terrorism,
+
+29:11.200 --> 29:16.240
+ which kills on average about six Americans a year with a one exception of 9 11. We invade
+
+29:16.240 --> 29:23.920
+ countries, we invent an entire new departments of government with massive, massive expenditure
+
+29:23.920 --> 29:30.800
+ of resources and lives to defend ourselves against a trivial risk. Whereas guaranteed risks,
+
+29:30.800 --> 29:36.720
+ you mentioned as one of them, you mentioned traffic fatalities and even risks that are
+
+29:36.720 --> 29:46.240
+ not here, but are plausible enough to worry about like pandemics, like nuclear war,
+
+29:47.120 --> 29:51.760
+ receive far too little attention. In presidential debates, there's no discussion of
+
+29:51.760 --> 29:56.720
+ how to minimize the risk of nuclear war, lots of discussion of terrorism, for example.
+
+29:57.840 --> 30:05.520
+ And so we, I think it's essential to calibrate our budget of fear, worry, concerned planning
+
+30:05.520 --> 30:12.640
+ to the actual probability of harm. Yep. So let me ask this in this question.
+
+30:13.520 --> 30:18.960
+ So speaking of imaginability, you said it's important to think about reason. And one of my
+
+30:18.960 --> 30:26.560
+ favorite people who likes to dip into the outskirts of reason through fascinating exploration of his
+
+30:26.560 --> 30:34.880
+ imagination is Joe Rogan. Oh, yes. So who has, through reason, used to believe a lot of conspiracies
+
+30:34.880 --> 30:40.000
+ and through reason has stripped away a lot of his beliefs in that way. So it's fascinating actually
+
+30:40.000 --> 30:47.920
+ to watch him through rationality, kind of throw away the ideas of Bigfoot and 911. I'm not sure
+
+30:47.920 --> 30:52.320
+ exactly. Kim Trails. I don't know what he believes in. Yes, okay. But he no longer believed in,
+
+30:52.320 --> 30:57.920
+ that's right. No, he's become a real force for good. So you were on the Joe Rogan podcast in
+
+30:57.920 --> 31:02.880
+ February and had a fascinating conversation, but as far as I remember, didn't talk much about
+
+31:02.880 --> 31:09.280
+ artificial intelligence. I will be on his podcast in a couple of weeks. Joe is very much concerned
+
+31:09.280 --> 31:14.640
+ about existential threat of AI. I'm not sure if you're, this is why I was hoping that you'll get
+
+31:14.640 --> 31:20.480
+ into that topic. And in this way, he represents quite a lot of people who look at the topic of AI
+
+31:20.480 --> 31:27.840
+ from 10,000 foot level. So as an exercise of communication, you said it's important to be
+
+31:27.840 --> 31:33.280
+ rational and reason about these things. Let me ask, if you were to coach me as an AI researcher
+
+31:33.280 --> 31:38.320
+ about how to speak to Joe and the general public about AI, what would you advise?
+
+31:38.320 --> 31:42.400
+ Well, the short answer would be to read the sections that I wrote in Enlightenment.
+
+31:44.080 --> 31:48.880
+ But longer reason would be, I think to emphasize, and I think you're very well positioned as an
+
+31:48.880 --> 31:54.800
+ engineer to remind people about the culture of engineering, that it really is safety oriented,
+
+31:54.800 --> 32:02.160
+ that another discussion in Enlightenment now, I plot rates of accidental death from various
+
+32:02.160 --> 32:09.280
+ causes, plane crashes, car crashes, occupational accidents, even death by lightning strikes,
+
+32:09.280 --> 32:16.560
+ and they all plummet. Because the culture of engineering is how do you squeeze out the lethal
+
+32:16.560 --> 32:23.360
+ risks, death by fire, death by drowning, death by asphyxiation, all of them drastically declined
+
+32:23.360 --> 32:28.160
+ because of advances in engineering, that I got to say, I did not appreciate until I saw those
+
+32:28.160 --> 32:34.000
+ graphs. And it is because exactly people like you who stay up at night thinking, oh my God,
+
+32:36.000 --> 32:42.560
+ what I'm inventing likely to hurt people and to deploy ingenuity to prevent that from happening.
+
+32:42.560 --> 32:47.360
+ Now, I'm not an engineer, although I spent 22 years at MIT, so I know something about the culture
+
+32:47.360 --> 32:51.360
+ of engineering. My understanding is that this is the way you think if you're an engineer.
+
+32:51.360 --> 32:58.160
+ And it's essential that that culture not be suddenly switched off when it comes to artificial
+
+32:58.160 --> 33:02.080
+ intelligence. So I mean, that could be a problem, but is there any reason to think it would be
+
+33:02.080 --> 33:07.360
+ switched off? I don't think so. And one, there's not enough engineers speaking up for this way,
+
+33:07.360 --> 33:13.680
+ for the excitement, for the positive view of human nature, what you're trying to create is
+
+33:13.680 --> 33:18.240
+ the positivity, like everything we try to invent is trying to do good for the world.
+
+33:18.240 --> 33:23.600
+ But let me ask you about the psychology of negativity. It seems just objectively,
+
+33:23.600 --> 33:27.680
+ not considering the topic, it seems that being negative about the future, it makes you sound
+
+33:27.680 --> 33:32.720
+ smarter than being positive about the future, in regard to this topic. Am I correct in this
+
+33:32.720 --> 33:39.120
+ observation? And if so, why do you think that is? Yeah, I think there is that phenomenon,
+
+33:39.120 --> 33:43.920
+ that as Tom Lehrer, the satirist said, always predict the worst and you'll be hailed as a
+
+33:43.920 --> 33:51.840
+ prophet. It may be part of our overall negativity bias. We are as a species more attuned to the
+
+33:51.840 --> 33:59.200
+ negative than the positive. We dread losses more than we enjoy gains. And that might open up a
+
+33:59.200 --> 34:06.560
+ space for prophets to remind us of harms and risks and losses that we may have overlooked.
+
+34:06.560 --> 34:15.040
+ So I think there is that asymmetry. So you've written some of my favorite books
+
+34:16.080 --> 34:21.680
+ all over the place. So starting from Enlightenment now, to the better ranges of our nature,
+
+34:21.680 --> 34:28.560
+ blank slate, how the mind works, the one about language, language instinct. Bill Gates,
+
+34:28.560 --> 34:37.840
+ big fan too, said of your most recent book that it's my new favorite book of all time. So for
+
+34:37.840 --> 34:44.000
+ you as an author, what was the book early on in your life that had a profound impact on the way
+
+34:44.000 --> 34:50.560
+ you saw the world? Certainly this book Enlightenment now is influenced by David Deutch's The Beginning
+
+34:50.560 --> 34:57.520
+ of Infinity. We have a rather deep reflection on knowledge and the power of knowledge to improve
+
+34:57.520 --> 35:02.960
+ the human condition. They end with bits of wisdom such as that problems are inevitable,
+
+35:02.960 --> 35:07.760
+ but problems are solvable given the right knowledge and that solutions create new problems
+
+35:07.760 --> 35:12.480
+ that have to be solved in their turn. That's I think a kind of wisdom about the human condition
+
+35:12.480 --> 35:16.960
+ that influenced the writing of this book. There's some books that are excellent but obscure,
+
+35:16.960 --> 35:22.080
+ some of which I have on my page on my website. I read a book called The History of Force,
+
+35:22.080 --> 35:27.920
+ self published by a political scientist named James Payne on the historical decline of violence and
+
+35:27.920 --> 35:35.120
+ that was one of the inspirations for the better angels of our nature. What about early on if
+
+35:35.120 --> 35:40.640
+ you look back when you were maybe a teenager? I loved a book called One, Two, Three, Infinity.
+
+35:40.640 --> 35:45.920
+ When I was a young adult, I read that book by George Gamov, the physicist, which had very
+
+35:45.920 --> 35:55.120
+ accessible and humorous explanations of relativity, of number theory, of dimensionality, high
+
+35:56.080 --> 36:02.240
+ multiple dimensional spaces in a way that I think is still delightful 70 years after it was published.
+
+36:03.120 --> 36:09.280
+ I like the Time Life Science series. These are books that arrive every month that my mother
+
+36:09.280 --> 36:15.600
+ subscribed to. Each one on a different topic. One would be on electricity, one would be on
+
+36:15.600 --> 36:21.440
+ forests, one would be on evolution, and then one was on the mind. I was just intrigued that there
+
+36:21.440 --> 36:27.040
+ could be a science of mind. That book, I would cite as an influence as well. Then later on.
+
+36:27.040 --> 36:30.960
+ That's when you fell in love with the idea of studying the mind. Was that the thing that grabbed
+
+36:30.960 --> 36:38.560
+ you? It was one of the things, I would say. I read as a college student the book Reflections on
+
+36:38.560 --> 36:44.800
+ Language by Noam Chomsky. He spent most of his career here at MIT. Richard Dawkins,
+
+36:44.800 --> 36:48.800
+ two books, The Blind Watchmaker and the Selfish Gene were enormously influential,
+
+36:49.520 --> 36:56.640
+ partly mainly for the content, but also for the writing style, the ability to explain
+
+36:56.640 --> 37:03.760
+ abstract concepts in lively prose. Stephen Jay Gould's first collection ever since Darwin, also
+
+37:05.040 --> 37:11.120
+ excellent example of lively writing. George Miller, the psychologist that most psychologists
+
+37:11.120 --> 37:17.440
+ are familiar with, came up with the idea that human memory has a capacity of seven plus or minus
+
+37:17.440 --> 37:21.920
+ two chunks. That's probably his biggest claim to fame. He wrote a couple of books on language
+
+37:21.920 --> 37:27.520
+ and communication that I'd read as an undergraduate. Again, beautifully written and intellectually deep.
+
+37:28.400 --> 37:31.840
+ Wonderful. Stephen, thank you so much for taking the time today.
+
+37:31.840 --> 37:42.960
+ My pleasure. Thanks a lot, Lex.
+

vtt/episode_004_small.vtt

@@ -0,0 +1,1109 @@
+WEBVTT
+
+00:00.000 --> 00:04.320
+ What difference between biological neural networks and artificial neural networks
+
+00:04.320 --> 00:07.680
+ is most mysterious, captivating and profound for you?
+
+00:11.120 --> 00:15.280
+ First of all, there's so much we don't know about biological neural networks,
+
+00:15.280 --> 00:21.840
+ and that's very mysterious and captivating because maybe it holds the key to improving
+
+00:21.840 --> 00:29.840
+ artificial neural networks. One of the things I studied recently is something that
+
+00:29.840 --> 00:36.160
+ we don't know how biological neural networks do, but would be really useful for artificial ones,
+
+00:37.120 --> 00:43.440
+ is the ability to do credit assignment through very long time spans.
+
+00:44.080 --> 00:49.680
+ There are things that we can in principle do with artificial neural nets, but it's not very
+
+00:49.680 --> 00:55.920
+ convenient and it's not biologically plausible. And this mismatch, I think this kind of mismatch,
+
+00:55.920 --> 01:03.600
+ maybe an interesting thing to study, to A, understand better how brains might do these
+
+01:03.600 --> 01:08.720
+ things because we don't have good corresponding theories with artificial neural nets, and B,
+
+01:10.240 --> 01:19.040
+ maybe provide new ideas that we could explore about things that brain do differently and
+
+01:19.040 --> 01:22.160
+ that we could incorporate in artificial neural nets.
+
+01:22.160 --> 01:27.680
+ So let's break credit assignment up a little bit. So what? It's a beautifully technical term,
+
+01:27.680 --> 01:34.560
+ but it could incorporate so many things. So is it more on the RNN memory side,
+
+01:35.840 --> 01:39.760
+ thinking like that, or is it something about knowledge, building up common sense knowledge
+
+01:39.760 --> 01:46.560
+ over time, or is it more in the reinforcement learning sense that you're picking up rewards
+
+01:46.560 --> 01:50.080
+ over time for a particular to achieve a certain kind of goal?
+
+01:50.080 --> 01:58.080
+ So I was thinking more about the first two meanings whereby we store all kinds of memories,
+
+01:59.120 --> 02:09.680
+ episodic memories in our brain, which we can access later in order to help us both infer
+
+02:10.560 --> 02:19.520
+ causes of things that we are observing now and assign credit to decisions or interpretations
+
+02:19.520 --> 02:26.960
+ we came up with a while ago when those memories were stored. And then we can change the way we
+
+02:26.960 --> 02:34.800
+ would have reacted or interpreted things in the past, and now that's credit assignment used for learning.
+
+02:36.320 --> 02:43.760
+ So in which way do you think artificial neural networks, the current LSTM,
+
+02:43.760 --> 02:52.240
+ the current architectures are not able to capture the presumably you're thinking of very long term?
+
+02:52.240 --> 03:00.720
+ Yes. So current, the current nets are doing a fairly good jobs for sequences with dozens or say
+
+03:00.720 --> 03:06.560
+ hundreds of time steps. And then it gets sort of harder and harder and depending on what you
+
+03:06.560 --> 03:13.120
+ have to remember and so on as you consider longer durations. Whereas humans seem to be able to
+
+03:13.120 --> 03:18.080
+ do credit assignment through essentially arbitrary times like I could remember something I did last
+
+03:18.080 --> 03:23.360
+ year. And then now because I see some new evidence, I'm going to change my mind about
+
+03:23.360 --> 03:29.040
+ the way I was thinking last year, and hopefully not do the same mistake again.
+
+03:31.040 --> 03:36.800
+ I think a big part of that is probably forgetting. You're only remembering the really important
+
+03:36.800 --> 03:43.680
+ things that's very efficient forgetting. Yes. So there's a selection of what we remember.
+
+03:43.680 --> 03:49.120
+ And I think there are really cool connection to higher level cognitions here regarding
+
+03:49.120 --> 03:55.760
+ consciousness, deciding and emotions. So deciding what comes to consciousness and what gets stored
+
+03:55.760 --> 04:04.800
+ in memory, which are not trivial either. So you've been at the forefront there all along
+
+04:04.800 --> 04:10.800
+ showing some of the amazing things that neural networks, deep neural networks can do in the
+
+04:10.800 --> 04:16.560
+ field of artificial intelligence is just broadly in all kinds of applications. But we can talk
+
+04:16.560 --> 04:23.200
+ about that forever. But what in your view, because we're thinking towards the future is the weakest
+
+04:23.200 --> 04:29.120
+ aspect of the way deep neural networks represent the world. What is that? What is in your view
+
+04:29.120 --> 04:41.200
+ is missing? So current state of the art neural nets trained on large quantities of images or texts
+
+04:43.840 --> 04:49.760
+ have some level of understanding of what explains those data sets, but it's very
+
+04:49.760 --> 05:01.440
+ basic. It's very low level. And it's not nearly as robust and abstract and general as our understanding.
+
+05:02.960 --> 05:09.760
+ Okay, so that doesn't tell us how to fix things. But I think it encourages us to think about
+
+05:09.760 --> 05:21.200
+ how we can maybe train our neural nets differently, so that they would focus, for example, on causal
+
+05:21.200 --> 05:30.000
+ explanations, something that we don't do currently with neural net training. Also, one thing I'll
+
+05:30.000 --> 05:37.920
+ talk about in my talk this afternoon is instead of learning separately from images and videos on
+
+05:37.920 --> 05:45.600
+ one hand and from texts on the other hand, we need to do a better job of jointly learning about
+
+05:45.600 --> 05:54.320
+ language and about the world to which it refers. So that, you know, both sides can help each other.
+
+05:54.880 --> 06:02.480
+ We need to have good world models in our neural nets for them to really understand sentences
+
+06:02.480 --> 06:10.000
+ which talk about what's going on in the world. And I think we need language input to help
+
+06:10.640 --> 06:17.760
+ provide clues about what high level concepts like semantic concepts should be represented
+
+06:17.760 --> 06:26.400
+ at the top levels of these neural nets. In fact, there is evidence that the purely unsupervised
+
+06:26.400 --> 06:33.840
+ learning of representations doesn't give rise to high level representations that are as powerful
+
+06:33.840 --> 06:40.320
+ as the ones we're getting from supervised learning. And so the clues we're getting just with the labels,
+
+06:40.320 --> 06:46.960
+ not even sentences, is already very powerful. Do you think that's an architecture challenge
+
+06:46.960 --> 06:55.920
+ or is it a data set challenge? Neither. I'm tempted to just end it there.
+
+07:02.960 --> 07:06.800
+ Of course, data sets and architectures are something you want to always play with. But
+
+07:06.800 --> 07:13.040
+ I think the crucial thing is more the training objectives, the training frameworks. For example,
+
+07:13.040 --> 07:20.240
+ going from passive observation of data to more active agents, which
+
+07:22.320 --> 07:27.280
+ learn by intervening in the world, the relationships between causes and effects,
+
+07:28.480 --> 07:36.240
+ the sort of objective functions which could be important to allow the highest level
+
+07:36.240 --> 07:44.000
+ of explanations to rise from the learning, which I don't think we have now. The kinds of
+
+07:44.000 --> 07:50.320
+ objective functions which could be used to reward exploration, the right kind of exploration. So
+
+07:50.320 --> 07:56.160
+ these kinds of questions are neither in the data set nor in the architecture, but more in
+
+07:56.800 --> 08:03.920
+ how we learn under what objectives and so on. Yeah, that's a, I've heard you mention in several
+
+08:03.920 --> 08:08.080
+ contexts, the idea of sort of the way children learn, they interact with objects in the world.
+
+08:08.080 --> 08:15.040
+ And it seems fascinating because in some sense, except with some cases in reinforcement learning,
+
+08:15.760 --> 08:23.600
+ that idea is not part of the learning process in artificial neural networks. It's almost like
+
+08:24.320 --> 08:33.120
+ do you envision something like an objective function saying, you know what, if you poke this
+
+08:33.120 --> 08:38.800
+ object in this kind of way, it would be really helpful for me to further, further learn.
+
+08:39.920 --> 08:44.880
+ Sort of almost guiding some aspect of learning. Right, right, right. So I was talking to Rebecca
+
+08:44.880 --> 08:54.240
+ Sachs just an hour ago and she was talking about lots and lots of evidence from infants seem to
+
+08:54.240 --> 09:04.880
+ clearly pick what interests them in a directed way. And so they're not passive learners.
+
+09:04.880 --> 09:11.680
+ They, they focus their attention on aspects of the world, which are most interesting,
+
+09:11.680 --> 09:17.760
+ surprising in a non trivial way that makes them change their theories of the world.
+
+09:17.760 --> 09:29.120
+ So that's a fascinating view of the future progress. But on a more maybe boring question,
+
+09:30.000 --> 09:37.440
+ do you think going deeper and larger? So do you think just increasing the size of the things
+
+09:37.440 --> 09:43.520
+ that have been increasing a lot in the past few years will, will also make significant progress?
+
+09:43.520 --> 09:49.760
+ So some of the representational issues that you, you mentioned, they're kind of shallow
+
+09:50.560 --> 09:54.880
+ in some sense. Oh, you mean in the sense of abstraction,
+
+09:54.880 --> 09:59.040
+ abstract in the sense of abstraction, they're not getting some, I don't think that having
+
+10:00.400 --> 10:05.520
+ more depth in the network in the sense of instead of 100 layers, we have 10,000 is going to solve
+
+10:05.520 --> 10:13.120
+ our problem. You don't think so? Is that obvious to you? Yes. What is clear to me is that
+
+10:13.120 --> 10:21.600
+ engineers and companies and labs, grad students will continue to tune architectures and explore
+
+10:21.600 --> 10:27.520
+ all kinds of tweaks to make the current state of the art slightly ever slightly better. But
+
+10:27.520 --> 10:31.840
+ I don't think that's going to be nearly enough. I think we need some fairly drastic changes in
+
+10:31.840 --> 10:39.680
+ the way that we're considering learning to achieve the goal that these learners actually
+
+10:39.680 --> 10:45.680
+ understand in a deep way the environment in which they are, you know, observing and acting.
+
+10:46.480 --> 10:51.920
+ But I guess I was trying to ask a question that's more interesting than just more layers
+
+10:53.040 --> 11:00.800
+ is basically once you figure out a way to learn through interacting, how many parameters does
+
+11:00.800 --> 11:07.760
+ it take to store that information? So I think our brain is quite bigger than most neural networks.
+
+11:07.760 --> 11:13.120
+ Right, right. Oh, I see what you mean. Oh, I'm with you there. So I agree that in order to
+
+11:14.240 --> 11:19.760
+ build neural nets with the kind of broad knowledge of the world that typical adult humans have,
+
+11:20.960 --> 11:24.880
+ probably the kind of computing power we have now is going to be insufficient.
+
+11:25.600 --> 11:30.320
+ So the good news is there are hardware companies building neural net chips. And so
+
+11:30.320 --> 11:39.280
+ it's going to get better. However, the good news in a way, which is also a bad news, is that even
+
+11:39.280 --> 11:47.840
+ our state of the art deep learning methods fail to learn models that understand even very simple
+
+11:47.840 --> 11:53.680
+ environments like some grid worlds that we have built. Even these fairly simple environments,
+
+11:53.680 --> 11:57.120
+ I mean, of course, if you train them with enough examples, eventually they get it,
+
+11:57.120 --> 12:05.200
+ but it's just like instead of what humans might need just dozens of examples, these things will
+
+12:05.200 --> 12:12.720
+ need millions, right, for very, very, very simple tasks. And so I think there's an opportunity
+
+12:13.520 --> 12:18.080
+ for academics who don't have the kind of computing power that say Google has
+
+12:19.280 --> 12:25.360
+ to do really important and exciting research to advance the state of the art in training
+
+12:25.360 --> 12:32.720
+ frameworks, learning models, agent learning in even simple environments that are synthetic,
+
+12:33.440 --> 12:37.200
+ that seem trivial, but yet current machine learning fails on.
+
+12:38.240 --> 12:48.240
+ We talked about priors and common sense knowledge. It seems like we humans take a lot of knowledge
+
+12:48.240 --> 12:57.040
+ for granted. So what's your view of these priors of forming this broad view of the world, this
+
+12:57.040 --> 13:02.560
+ accumulation of information, and how we can teach neural networks or learning systems to pick that
+
+13:02.560 --> 13:10.880
+ knowledge up? So knowledge, you know, for a while, the artificial intelligence, maybe in the 80,
+
+13:10.880 --> 13:16.880
+ like there's a time where knowledge representation, knowledge, acquisition, expert systems, I mean,
+
+13:16.880 --> 13:24.080
+ though, the symbolic AI was a view, was an interesting problem set to solve. And it was kind
+
+13:24.080 --> 13:29.440
+ of put on hold a little bit, it seems like because it doesn't work. It doesn't work. That's right.
+
+13:29.440 --> 13:37.840
+ But that's right. But the goals of that remain important. Yes, remain important. And how do you
+
+13:37.840 --> 13:45.920
+ think those goals can be addressed? Right. So first of all, I believe that one reason why the
+
+13:45.920 --> 13:52.560
+ classical expert systems approach failed is because a lot of the knowledge we have, so you talked
+
+13:52.560 --> 14:01.760
+ about common sense and tuition, there's a lot of knowledge like this, which is not consciously
+
+14:01.760 --> 14:06.320
+ accessible. There are lots of decisions we're taking that we can't really explain, even if
+
+14:06.320 --> 14:16.160
+ sometimes we make up a story. And that knowledge is also necessary for machines to take good
+
+14:16.160 --> 14:22.320
+ decisions. And that knowledge is hard to codify in expert systems, rule based systems, and, you
+
+14:22.320 --> 14:27.920
+ know, classical AI formalism. And there are other issues, of course, with the old AI, like,
+
+14:29.680 --> 14:34.320
+ not really good ways of handling uncertainty, I would say something more subtle,
+
+14:34.320 --> 14:40.480
+ which we understand better now, but I think still isn't enough in the minds of people.
+
+14:41.360 --> 14:48.480
+ There's something really powerful that comes from distributed representations, the thing that really
+
+14:49.120 --> 14:58.480
+ makes neural nets work so well. And it's hard to replicate that kind of power in a symbolic world.
+
+14:58.480 --> 15:05.200
+ The knowledge in expert systems and so on is nicely decomposed into like a bunch of rules.
+
+15:05.760 --> 15:11.280
+ Whereas if you think about a neural net, it's the opposite. You have this big blob of parameters
+
+15:11.280 --> 15:16.480
+ which work intensely together to represent everything the network knows. And it's not
+
+15:16.480 --> 15:22.880
+ sufficiently factorized. And so I think this is one of the weaknesses of current neural nets,
+
+15:22.880 --> 15:30.080
+ that we have to take lessons from classical AI in order to bring in another kind of
+
+15:30.080 --> 15:35.920
+ compositionality, which is common in language, for example, and in these rules. But that isn't
+
+15:35.920 --> 15:45.040
+ so native to neural nets. And on that line of thinking, disentangled representations. Yes. So
+
+15:46.320 --> 15:51.680
+ let me connect with disentangled representations. If you might, if you don't mind. Yes, exactly.
+
+15:51.680 --> 15:58.080
+ Yeah. So for many years, I thought, and I still believe that it's really important that we come
+
+15:58.080 --> 16:04.080
+ up with learning algorithms, either unsupervised or supervised, but reinforcement, whatever,
+
+16:04.720 --> 16:11.600
+ that build representations in which the important factors, hopefully causal factors are nicely
+
+16:11.600 --> 16:16.240
+ separated and easy to pick up from the representation. So that's the idea of disentangled
+
+16:16.240 --> 16:22.560
+ representations. It says transfer the data into a space where everything becomes easy, we can maybe
+
+16:22.560 --> 16:29.360
+ just learn with linear models about the things we care about. And I still think this is important,
+
+16:29.360 --> 16:36.880
+ but I think this is missing out on a very important ingredient, which classical AI systems can remind
+
+16:36.880 --> 16:41.920
+ us of. So let's say we have these disentangled representations, you still need to learn about
+
+16:41.920 --> 16:47.120
+ the, the relationships between the variables, those high level semantic variables, they're not
+
+16:47.120 --> 16:52.000
+ going to be independent. I mean, this is like too much of an assumption. They're going to have some
+
+16:52.000 --> 16:56.400
+ interesting relationships that allow to predict things in the future to explain what happened in
+
+16:56.400 --> 17:01.840
+ the past. The kind of knowledge about those relationships in a classical AI system is
+
+17:01.840 --> 17:06.640
+ encoded in the rules, like a rule is just like a little piece of knowledge that says, oh, I have
+
+17:06.640 --> 17:12.160
+ these two, three, four variables that are linked in this interesting way. Then I can say something
+
+17:12.160 --> 17:17.280
+ about one or two of them given a couple of others, right? In addition to disentangling the,
+
+17:18.880 --> 17:23.520
+ the elements of the representation, which are like the variables in a rule based system,
+
+17:24.080 --> 17:33.200
+ you also need to disentangle the, the mechanisms that relate those variables to each other.
+
+17:33.200 --> 17:37.760
+ So like the rules. So if the rules are neatly separated, like each rule is, you know, living
+
+17:37.760 --> 17:44.960
+ on its own. And when I, I change a rule because I'm learning, it doesn't need to break other rules.
+
+17:44.960 --> 17:49.280
+ Whereas current neural nets, for example, are very sensitive to what's called catastrophic
+
+17:49.280 --> 17:54.800
+ forgetting, where after I've learned some things, and then they learn new things, they can destroy
+
+17:54.800 --> 18:00.480
+ the old things that I had learned, right? If the knowledge was better factorized and, and
+
+18:00.480 --> 18:08.240
+ and separated disentangled, then you would avoid a lot of that. Now you can't do this in the
+
+18:08.880 --> 18:17.200
+ sensory domain, but my idea in like a pixel space, but, but my idea is that when you project the
+
+18:17.200 --> 18:22.560
+ data in the right semantic space, it becomes possible to now represent this extra knowledge
+
+18:23.440 --> 18:27.760
+ beyond the transformation from input to representations, which is how representations
+
+18:27.760 --> 18:33.120
+ act on each other and predict the future and so on, in a way that can be neatly
+
+18:34.560 --> 18:38.560
+ disentangled. So now it's the rules that are disentangled from each other and not just the
+
+18:38.560 --> 18:43.680
+ variables that are disentangled from each other. And you draw distinction between semantic space
+
+18:43.680 --> 18:48.400
+ and pixel, like, does there need to be an architectural difference? Well, yeah. So, so
+
+18:48.400 --> 18:51.840
+ there's the sensory space like pixels, which where everything is entangled,
+
+18:51.840 --> 18:58.000
+ and the information, like the variables are completely interdependent in very complicated
+
+18:58.000 --> 19:03.760
+ ways. And also computation, like the, it's not just variables, it's also how they are
+
+19:03.760 --> 19:10.240
+ related to each other is, is all intertwined. But, but I'm hypothesizing that in the right
+
+19:10.240 --> 19:16.800
+ high level representation space, both the variables and how they relate to each other
+
+19:16.800 --> 19:22.960
+ can be disentangled and that will provide a lot of generalization power. Generalization power.
+
+19:22.960 --> 19:29.760
+ Yes. Distribution of the test set, it's assumed to be the same as a distribution of the training
+
+19:29.760 --> 19:36.640
+ set. Right. This is where current machine learning is too weak. It doesn't tell us anything,
+
+19:36.640 --> 19:41.120
+ is not able to tell us anything about how our neural nets, say, are going to generalize to a
+
+19:41.120 --> 19:46.160
+ new distribution. And, and, you know, people may think, well, but there's nothing we can say if
+
+19:46.160 --> 19:51.840
+ we don't know what the new distribution will be. The truth is, humans are able to generalize to
+
+19:51.840 --> 19:56.560
+ new distributions. Yeah, how are we able to do that? So yeah, because there is something, these
+
+19:56.560 --> 20:00.720
+ new distributions, even though they could look very different from the training distributions,
+
+20:01.520 --> 20:05.360
+ they have things in common. So let me give you a concrete example. You read a science fiction
+
+20:05.360 --> 20:12.560
+ novel, the science fiction novel, maybe, you know, brings you in some other planet where
+
+20:12.560 --> 20:17.760
+ things look very different on the surface, but it's still the same laws of physics.
+
+20:18.560 --> 20:21.440
+ All right. And so you can read the book and you understand what's going on.
+
+20:22.960 --> 20:29.200
+ So the distribution is very different. But because you can transport a lot of the knowledge you had
+
+20:29.200 --> 20:35.680
+ from Earth about the underlying cause and effect relationships and physical mechanisms and all
+
+20:35.680 --> 20:40.880
+ that, and maybe even social interactions, you can now make sense of what is going on on this
+
+20:40.880 --> 20:43.920
+ planet where like visually, for example, things are totally different.
+
+20:45.920 --> 20:52.000
+ Taking that analogy further and distorting it, let's enter a science fiction world of, say,
+
+20:52.000 --> 21:00.720
+ Space Odyssey 2001 with Hal. Yeah. Or maybe, which is probably one of my favorite AI movies.
+
+21:00.720 --> 21:06.080
+ Me too. And then there's another one that a lot of people love that may be a little bit outside
+
+21:06.080 --> 21:13.120
+ of the AI community is Ex Machina. I don't know if you've seen it. Yes. By the way, what are your
+
+21:13.120 --> 21:19.600
+ reviews on that movie? Are you able to enjoy it? So there are things I like and things I hate.
+
+21:21.120 --> 21:25.760
+ So let me, you could talk about that in the context of a question I want to ask,
+
+21:25.760 --> 21:31.920
+ which is there's quite a large community of people from different backgrounds off and outside of AI
+
+21:31.920 --> 21:36.480
+ who are concerned about existential threat of artificial intelligence. Right. You've seen
+
+21:36.480 --> 21:41.920
+ now this community develop over time. You've seen you have a perspective. So what do you think is
+
+21:41.920 --> 21:47.680
+ the best way to talk about AI safety, to think about it, to have discourse about it within AI
+
+21:47.680 --> 21:53.920
+ community and outside and grounded in the fact that Ex Machina is one of the main sources of
+
+21:53.920 --> 21:59.040
+ information for the general public about AI. So I think you're putting it right. There's a big
+
+21:59.040 --> 22:04.400
+ difference between the sort of discussion we ought to have within the AI community
+
+22:05.200 --> 22:11.600
+ and the sort of discussion that really matter in the general public. So I think the picture of
+
+22:11.600 --> 22:19.040
+ Terminator and, you know, AI loose and killing people and super intelligence that's going to
+
+22:19.040 --> 22:26.320
+ destroy us, whatever we try, isn't really so useful for the public discussion because
+
+22:26.320 --> 22:32.960
+ for the public discussion that things I believe really matter are the short term and
+
+22:32.960 --> 22:40.560
+ mini term, very likely negative impacts of AI on society, whether it's from security,
+
+22:40.560 --> 22:45.680
+ like, you know, big brother scenarios with face recognition or killer robots, or the impact on
+
+22:45.680 --> 22:52.400
+ the job market, or concentration of power and discrimination, all kinds of social issues,
+
+22:52.400 --> 22:58.240
+ which could actually, some of them could really threaten democracy, for example.
+
+22:58.800 --> 23:04.000
+ Just to clarify, when you said killer robots, you mean autonomous weapons as a weapon system?
+
+23:04.000 --> 23:10.400
+ Yes, I don't mean, no, that's right. So I think these short and medium term concerns
+
+23:11.280 --> 23:18.560
+ should be important parts of the public debate. Now, existential risk, for me, is a very unlikely
+
+23:18.560 --> 23:26.880
+ consideration, but still worth academic investigation. In the same way that you could say,
+
+23:26.880 --> 23:32.640
+ should we study what could happen if meteorite, you know, came to earth and destroyed it.
+
+23:32.640 --> 23:37.680
+ So I think it's very unlikely that this is going to happen in or happen in a reasonable future.
+
+23:37.680 --> 23:45.520
+ It's very, the sort of scenario of an AI getting loose goes against my understanding of at least
+
+23:45.520 --> 23:50.160
+ current machine learning and current neural nets and so on. It's not plausible to me.
+
+23:50.160 --> 23:54.320
+ But of course, I don't have a crystal ball and who knows what AI will be in 50 years from now.
+
+23:54.320 --> 23:59.280
+ So I think it is worth that scientists study those problems. It's just not a pressing question,
+
+23:59.280 --> 24:04.880
+ as far as I'm concerned. So before I continue down that line, I have a few questions there, but
+
+24:06.640 --> 24:11.440
+ what do you like and not like about X Machina as a movie? Because I actually watched it for the
+
+24:11.440 --> 24:17.840
+ second time and enjoyed it. I hated it the first time and I enjoyed it quite a bit more the second
+
+24:17.840 --> 24:26.080
+ time when I sort of learned to accept certain pieces of it. See it as a concept movie. What
+
+24:26.080 --> 24:36.160
+ was your experience? What were your thoughts? So the negative is the picture it paints of science
+
+24:36.160 --> 24:41.760
+ is totally wrong. Science in general and AI in particular. Science is not happening
+
+24:43.120 --> 24:51.840
+ in some hidden place by some really smart guy. One person. One person. This is totally unrealistic.
+
+24:51.840 --> 24:58.240
+ This is not how it happens. Even a team of people in some isolated place will not make it.
+
+24:58.240 --> 25:07.920
+ Science moves by small steps thanks to the collaboration and community of a large number
+
+25:07.920 --> 25:16.000
+ of people interacting and all the scientists who are expert in their field kind of know what is
+
+25:16.000 --> 25:24.000
+ going on even in the industrial labs. Information flows and leaks and so on. And the spirit of
+
+25:24.000 --> 25:30.320
+ it is very different from the way science is painted in this movie. Yeah, let me ask on that
+
+25:30.320 --> 25:36.400
+ point. It's been the case to this point that kind of even if the research happens inside
+
+25:36.400 --> 25:42.000
+ Google or Facebook, inside companies, it still kind of comes out. Do you think that will always be
+
+25:42.000 --> 25:48.960
+ the case with AI? Is it possible to bottle ideas to the point where there's a set of breakthroughs
+
+25:48.960 --> 25:53.120
+ that go completely undiscovered by the general research community? Do you think that's even
+
+25:53.120 --> 26:02.240
+ possible? It's possible, but it's unlikely. It's not how it is done now. It's not how I can force
+
+26:02.240 --> 26:13.120
+ it in in the foreseeable future. But of course, I don't have a crystal ball. And so who knows,
+
+26:13.120 --> 26:18.240
+ this is science fiction after all. But but usually ominous that the lights went off during
+
+26:18.240 --> 26:24.320
+ during that discussion. So the problem again, there's a you know, one thing is the movie and
+
+26:24.320 --> 26:28.720
+ you could imagine all kinds of science fiction. The problem with for me, maybe similar to the
+
+26:28.720 --> 26:37.120
+ question about existential risk is that this kind of movie paints such a wrong picture of what is
+
+26:37.120 --> 26:43.520
+ actual, you know, the actual science and how it's going on that that it can have unfortunate effects
+
+26:43.520 --> 26:49.040
+ on people's understanding of current science. And so that's kind of sad.
+
+26:50.560 --> 26:56.800
+ There's an important principle in research, which is diversity. So in other words,
+
+26:58.000 --> 27:02.720
+ research is exploration, research is exploration in the space of ideas. And different people
+
+27:03.440 --> 27:09.920
+ will focus on different directions. And this is not just good, it's essential. So I'm totally fine
+
+27:09.920 --> 27:16.640
+ with people exploring directions that are contrary to mine or look orthogonal to mine.
+
+27:18.560 --> 27:24.880
+ I am more than fine, I think it's important. I and my friends don't claim we have universal
+
+27:24.880 --> 27:29.680
+ truth about what will especially about what will happen in the future. Now that being said,
+
+27:30.320 --> 27:37.600
+ we have our intuitions and then we act accordingly, according to where we think we can be most useful
+
+27:37.600 --> 27:43.360
+ and where society has the most to gain or to lose. We should have those debates and
+
+27:45.920 --> 27:50.080
+ and not end up in a society where there's only one voice and one way of thinking and
+
+27:51.360 --> 27:59.120
+ research money is spread out. So this agreement is a sign of good research, good science. So
+
+27:59.120 --> 28:08.560
+ yes. The idea of bias in the human sense of bias. How do you think about instilling in machine
+
+28:08.560 --> 28:15.440
+ learning something that's aligned with human values in terms of bias? We intuitively assume
+
+28:15.440 --> 28:21.680
+ beings have a concept of what bias means, of what fundamental respect for other human beings means,
+
+28:21.680 --> 28:25.280
+ but how do we instill that into machine learning systems, do you think?
+
+28:25.280 --> 28:32.720
+ So I think there are short term things that are already happening and then there are long term
+
+28:32.720 --> 28:39.040
+ things that we need to do. In the short term, there are techniques that have been proposed and
+
+28:39.040 --> 28:44.800
+ I think will continue to be improved and maybe alternatives will come up to take data sets
+
+28:45.600 --> 28:51.200
+ in which we know there is bias, we can measure it. Pretty much any data set where humans are
+
+28:51.200 --> 28:56.080
+ being observed taking decisions will have some sort of bias discrimination against particular
+
+28:56.080 --> 29:04.000
+ groups and so on. And we can use machine learning techniques to try to build predictors, classifiers
+
+29:04.000 --> 29:11.920
+ that are going to be less biased. We can do it for example using adversarial methods to make our
+
+29:11.920 --> 29:19.520
+ systems less sensitive to these variables we should not be sensitive to. So these are clear,
+
+29:19.520 --> 29:24.240
+ well defined ways of trying to address the problem, maybe they have weaknesses and more
+
+29:24.240 --> 29:30.400
+ research is needed and so on, but I think in fact they're sufficiently mature that governments should
+
+29:30.400 --> 29:36.160
+ start regulating companies where it matters say like insurance companies so that they use those
+
+29:36.160 --> 29:43.840
+ techniques because those techniques will probably reduce the bias, but at a cost for example maybe
+
+29:43.840 --> 29:47.920
+ their predictions will be less accurate and so companies will not do it until you force them.
+
+29:47.920 --> 29:56.000
+ All right, so this is short term. Long term, I'm really interested in thinking how we can
+
+29:56.000 --> 30:02.160
+ instill moral values into computers. Obviously this is not something we'll achieve in the next five
+
+30:02.160 --> 30:11.680
+ or 10 years. There's already work in detecting emotions for example in images and sounds and
+
+30:11.680 --> 30:21.520
+ texts and also studying how different agents interacting in different ways may correspond to
+
+30:22.960 --> 30:30.000
+ patterns of say injustice which could trigger anger. So these are things we can do in the
+
+30:30.000 --> 30:42.160
+ medium term and eventually train computers to model for example how humans react emotionally. I would
+
+30:42.160 --> 30:49.920
+ say the simplest thing is unfair situations which trigger anger. This is one of the most basic
+
+30:49.920 --> 30:55.360
+ emotions that we share with other animals. I think it's quite feasible within the next few years so
+
+30:55.360 --> 31:00.800
+ we can build systems that can detect these kind of things to the extent unfortunately that they
+
+31:00.800 --> 31:07.840
+ understand enough about the world around us which is a long time away but maybe we can initially do
+
+31:07.840 --> 31:14.800
+ this in virtual environments so you can imagine like a video game where agents interact in some
+
+31:14.800 --> 31:21.760
+ ways and then some situations trigger an emotion. I think we could train machines to detect those
+
+31:21.760 --> 31:27.920
+ situations and predict that the particular emotion will likely be felt if a human was playing one
+
+31:27.920 --> 31:34.080
+ of the characters. You have shown excitement and done a lot of excellent work with unsupervised
+
+31:34.080 --> 31:42.800
+ learning but there's been a lot of success on the supervised learning. One of the things I'm
+
+31:42.800 --> 31:48.800
+ really passionate about is how humans and robots work together and in the context of supervised
+
+31:48.800 --> 31:54.800
+ learning that means the process of annotation. Do you think about the problem of annotation of
+
+31:55.520 --> 32:04.080
+ put in a more interesting way is humans teaching machines? Yes, I think it's an important subject.
+
+32:04.880 --> 32:11.280
+ Reducing it to annotation may be useful for somebody building a system tomorrow but
+
+32:12.560 --> 32:17.600
+ longer term the process of teaching I think is something that deserves a lot more attention
+
+32:17.600 --> 32:21.840
+ from the machine learning community so there are people of coin the term machine teaching.
+
+32:22.560 --> 32:30.480
+ So what are good strategies for teaching a learning agent and can we design, train a system
+
+32:30.480 --> 32:38.000
+ that is going to be a good teacher? So in my group we have a project called a BBI or BBI game
+
+32:38.640 --> 32:46.000
+ where there is a game or a scenario where there's a learning agent and a teaching agent
+
+32:46.000 --> 32:54.400
+ presumably the teaching agent would eventually be a human but we're not there yet and the
+
+32:56.000 --> 33:00.880
+ role of the teacher is to use its knowledge of the environment which it can acquire using
+
+33:00.880 --> 33:09.680
+ whatever way brute force to help the learner learn as quickly as possible. So the learner
+
+33:09.680 --> 33:13.920
+ is going to try to learn by itself maybe using some exploration and whatever
+
+33:13.920 --> 33:21.520
+ but the teacher can choose, can have an influence on the interaction with the learner
+
+33:21.520 --> 33:28.960
+ so as to guide the learner maybe teach it the things that the learner has most trouble with
+
+33:28.960 --> 33:34.320
+ or just add the boundary between what it knows and doesn't know and so on. So there's a tradition
+
+33:34.320 --> 33:41.280
+ of these kind of ideas from other fields and like tutorial systems for example and AI
+
+33:41.280 --> 33:46.880
+ and of course people in the humanities have been thinking about these questions but I think
+
+33:46.880 --> 33:52.560
+ it's time that machine learning people look at this because in the future we'll have more and more
+
+33:53.760 --> 33:59.680
+ human machine interaction with the human in the loop and I think understanding how to make this
+
+33:59.680 --> 34:04.080
+ work better. Oh the problems around that are very interesting and not sufficiently addressed.
+
+34:04.080 --> 34:11.440
+ You've done a lot of work with language too, what aspect of the traditionally formulated
+
+34:11.440 --> 34:17.040
+ touring test, a test of natural language understanding in generation in your eyes is the
+
+34:17.040 --> 34:22.960
+ most difficult of conversation, what in your eyes is the hardest part of conversation to solve for
+
+34:22.960 --> 34:30.640
+ machines. So I would say it's everything having to do with the non linguistic knowledge which
+
+34:30.640 --> 34:36.400
+ implicitly you need in order to make sense of sentences. Things like the winner grad schemas
+
+34:36.400 --> 34:42.400
+ so these sentences that are semantically ambiguous. In other words you need to understand enough about
+
+34:42.400 --> 34:48.720
+ the world in order to really interpret properly those sentences. I think these are interesting
+
+34:48.720 --> 34:55.840
+ challenges for machine learning because they point in the direction of building systems that
+
+34:55.840 --> 35:02.880
+ both understand how the world works and there's causal relationships in the world and associate
+
+35:03.520 --> 35:09.760
+ that knowledge with how to express it in language either for reading or writing.
+
+35:11.840 --> 35:17.600
+ You speak French? Yes, it's my mother tongue. It's one of the romance languages. Do you think
+
+35:17.600 --> 35:23.040
+ passing the touring test and all the underlying challenges we just mentioned depend on language?
+
+35:23.040 --> 35:28.000
+ Do you think it might be easier in French than it is in English or is independent of language?
+
+35:28.800 --> 35:37.680
+ I think it's independent of language. I would like to build systems that can use the same
+
+35:37.680 --> 35:45.840
+ principles, the same learning mechanisms to learn from human agents, whatever their language.
+
+35:45.840 --> 35:53.600
+ Well, certainly us humans can talk more beautifully and smoothly in poetry. So I'm Russian originally.
+
+35:53.600 --> 36:01.360
+ I know poetry in Russian is maybe easier to convey complex ideas than it is in English
+
+36:02.320 --> 36:09.520
+ but maybe I'm showing my bias and some people could say that about French. But of course the
+
+36:09.520 --> 36:16.400
+ goal ultimately is our human brain is able to utilize any kind of those languages to use them
+
+36:16.400 --> 36:21.040
+ as tools to convey meaning. Yeah, of course there are differences between languages and maybe some
+
+36:21.040 --> 36:25.920
+ are slightly better at some things but in the grand scheme of things where we're trying to understand
+
+36:25.920 --> 36:31.040
+ how the brain works and language and so on, I think these differences are minute.
+
+36:31.040 --> 36:42.880
+ So you've lived perhaps through an AI winter of sorts. Yes. How did you stay warm and continue
+
+36:42.880 --> 36:48.480
+ with your research? Stay warm with friends. With friends. Okay, so it's important to have friends
+
+36:48.480 --> 36:57.200
+ and what have you learned from the experience? Listen to your inner voice. Don't, you know, be
+
+36:57.200 --> 37:07.680
+ trying to just please the crowds and the fashion and if you have a strong intuition about something
+
+37:08.480 --> 37:15.520
+ that is not contradicted by actual evidence, go for it. I mean, it could be contradicted by people.
+
+37:16.960 --> 37:21.920
+ Not your own instinct of based on everything you've learned. So of course you have to adapt
+
+37:21.920 --> 37:29.440
+ your beliefs when your experiments contradict those beliefs but you have to stick to your
+
+37:29.440 --> 37:36.160
+ beliefs otherwise. It's what allowed me to go through those years. It's what allowed me to
+
+37:37.120 --> 37:44.480
+ persist in directions that, you know, took time, whatever other people think, took time to mature
+
+37:44.480 --> 37:53.680
+ and bring fruits. So history of AI is marked with these, of course it's marked with technical
+
+37:53.680 --> 37:58.880
+ breakthroughs but it's also marked with these seminal events that capture the imagination
+
+37:58.880 --> 38:06.000
+ of the community. Most recent, I would say AlphaGo beating the world champion human go player
+
+38:06.000 --> 38:14.000
+ was one of those moments. What do you think the next such moment might be? Okay, sir, first of all,
+
+38:14.000 --> 38:24.880
+ I think that these so called seminal events are overrated. As I said, science really moves by
+
+38:24.880 --> 38:33.760
+ small steps. Now what happens is you make one more small step and it's like the drop that,
+
+38:33.760 --> 38:40.560
+ you know, allows to, that fills the bucket and then you have drastic consequences because now
+
+38:40.560 --> 38:46.240
+ you're able to do something you were not able to do before or now say the cost of building some
+
+38:46.240 --> 38:51.920
+ device or solving a problem becomes cheaper than what existed and you have a new market that opens
+
+38:51.920 --> 39:00.080
+ up. So especially in the world of commerce and applications, the impact of a small scientific
+
+39:00.080 --> 39:07.520
+ progress could be huge but in the science itself, I think it's very, very gradual and
+
+39:07.520 --> 39:15.280
+ where are these steps being taken now? So there's unsupervised, right? So if I look at one trend
+
+39:15.280 --> 39:24.080
+ that I like in my community, for example, and at me line, my institute, what are the two hardest
+
+39:24.080 --> 39:32.800
+ topics? GANs and reinforcement learning, even though in Montreal in particular, like reinforcement
+
+39:32.800 --> 39:39.600
+ learning was something pretty much absent just two or three years ago. So it is really a big
+
+39:39.600 --> 39:48.400
+ interest from students and there's a big interest from people like me. So I would say this is
+
+39:48.400 --> 39:54.960
+ something where we're going to see more progress even though it hasn't yet provided much in terms of
+
+39:54.960 --> 40:01.280
+ actual industrial fallout. Like even though there's Alpha Gold, there's no, like Google is not making
+
+40:01.280 --> 40:06.320
+ money on this right now. But I think over the long term, this is really, really important for many
+
+40:06.320 --> 40:13.760
+ reasons. So in other words, I would say reinforcement learning maybe more generally agent learning
+
+40:13.760 --> 40:17.520
+ because it doesn't have to be with rewards. It could be in all kinds of ways that an agent
+
+40:17.520 --> 40:23.040
+ is learning about its environment. Now, reinforcement learning, you're excited about. Do you think
+
+40:23.040 --> 40:32.320
+ GANs could provide something? Yes. Some moment in it. Well, GANs or other
+
+40:33.760 --> 40:41.360
+ generative models, I believe, will be crucial ingredients in building agents that can understand
+
+40:41.360 --> 40:48.880
+ the world. A lot of the successes in reinforcement learning in the past has been with policy
+
+40:48.880 --> 40:53.360
+ gradient where you'll just learn a policy. You don't actually learn a model of the world. But
+
+40:53.360 --> 40:58.640
+ there are lots of issues with that. And we don't know how to do model based RL right now. But I
+
+40:58.640 --> 41:06.080
+ think this is where we have to go in order to build models that can generalize faster and better,
+
+41:06.080 --> 41:13.200
+ like to new distributions that capture, to some extent, at least the underlying causal
+
+41:13.200 --> 41:20.320
+ mechanisms in the world. Last question. What made you fall in love with artificial intelligence?
+
+41:20.960 --> 41:28.400
+ If you look back, what was the first moment in your life when you were fascinated by either
+
+41:28.400 --> 41:33.600
+ the human mind or the artificial mind? You know, when I was an adolescent, I was reading a lot.
+
+41:33.600 --> 41:41.920
+ And then I started reading science fiction. There you go. That's it. That's where I got hooked.
+
+41:41.920 --> 41:50.160
+ And then, you know, I had one of the first personal computers and I got hooked in programming.
+
+41:50.960 --> 41:55.040
+ And so it just, you know, start with fiction and then make it a reality. That's right.
+
+41:55.040 --> 42:12.080
+ Yosha, thank you so much for talking to me. My pleasure.
+

vtt/episode_005_small.vtt

@@ -0,0 +1,2654 @@
+WEBVTT
+
+00:00.000 --> 00:03.040
+ The following is a conversation with Vladimir Vapnik.
+
+00:03.040 --> 00:05.280
+ He's the coinventor of the Support Vector Machines,
+
+00:05.280 --> 00:07.920
+ Support Vector Clustering, VC Theory,
+
+00:07.920 --> 00:11.200
+ and many foundational ideas in statistical learning.
+
+00:11.200 --> 00:13.640
+ He was born in the Soviet Union and worked
+
+00:13.640 --> 00:16.320
+ at the Institute of Control Sciences in Moscow.
+
+00:16.320 --> 00:20.640
+ Then in the United States, he worked at AT&T, NEC Labs,
+
+00:20.640 --> 00:24.280
+ Facebook Research, and now as a professor at Columbia
+
+00:24.280 --> 00:25.960
+ University.
+
+00:25.960 --> 00:30.320
+ His work has been cited over 170,000 times.
+
+00:30.320 --> 00:31.840
+ He has some very interesting ideas
+
+00:31.840 --> 00:34.800
+ about artificial intelligence and the nature of learning,
+
+00:34.800 --> 00:37.600
+ especially on the limits of our current approaches
+
+00:37.600 --> 00:40.440
+ and the open problems in the field.
+
+00:40.440 --> 00:42.520
+ This conversation is part of MIT course
+
+00:42.520 --> 00:44.440
+ on artificial general intelligence
+
+00:44.440 --> 00:46.840
+ and the Artificial Intelligence Podcast.
+
+00:46.840 --> 00:49.600
+ If you enjoy it, please subscribe on YouTube
+
+00:49.600 --> 00:53.040
+ or rate it on iTunes or your podcast provider of choice
+
+00:53.040 --> 00:55.320
+ or simply connect with me on Twitter
+
+00:55.320 --> 01:00.200
+ or other social networks at Lex Friedman, spelled F R I D.
+
+01:00.200 --> 01:04.800
+ And now here's my conversation with Vladimir Vapnik.
+
+01:04.800 --> 01:08.840
+ Einstein famously said that God doesn't play dice.
+
+01:08.840 --> 01:10.000
+ Yeah.
+
+01:10.000 --> 01:12.880
+ You have studied the world through the eyes of statistics.
+
+01:12.880 --> 01:17.320
+ So let me ask you, in terms of the nature of reality,
+
+01:17.320 --> 01:21.360
+ fundamental nature of reality, does God play dice?
+
+01:21.360 --> 01:26.200
+ We don't know some factors, and because we
+
+01:26.200 --> 01:30.520
+ don't know some factors, which could be important,
+
+01:30.520 --> 01:38.000
+ it looks like God play dice, but we should describe it.
+
+01:38.000 --> 01:42.080
+ In philosophy, they distinguish between two positions,
+
+01:42.080 --> 01:45.480
+ positions of instrumentalism, where
+
+01:45.480 --> 01:48.720
+ you're creating theory for prediction
+
+01:48.720 --> 01:51.400
+ and position of realism, where you're
+
+01:51.400 --> 01:54.640
+ trying to understand what God's big.
+
+01:54.640 --> 01:56.800
+ Can you describe instrumentalism and realism
+
+01:56.800 --> 01:58.400
+ a little bit?
+
+01:58.400 --> 02:06.320
+ For example, if you have some mechanical laws, what is that?
+
+02:06.320 --> 02:11.480
+ Is it law which is true always and everywhere?
+
+02:11.480 --> 02:14.880
+ Or it is law which allows you to predict
+
+02:14.880 --> 02:22.920
+ the position of moving element, what you believe.
+
+02:22.920 --> 02:28.480
+ You believe that it is God's law, that God created the world,
+
+02:28.480 --> 02:33.160
+ which obeyed to this physical law,
+
+02:33.160 --> 02:36.240
+ or it is just law for predictions?
+
+02:36.240 --> 02:38.400
+ And which one is instrumentalism?
+
+02:38.400 --> 02:39.880
+ For predictions.
+
+02:39.880 --> 02:45.400
+ If you believe that this is law of God, and it's always
+
+02:45.400 --> 02:50.040
+ true everywhere, that means that you're a realist.
+
+02:50.040 --> 02:55.480
+ So you're trying to really understand that God's thought.
+
+02:55.480 --> 03:00.040
+ So the way you see the world as an instrumentalist?
+
+03:00.040 --> 03:03.240
+ You know, I'm working for some models,
+
+03:03.240 --> 03:06.960
+ model of machine learning.
+
+03:06.960 --> 03:12.760
+ So in this model, we can see setting,
+
+03:12.760 --> 03:16.440
+ and we try to solve, resolve the setting,
+
+03:16.440 --> 03:18.240
+ to solve the problem.
+
+03:18.240 --> 03:20.760
+ And you can do it in two different ways,
+
+03:20.760 --> 03:23.840
+ from the point of view of instrumentalists.
+
+03:23.840 --> 03:27.120
+ And that's what everybody does now,
+
+03:27.120 --> 03:31.560
+ because they say that the goal of machine learning
+
+03:31.560 --> 03:36.800
+ is to find the rule for classification.
+
+03:36.800 --> 03:40.920
+ That is true, but it is an instrument for prediction.
+
+03:40.920 --> 03:46.160
+ But I can say the goal of machine learning
+
+03:46.160 --> 03:50.040
+ is to learn about conditional probability.
+
+03:50.040 --> 03:54.440
+ So how God played use, and is He play?
+
+03:54.440 --> 03:55.960
+ What is probability for one?
+
+03:55.960 --> 03:59.960
+ What is probability for another given situation?
+
+03:59.960 --> 04:02.600
+ But for prediction, I don't need this.
+
+04:02.600 --> 04:04.240
+ I need the rule.
+
+04:04.240 --> 04:08.480
+ But for understanding, I need conditional probability.
+
+04:08.480 --> 04:11.800
+ So let me just step back a little bit first to talk about.
+
+04:11.800 --> 04:13.960
+ You mentioned, which I read last night,
+
+04:13.960 --> 04:21.280
+ the parts of the 1960 paper by Eugene Wigner,
+
+04:21.280 --> 04:23.520
+ unreasonable effectiveness of mathematics
+
+04:23.520 --> 04:24.880
+ and natural sciences.
+
+04:24.880 --> 04:29.400
+ Such a beautiful paper, by the way.
+
+04:29.400 --> 04:34.480
+ It made me feel, to be honest, to confess my own work
+
+04:34.480 --> 04:38.400
+ in the past few years on deep learning, heavily applied.
+
+04:38.400 --> 04:40.320
+ It made me feel that I was missing out
+
+04:40.320 --> 04:43.960
+ on some of the beauty of nature in the way
+
+04:43.960 --> 04:45.560
+ that math can uncover.
+
+04:45.560 --> 04:50.360
+ So let me just step away from the poetry of that for a second.
+
+04:50.360 --> 04:53.040
+ How do you see the role of math in your life?
+
+04:53.040 --> 04:54.080
+ Is it a tool?
+
+04:54.080 --> 04:55.840
+ Is it poetry?
+
+04:55.840 --> 04:56.960
+ Where does it sit?
+
+04:56.960 --> 05:01.400
+ And does math for you have limits of what it can describe?
+
+05:01.400 --> 05:08.280
+ Some people saying that math is language which use God.
+
+05:08.280 --> 05:10.280
+ So I believe in that.
+
+05:10.280 --> 05:12.000
+ Speak to God or use God.
+
+05:12.000 --> 05:12.760
+ Or use God.
+
+05:12.760 --> 05:14.080
+ Use God.
+
+05:14.080 --> 05:15.560
+ Yeah.
+
+05:15.560 --> 05:25.680
+ So I believe that this article about unreasonable
+
+05:25.680 --> 05:29.960
+ effectiveness of math is that if you're
+
+05:29.960 --> 05:33.960
+ looking in mathematical structures,
+
+05:33.960 --> 05:37.720
+ they know something about reality.
+
+05:37.720 --> 05:42.480
+ And the most scientists from natural science,
+
+05:42.480 --> 05:48.440
+ they're looking on equation and trying to understand reality.
+
+05:48.440 --> 05:51.280
+ So the same in machine learning.
+
+05:51.280 --> 05:57.560
+ If you're trying very carefully look on all equations
+
+05:57.560 --> 06:00.640
+ which define conditional probability,
+
+06:00.640 --> 06:05.680
+ you can understand something about reality more
+
+06:05.680 --> 06:08.160
+ than from your fantasy.
+
+06:08.160 --> 06:12.480
+ So math can reveal the simple underlying principles
+
+06:12.480 --> 06:13.880
+ of reality, perhaps.
+
+06:13.880 --> 06:16.880
+ You know, what means simple?
+
+06:16.880 --> 06:20.320
+ It is very hard to discover them.
+
+06:20.320 --> 06:23.800
+ But then when you discover them and look at them,
+
+06:23.800 --> 06:27.440
+ you see how beautiful they are.
+
+06:27.440 --> 06:33.560
+ And it is surprising why people did not see that before.
+
+06:33.560 --> 06:37.480
+ You're looking on equation and derive it from equations.
+
+06:37.480 --> 06:43.360
+ For example, I talked yesterday about least squirmated.
+
+06:43.360 --> 06:48.120
+ And people had a lot of fantasy have to improve least squirmated.
+
+06:48.120 --> 06:52.360
+ But if you're going step by step by solving some equations,
+
+06:52.360 --> 06:57.680
+ you suddenly will get some term which,
+
+06:57.680 --> 07:01.040
+ after thinking, you understand that it described
+
+07:01.040 --> 07:04.360
+ position of observation point.
+
+07:04.360 --> 07:08.240
+ In least squirmated, we throw out a lot of information.
+
+07:08.240 --> 07:11.760
+ We don't look in composition of point of observations.
+
+07:11.760 --> 07:14.600
+ We're looking only on residuals.
+
+07:14.600 --> 07:19.400
+ But when you understood that, that's a very simple idea.
+
+07:19.400 --> 07:22.320
+ But it's not too simple to understand.
+
+07:22.320 --> 07:25.680
+ And you can derive this just from equations.
+
+07:25.680 --> 07:28.120
+ So some simple algebra, a few steps
+
+07:28.120 --> 07:31.040
+ will take you to something surprising
+
+07:31.040 --> 07:34.360
+ that when you think about, you understand.
+
+07:34.360 --> 07:41.120
+ And that is proof that human intuition not to reach
+
+07:41.120 --> 07:42.640
+ and very primitive.
+
+07:42.640 --> 07:48.520
+ And it does not see very simple situations.
+
+07:48.520 --> 07:51.760
+ So let me take a step back in general.
+
+07:51.760 --> 07:54.480
+ Yes, right?
+
+07:54.480 --> 07:58.840
+ But what about human as opposed to intuition and ingenuity?
+
+08:01.600 --> 08:02.960
+ Moments of brilliance.
+
+08:02.960 --> 08:09.480
+ So do you have to be so hard on human intuition?
+
+08:09.480 --> 08:11.840
+ Are there moments of brilliance in human intuition?
+
+08:11.840 --> 08:17.520
+ They can leap ahead of math, and then the math will catch up?
+
+08:17.520 --> 08:19.400
+ I don't think so.
+
+08:19.400 --> 08:23.560
+ I think that the best human intuition,
+
+08:23.560 --> 08:26.440
+ it is putting in axioms.
+
+08:26.440 --> 08:28.600
+ And then it is technical.
+
+08:28.600 --> 08:31.880
+ See where the axioms take you.
+
+08:31.880 --> 08:34.920
+ But if they correctly take axioms,
+
+08:34.920 --> 08:41.400
+ but it axiom polished during generations of scientists.
+
+08:41.400 --> 08:45.040
+ And this is integral wisdom.
+
+08:45.040 --> 08:47.480
+ So that's beautifully put.
+
+08:47.480 --> 08:54.040
+ But if you maybe look at when you think of Einstein
+
+08:54.040 --> 08:58.960
+ and special relativity, what is the role of imagination
+
+08:58.960 --> 09:04.480
+ coming first there in the moment of discovery of an idea?
+
+09:04.480 --> 09:06.440
+ So there is obviously a mix of math
+
+09:06.440 --> 09:10.800
+ and out of the box imagination there.
+
+09:10.800 --> 09:12.600
+ That I don't know.
+
+09:12.600 --> 09:18.080
+ Whatever I did, I exclude any imagination.
+
+09:18.080 --> 09:21.080
+ Because whatever I saw in machine learning that
+
+09:21.080 --> 09:26.440
+ come from imagination, like features, like deep learning,
+
+09:26.440 --> 09:29.320
+ they are not relevant to the problem.
+
+09:29.320 --> 09:31.960
+ When you're looking very carefully
+
+09:31.960 --> 09:34.280
+ for mathematical equations, you're
+
+09:34.280 --> 09:38.000
+ deriving very simple theory, which goes far by
+
+09:38.000 --> 09:42.040
+ no theory at school than whatever people can imagine.
+
+09:42.040 --> 09:44.760
+ Because it is not good fantasy.
+
+09:44.760 --> 09:46.720
+ It is just interpretation.
+
+09:46.720 --> 09:48.000
+ It is just fantasy.
+
+09:48.000 --> 09:51.320
+ But it is not what you need.
+
+09:51.320 --> 09:56.960
+ You don't need any imagination to derive, say,
+
+09:56.960 --> 10:00.040
+ main principle of machine learning.
+
+10:00.040 --> 10:02.760
+ When you think about learning and intelligence,
+
+10:02.760 --> 10:04.560
+ maybe thinking about the human brain
+
+10:04.560 --> 10:09.200
+ and trying to describe mathematically the process of learning
+
+10:09.200 --> 10:13.160
+ that is something like what happens in the human brain,
+
+10:13.160 --> 10:17.200
+ do you think we have the tools currently?
+
+10:17.200 --> 10:19.000
+ Do you think we will ever have the tools
+
+10:19.000 --> 10:22.680
+ to try to describe that process of learning?
+
+10:22.680 --> 10:25.800
+ It is not description of what's going on.
+
+10:25.800 --> 10:27.360
+ It is interpretation.
+
+10:27.360 --> 10:29.400
+ It is your interpretation.
+
+10:29.400 --> 10:32.080
+ Your vision can be wrong.
+
+10:32.080 --> 10:36.160
+ You know, when a guy invent microscope,
+
+10:36.160 --> 10:40.560
+ Levin Cook for the first time, only he got this instrument
+
+10:40.560 --> 10:45.440
+ and nobody, he kept secrets about microscope.
+
+10:45.440 --> 10:49.080
+ But he wrote reports in London Academy of Science.
+
+10:49.080 --> 10:52.040
+ In his report, when he looked into the blood,
+
+10:52.040 --> 10:54.480
+ he looked everywhere, on the water, on the blood,
+
+10:54.480 --> 10:56.320
+ on the spin.
+
+10:56.320 --> 11:04.040
+ But he described blood like fight between queen and king.
+
+11:04.040 --> 11:08.120
+ So he saw blood cells, red cells,
+
+11:08.120 --> 11:12.400
+ and he imagines that it is army fighting each other.
+
+11:12.400 --> 11:16.960
+ And it was his interpretation of situation.
+
+11:16.960 --> 11:19.760
+ And he sent this report in Academy of Science.
+
+11:19.760 --> 11:22.640
+ They very carefully looked because they believed
+
+11:22.640 --> 11:25.160
+ that he is right, he saw something.
+
+11:25.160 --> 11:28.240
+ But he gave wrong interpretation.
+
+11:28.240 --> 11:32.280
+ And I believe the same can happen with brain.
+
+11:32.280 --> 11:35.280
+ Because the most important part, you know,
+
+11:35.280 --> 11:38.840
+ I believe in human language.
+
+11:38.840 --> 11:43.000
+ In some proverb, it's so much wisdom.
+
+11:43.000 --> 11:50.240
+ For example, people say that it is better than 1,000 days
+
+11:50.240 --> 11:53.960
+ of diligent studies one day with great teacher.
+
+11:53.960 --> 11:59.480
+ But if I will ask you what teacher does, nobody knows.
+
+11:59.480 --> 12:01.400
+ And that is intelligence.
+
+12:01.400 --> 12:07.320
+ And what we know from history, and now from mass
+
+12:07.320 --> 12:12.080
+ and machine learning, that teacher can do a lot.
+
+12:12.080 --> 12:14.400
+ So what, from a mathematical point of view,
+
+12:14.400 --> 12:16.080
+ is the great teacher?
+
+12:16.080 --> 12:17.240
+ I don't know.
+
+12:17.240 --> 12:18.880
+ That's an awful question.
+
+12:18.880 --> 12:25.120
+ Now, what we can say what teacher can do,
+
+12:25.120 --> 12:29.440
+ he can introduce some invariance, some predicate
+
+12:29.440 --> 12:32.280
+ for creating invariance.
+
+12:32.280 --> 12:33.520
+ How he doing it?
+
+12:33.520 --> 12:34.080
+ I don't know.
+
+12:34.080 --> 12:37.560
+ Because teacher knows reality and can describe
+
+12:37.560 --> 12:41.200
+ from this reality a predicate invariance.
+
+12:41.200 --> 12:43.480
+ But he knows that when you're using invariant,
+
+12:43.480 --> 12:47.960
+ he can decrease number of observations 100 times.
+
+12:47.960 --> 12:52.960
+ But maybe try to pull that apart a little bit.
+
+12:52.960 --> 12:58.120
+ I think you mentioned a piano teacher saying to the student,
+
+12:58.120 --> 12:59.880
+ play like a butterfly.
+
+12:59.880 --> 13:03.720
+ I played piano, I played guitar for a long time.
+
+13:03.720 --> 13:09.800
+ Yeah, maybe it's romantic, poetic.
+
+13:09.800 --> 13:13.160
+ But it feels like there's a lot of truth in that statement.
+
+13:13.160 --> 13:15.440
+ There is a lot of instruction in that statement.
+
+13:15.440 --> 13:17.320
+ And so can you pull that apart?
+
+13:17.320 --> 13:19.760
+ What is that?
+
+13:19.760 --> 13:22.520
+ The language itself may not contain this information.
+
+13:22.520 --> 13:24.160
+ It's not blah, blah, blah.
+
+13:24.160 --> 13:25.640
+ It does not blah, blah, blah, yeah.
+
+13:25.640 --> 13:26.960
+ It affects you.
+
+13:26.960 --> 13:27.600
+ It's what?
+
+13:27.600 --> 13:28.600
+ It affects you.
+
+13:28.600 --> 13:29.800
+ It affects your playing.
+
+13:29.800 --> 13:30.640
+ Yes, it does.
+
+13:30.640 --> 13:33.640
+ But it's not the language.
+
+13:33.640 --> 13:38.000
+ It feels like what is the information being exchanged there?
+
+13:38.000 --> 13:39.760
+ What is the nature of information?
+
+13:39.760 --> 13:41.880
+ What is the representation of that information?
+
+13:41.880 --> 13:44.000
+ I believe that it is sort of predicate.
+
+13:44.000 --> 13:45.400
+ But I don't know.
+
+13:45.400 --> 13:48.880
+ That is exactly what intelligence in machine learning
+
+13:48.880 --> 13:50.080
+ should be.
+
+13:50.080 --> 13:53.200
+ Because the rest is just mathematical technique.
+
+13:53.200 --> 13:57.920
+ I think that what was discovered recently
+
+13:57.920 --> 14:03.280
+ is that there is two mechanisms of learning.
+
+14:03.280 --> 14:06.040
+ One called strong convergence mechanism
+
+14:06.040 --> 14:08.560
+ and weak convergence mechanism.
+
+14:08.560 --> 14:11.200
+ Before, people use only one convergence.
+
+14:11.200 --> 14:15.840
+ In weak convergence mechanism, you can use predicate.
+
+14:15.840 --> 14:19.360
+ That's what play like butterfly.
+
+14:19.360 --> 14:23.640
+ And it will immediately affect your playing.
+
+14:23.640 --> 14:26.360
+ You know, there is English proverb.
+
+14:26.360 --> 14:27.320
+ Great.
+
+14:27.320 --> 14:31.680
+ If it looks like a duck, swims like a duck,
+
+14:31.680 --> 14:35.200
+ and quack like a duck, then it is probably duck.
+
+14:35.200 --> 14:36.240
+ Yes.
+
+14:36.240 --> 14:40.400
+ But this is exact about predicate.
+
+14:40.400 --> 14:42.920
+ Looks like a duck, what it means.
+
+14:42.920 --> 14:46.720
+ So you saw many ducks that you're training data.
+
+14:46.720 --> 14:56.480
+ So you have description of how looks integral looks ducks.
+
+14:56.480 --> 14:59.360
+ Yeah, the visual characteristics of a duck.
+
+14:59.360 --> 15:00.840
+ Yeah, but you won't.
+
+15:00.840 --> 15:04.200
+ And you have model for the cognition ducks.
+
+15:04.200 --> 15:07.880
+ So you would like that theoretical description
+
+15:07.880 --> 15:12.720
+ from model coincide with empirical description, which
+
+15:12.720 --> 15:14.520
+ you saw on Territax there.
+
+15:14.520 --> 15:18.440
+ So about looks like a duck, it is general.
+
+15:18.440 --> 15:21.480
+ But what about swims like a duck?
+
+15:21.480 --> 15:23.560
+ You should know that duck swims.
+
+15:23.560 --> 15:26.960
+ You can say it play chess like a duck, OK?
+
+15:26.960 --> 15:28.880
+ Duck doesn't play chess.
+
+15:28.880 --> 15:35.560
+ And it is completely legal predicate, but it is useless.
+
+15:35.560 --> 15:41.040
+ So half teacher can recognize not useless predicate.
+
+15:41.040 --> 15:44.640
+ So up to now, we don't use this predicate
+
+15:44.640 --> 15:46.680
+ in existing machine learning.
+
+15:46.680 --> 15:47.200
+ And you think that's not so useful?
+
+15:47.200 --> 15:50.600
+ So why we need billions of data?
+
+15:50.600 --> 15:55.560
+ But in this English proverb, they use only three predicate.
+
+15:55.560 --> 15:59.080
+ Looks like a duck, swims like a duck, and quack like a duck.
+
+15:59.080 --> 16:02.040
+ So you can't deny the fact that swims like a duck
+
+16:02.040 --> 16:08.520
+ and quacks like a duck has humor in it, has ambiguity.
+
+16:08.520 --> 16:12.600
+ Let's talk about swim like a duck.
+
+16:12.600 --> 16:16.520
+ It does not say jumps like a duck.
+
+16:16.520 --> 16:17.680
+ Why?
+
+16:17.680 --> 16:20.760
+ Because it's not relevant.
+
+16:20.760 --> 16:25.880
+ But that means that you know ducks, you know different birds,
+
+16:25.880 --> 16:27.600
+ you know animals.
+
+16:27.600 --> 16:32.440
+ And you derive from this that it is relevant to say swim like a duck.
+
+16:32.440 --> 16:36.680
+ So underneath, in order for us to understand swims like a duck,
+
+16:36.680 --> 16:41.200
+ it feels like we need to know millions of other little pieces
+
+16:41.200 --> 16:43.000
+ of information.
+
+16:43.000 --> 16:44.280
+ We pick up along the way.
+
+16:44.280 --> 16:45.120
+ You don't think so.
+
+16:45.120 --> 16:48.480
+ There doesn't need to be this knowledge base.
+
+16:48.480 --> 16:52.600
+ In those statements, carries some rich information
+
+16:52.600 --> 16:57.280
+ that helps us understand the essence of duck.
+
+16:57.280 --> 17:01.920
+ How far are we from integrating predicates?
+
+17:01.920 --> 17:06.000
+ You know that when you consider complete theory,
+
+17:06.000 --> 17:09.320
+ machine learning, so what it does,
+
+17:09.320 --> 17:12.400
+ you have a lot of functions.
+
+17:12.400 --> 17:17.480
+ And then you're talking, it looks like a duck.
+
+17:17.480 --> 17:20.720
+ You see your training data.
+
+17:20.720 --> 17:31.040
+ From training data, you recognize like expected duck should look.
+
+17:31.040 --> 17:37.640
+ Then you remove all functions, which does not look like you think
+
+17:37.640 --> 17:40.080
+ it should look from training data.
+
+17:40.080 --> 17:45.800
+ So you decrease amount of function from which you pick up one.
+
+17:45.800 --> 17:48.320
+ Then you give a second predicate.
+
+17:48.320 --> 17:51.840
+ And then, again, decrease the set of function.
+
+17:51.840 --> 17:55.800
+ And after that, you pick up the best function you can find.
+
+17:55.800 --> 17:58.120
+ It is standard machine learning.
+
+17:58.120 --> 18:03.280
+ So why you need not too many examples?
+
+18:03.280 --> 18:06.600
+ Because your predicates aren't very good, or you're not.
+
+18:06.600 --> 18:09.200
+ That means that predicate very good.
+
+18:09.200 --> 18:12.520
+ Because every predicate is invented
+
+18:12.520 --> 18:17.720
+ to decrease a divisible set of functions.
+
+18:17.720 --> 18:20.320
+ So you talk about admissible set of functions,
+
+18:20.320 --> 18:22.440
+ and you talk about good functions.
+
+18:22.440 --> 18:24.280
+ So what makes a good function?
+
+18:24.280 --> 18:28.600
+ So admissible set of function is set of function
+
+18:28.600 --> 18:32.760
+ which has small capacity, or small diversity,
+
+18:32.760 --> 18:36.960
+ small VC dimension example, which contain good function.
+
+18:36.960 --> 18:38.760
+ So by the way, for people who don't know,
+
+18:38.760 --> 18:42.440
+ VC, you're the V in the VC.
+
+18:42.440 --> 18:50.440
+ So how would you describe to a lay person what VC theory is?
+
+18:50.440 --> 18:51.440
+ How would you describe VC?
+
+18:51.440 --> 18:56.480
+ So when you have a machine, so a machine
+
+18:56.480 --> 19:00.240
+ capable to pick up one function from the admissible set
+
+19:00.240 --> 19:02.520
+ of function.
+
+19:02.520 --> 19:07.640
+ But set of admissibles function can be big.
+
+19:07.640 --> 19:11.600
+ They contain all continuous functions and it's useless.
+
+19:11.600 --> 19:15.280
+ You don't have so many examples to pick up function.
+
+19:15.280 --> 19:17.280
+ But it can be small.
+
+19:17.280 --> 19:24.560
+ Small, we call it capacity, but maybe better called diversity.
+
+19:24.560 --> 19:27.160
+ So not very different function in the set
+
+19:27.160 --> 19:31.280
+ is infinite set of function, but not very diverse.
+
+19:31.280 --> 19:34.280
+ So it is small VC dimension.
+
+19:34.280 --> 19:39.360
+ When VC dimension is small, you need small amount
+
+19:39.360 --> 19:41.760
+ of training data.
+
+19:41.760 --> 19:47.360
+ So the goal is to create admissible set of functions
+
+19:47.360 --> 19:53.200
+ which have small VC dimension and contain good function.
+
+19:53.200 --> 19:58.160
+ Then you will be able to pick up the function
+
+19:58.160 --> 20:02.400
+ using small amount of observations.
+
+20:02.400 --> 20:06.760
+ So that is the task of learning.
+
+20:06.760 --> 20:11.360
+ It is creating a set of admissible functions
+
+20:11.360 --> 20:13.120
+ that has a small VC dimension.
+
+20:13.120 --> 20:17.320
+ And then you've figured out a clever way of picking up.
+
+20:17.320 --> 20:22.440
+ No, that is goal of learning, which I formulated yesterday.
+
+20:22.440 --> 20:25.760
+ Statistical learning theory does not
+
+20:25.760 --> 20:30.360
+ involve in creating admissible set of function.
+
+20:30.360 --> 20:35.520
+ In classical learning theory, everywhere, 100% in textbook,
+
+20:35.520 --> 20:39.200
+ the set of function admissible set of function is given.
+
+20:39.200 --> 20:41.760
+ But this is science about nothing,
+
+20:41.760 --> 20:44.040
+ because the most difficult problem
+
+20:44.040 --> 20:50.120
+ to create admissible set of functions, given, say,
+
+20:50.120 --> 20:53.080
+ a lot of functions, continuum set of functions,
+
+20:53.080 --> 20:54.960
+ create admissible set of functions,
+
+20:54.960 --> 20:58.760
+ that means that it has finite VC dimension,
+
+20:58.760 --> 21:02.280
+ small VC dimension, and contain good function.
+
+21:02.280 --> 21:05.280
+ So this was out of consideration.
+
+21:05.280 --> 21:07.240
+ So what's the process of doing that?
+
+21:07.240 --> 21:08.240
+ I mean, it's fascinating.
+
+21:08.240 --> 21:13.200
+ What is the process of creating this admissible set of functions?
+
+21:13.200 --> 21:14.920
+ That is invariant.
+
+21:14.920 --> 21:15.760
+ That's invariance.
+
+21:15.760 --> 21:17.280
+ Can you describe invariance?
+
+21:17.280 --> 21:22.440
+ Yeah, you're looking of properties of training data.
+
+21:22.440 --> 21:30.120
+ And properties means that you have some function,
+
+21:30.120 --> 21:36.520
+ and you just count what is the average value of function
+
+21:36.520 --> 21:38.960
+ on training data.
+
+21:38.960 --> 21:43.040
+ You have a model, and what is the expectation
+
+21:43.040 --> 21:44.960
+ of this function on the model.
+
+21:44.960 --> 21:46.720
+ And they should coincide.
+
+21:46.720 --> 21:51.800
+ So the problem is about how to pick up functions.
+
+21:51.800 --> 21:53.200
+ It can be any function.
+
+21:53.200 --> 21:59.280
+ In fact, it is true for all functions.
+
+21:59.280 --> 22:05.000
+ But because when I talking set, say,
+
+22:05.000 --> 22:09.920
+ duck does not jumping, so you don't ask question, jump like a duck.
+
+22:09.920 --> 22:13.360
+ Because it is trivial, it does not jumping,
+
+22:13.360 --> 22:15.560
+ it doesn't help you to recognize jump.
+
+22:15.560 --> 22:19.000
+ But you know something, which question to ask,
+
+22:19.000 --> 22:23.840
+ when you're asking, it swims like a jump, like a duck.
+
+22:23.840 --> 22:26.840
+ But looks like a duck, it is general situation.
+
+22:26.840 --> 22:34.440
+ Looks like, say, guy who have this illness, this disease,
+
+22:34.440 --> 22:42.280
+ it is legal, so there is a general type of predicate
+
+22:42.280 --> 22:46.440
+ looks like, and special type of predicate,
+
+22:46.440 --> 22:50.040
+ which related to this specific problem.
+
+22:50.040 --> 22:53.440
+ And that is intelligence part of all this business.
+
+22:53.440 --> 22:55.440
+ And that we are teachers in world.
+
+22:55.440 --> 22:58.440
+ Incorporating those specialized predicates.
+
+22:58.440 --> 23:04.840
+ What do you think about deep learning as neural networks,
+
+23:04.840 --> 23:11.440
+ these arbitrary architectures as helping accomplish some of the tasks
+
+23:11.440 --> 23:14.440
+ you're thinking about, their effectiveness or lack thereof,
+
+23:14.440 --> 23:19.440
+ what are the weaknesses and what are the possible strengths?
+
+23:19.440 --> 23:22.440
+ You know, I think that this is fantasy.
+
+23:22.440 --> 23:28.440
+ Everything which like deep learning, like features.
+
+23:28.440 --> 23:32.440
+ Let me give you this example.
+
+23:32.440 --> 23:38.440
+ One of the greatest book, this Churchill book about history of Second World War.
+
+23:38.440 --> 23:47.440
+ And he's starting this book describing that in all time, when war is over,
+
+23:47.440 --> 23:54.440
+ so the great kings, they gathered together,
+
+23:54.440 --> 23:57.440
+ almost all of them were relatives,
+
+23:57.440 --> 24:02.440
+ and they discussed what should be done, how to create peace.
+
+24:02.440 --> 24:04.440
+ And they came to agreement.
+
+24:04.440 --> 24:13.440
+ And when happens First World War, the general public came in power.
+
+24:13.440 --> 24:17.440
+ And they were so greedy that robbed Germany.
+
+24:17.440 --> 24:21.440
+ And it was clear for everybody that it is not peace.
+
+24:21.440 --> 24:28.440
+ That peace will last only 20 years, because they were not professionals.
+
+24:28.440 --> 24:31.440
+ It's the same I see in machine learning.
+
+24:31.440 --> 24:37.440
+ There are mathematicians who are looking for the problem from a very deep point of view,
+
+24:37.440 --> 24:39.440
+ a mathematical point of view.
+
+24:39.440 --> 24:45.440
+ And there are computer scientists who mostly does not know mathematics.
+
+24:45.440 --> 24:48.440
+ They just have interpretation of that.
+
+24:48.440 --> 24:53.440
+ And they invented a lot of blah, blah, blah interpretations like deep learning.
+
+24:53.440 --> 24:55.440
+ Why you need deep learning?
+
+24:55.440 --> 24:57.440
+ Mathematics does not know deep learning.
+
+24:57.440 --> 25:00.440
+ Mathematics does not know neurons.
+
+25:00.440 --> 25:02.440
+ It is just function.
+
+25:02.440 --> 25:06.440
+ If you like to say piecewise linear function, say that,
+
+25:06.440 --> 25:10.440
+ and do it in class of piecewise linear function.
+
+25:10.440 --> 25:12.440
+ But they invent something.
+
+25:12.440 --> 25:20.440
+ And then they try to prove the advantage of that through interpretations,
+
+25:20.440 --> 25:22.440
+ which mostly wrong.
+
+25:22.440 --> 25:25.440
+ And when not enough they appeal to brain,
+
+25:25.440 --> 25:27.440
+ which they know nothing about that.
+
+25:27.440 --> 25:29.440
+ Nobody knows what's going on in the brain.
+
+25:29.440 --> 25:34.440
+ So I think that more reliable look on maths.
+
+25:34.440 --> 25:36.440
+ This is a mathematical problem.
+
+25:36.440 --> 25:38.440
+ Do your best to solve this problem.
+
+25:38.440 --> 25:43.440
+ Try to understand that there is not only one way of convergence,
+
+25:43.440 --> 25:45.440
+ which is strong way of convergence.
+
+25:45.440 --> 25:49.440
+ There is a weak way of convergence, which requires predicate.
+
+25:49.440 --> 25:52.440
+ And if you will go through all this stuff,
+
+25:52.440 --> 25:55.440
+ you will see that you don't need deep learning.
+
+25:55.440 --> 26:00.440
+ Even more, I would say one of the theorem,
+
+26:00.440 --> 26:02.440
+ which is called representor theorem.
+
+26:02.440 --> 26:10.440
+ It says that optimal solution of mathematical problem,
+
+26:10.440 --> 26:20.440
+ which described learning, is on shadow network, not on deep learning.
+
+26:20.440 --> 26:22.440
+ And a shallow network, yeah.
+
+26:22.440 --> 26:24.440
+ The ultimate problem is there.
+
+26:24.440 --> 26:25.440
+ Absolutely.
+
+26:25.440 --> 26:29.440
+ So in the end, what you're saying is exactly right.
+
+26:29.440 --> 26:35.440
+ The question is, you have no value for throwing something on the table,
+
+26:35.440 --> 26:38.440
+ playing with it, not math.
+
+26:38.440 --> 26:41.440
+ It's like in your old network where you said throwing something in the bucket
+
+26:41.440 --> 26:45.440
+ or the biological example and looking at kings and queens
+
+26:45.440 --> 26:47.440
+ or the cells or the microscope.
+
+26:47.440 --> 26:52.440
+ You don't see value in imagining the cells or kings and queens
+
+26:52.440 --> 26:56.440
+ and using that as inspiration and imagination
+
+26:56.440 --> 26:59.440
+ for where the math will eventually lead you.
+
+26:59.440 --> 27:06.440
+ You think that interpretation basically deceives you in a way that's not productive.
+
+27:06.440 --> 27:14.440
+ I think that if you're trying to analyze this business of learning
+
+27:14.440 --> 27:18.440
+ and especially discussion about deep learning,
+
+27:18.440 --> 27:21.440
+ it is discussion about interpretation.
+
+27:21.440 --> 27:26.440
+ It's discussion about things, about what you can say about things.
+
+27:26.440 --> 27:29.440
+ That's right, but aren't you surprised by the beauty of it?
+
+27:29.440 --> 27:36.440
+ Not mathematical beauty, but the fact that it works at all.
+
+27:36.440 --> 27:39.440
+ Or are you criticizing that very beauty,
+
+27:39.440 --> 27:45.440
+ our human desire to interpret,
+
+27:45.440 --> 27:49.440
+ to find our silly interpretations in these constructs?
+
+27:49.440 --> 27:51.440
+ Let me ask you this.
+
+27:51.440 --> 27:55.440
+ Are you surprised?
+
+27:55.440 --> 27:57.440
+ Does it inspire you?
+
+27:57.440 --> 28:00.440
+ How do you feel about the success of a system like AlphaGo
+
+28:00.440 --> 28:03.440
+ at beating the game of Go?
+
+28:03.440 --> 28:09.440
+ Using neural networks to estimate the quality of a board
+
+28:09.440 --> 28:11.440
+ and the quality of the board?
+
+28:11.440 --> 28:14.440
+ That is your interpretation quality of the board.
+
+28:14.440 --> 28:17.440
+ Yes.
+
+28:17.440 --> 28:20.440
+ It's not our interpretation.
+
+28:20.440 --> 28:23.440
+ The fact is, a neural network system doesn't matter.
+
+28:23.440 --> 28:27.440
+ A learning system that we don't mathematically understand
+
+28:27.440 --> 28:29.440
+ that beats the best human player.
+
+28:29.440 --> 28:31.440
+ It does something that was thought impossible.
+
+28:31.440 --> 28:35.440
+ That means that it's not very difficult problem.
+
+28:35.440 --> 28:41.440
+ We've empirically discovered that this is not a very difficult problem.
+
+28:41.440 --> 28:43.440
+ That's true.
+
+28:43.440 --> 28:49.440
+ Maybe I can't argue.
+
+28:49.440 --> 28:52.440
+ Even more, I would say,
+
+28:52.440 --> 28:54.440
+ that if they use deep learning,
+
+28:54.440 --> 28:59.440
+ it is not the most effective way of learning theory.
+
+28:59.440 --> 29:03.440
+ Usually, when people use deep learning,
+
+29:03.440 --> 29:09.440
+ they're using zillions of training data.
+
+29:09.440 --> 29:13.440
+ But you don't need this.
+
+29:13.440 --> 29:15.440
+ I describe the challenge.
+
+29:15.440 --> 29:22.440
+ Can we do some problems with deep learning method
+
+29:22.440 --> 29:27.440
+ with deep net using 100 times less training data?
+
+29:27.440 --> 29:33.440
+ Even more, some problems deep learning cannot solve
+
+29:33.440 --> 29:37.440
+ because it's not necessary.
+
+29:37.440 --> 29:40.440
+ They create admissible set of functions.
+
+29:40.440 --> 29:45.440
+ Deep architecture means to create admissible set of functions.
+
+29:45.440 --> 29:49.440
+ You cannot say that you're creating good admissible set of functions.
+
+29:49.440 --> 29:52.440
+ It's your fantasy.
+
+29:52.440 --> 29:54.440
+ It does not come from mass.
+
+29:54.440 --> 29:58.440
+ But it is possible to create admissible set of functions
+
+29:58.440 --> 30:01.440
+ because you have your training data.
+
+30:01.440 --> 30:08.440
+ Actually, for mathematicians, when you consider a variant,
+
+30:08.440 --> 30:11.440
+ you need to use law of large numbers.
+
+30:11.440 --> 30:17.440
+ When you're making training in existing algorithm,
+
+30:17.440 --> 30:20.440
+ you need uniform law of large numbers,
+
+30:20.440 --> 30:22.440
+ which is much more difficult.
+
+30:22.440 --> 30:24.440
+ You see dimension and all this stuff.
+
+30:24.440 --> 30:32.440
+ Nevertheless, if you use both weak and strong way of convergence,
+
+30:32.440 --> 30:34.440
+ you can decrease a lot of training data.
+
+30:34.440 --> 30:39.440
+ You could do the three, the Swims like a duck and Quacks like a duck.
+
+30:39.440 --> 30:47.440
+ Let's step back and think about human intelligence in general.
+
+30:47.440 --> 30:52.440
+ Clearly, that has evolved in a nonmathematical way.
+
+30:52.440 --> 31:00.440
+ As far as we know, God, or whoever,
+
+31:00.440 --> 31:05.440
+ didn't come up with a model in place in our brain of admissible functions.
+
+31:05.440 --> 31:06.440
+ It kind of evolved.
+
+31:06.440 --> 31:07.440
+ I don't know.
+
+31:07.440 --> 31:08.440
+ Maybe you have a view on this.
+
+31:08.440 --> 31:15.440
+ Alan Turing in the 50s in his paper asked and rejected the question,
+
+31:15.440 --> 31:16.440
+ can machines think?
+
+31:16.440 --> 31:18.440
+ It's not a very useful question.
+
+31:18.440 --> 31:23.440
+ But can you briefly entertain this useless question?
+
+31:23.440 --> 31:25.440
+ Can machines think?
+
+31:25.440 --> 31:28.440
+ So talk about intelligence and your view of it.
+
+31:28.440 --> 31:29.440
+ I don't know that.
+
+31:29.440 --> 31:34.440
+ I know that Turing described imitation.
+
+31:34.440 --> 31:41.440
+ If computer can imitate human being, let's call it intelligent.
+
+31:41.440 --> 31:45.440
+ And he understands that it is not thinking computer.
+
+31:45.440 --> 31:46.440
+ Yes.
+
+31:46.440 --> 31:49.440
+ He completely understands what he's doing.
+
+31:49.440 --> 31:53.440
+ But he's set up a problem of imitation.
+
+31:53.440 --> 31:57.440
+ So now we understand that the problem is not in imitation.
+
+31:57.440 --> 32:04.440
+ I'm not sure that intelligence is just inside of us.
+
+32:04.440 --> 32:06.440
+ It may be also outside of us.
+
+32:06.440 --> 32:09.440
+ I have several observations.
+
+32:09.440 --> 32:15.440
+ So when I prove some theorem, it's a very difficult theorem.
+
+32:15.440 --> 32:22.440
+ But in a couple of years, in several places, people proved the same theorem.
+
+32:22.440 --> 32:26.440
+ Say, soil lemma after us was done.
+
+32:26.440 --> 32:29.440
+ Then another guy proved the same theorem.
+
+32:29.440 --> 32:32.440
+ In the history of science, it's happened all the time.
+
+32:32.440 --> 32:35.440
+ For example, geometry.
+
+32:35.440 --> 32:37.440
+ It's happened simultaneously.
+
+32:37.440 --> 32:43.440
+ First it did Lobachevsky and then Gauss and Boyai and other guys.
+
+32:43.440 --> 32:48.440
+ It happened simultaneously in 10 years period of time.
+
+32:48.440 --> 32:51.440
+ And I saw a lot of examples like that.
+
+32:51.440 --> 32:56.440
+ And many mathematicians think that when they develop something,
+
+32:56.440 --> 33:01.440
+ they develop something in general which affects everybody.
+
+33:01.440 --> 33:07.440
+ So maybe our model that intelligence is only inside of us is incorrect.
+
+33:07.440 --> 33:09.440
+ It's our interpretation.
+
+33:09.440 --> 33:15.440
+ Maybe there exists some connection with world intelligence.
+
+33:15.440 --> 33:16.440
+ I don't know.
+
+33:16.440 --> 33:19.440
+ You're almost like plugging in into...
+
+33:19.440 --> 33:20.440
+ Yeah, exactly.
+
+33:20.440 --> 33:22.440
+...and contributing to this...
+
+33:22.440 --> 33:23.440
+ Into a big network.
+
+33:23.440 --> 33:26.440
+...into a big, maybe in your own network.
+
+33:26.440 --> 33:27.440
+ No, no, no.
+
+33:27.440 --> 33:34.440
+ On the flip side of that, maybe you can comment on big O complexity
+
+33:34.440 --> 33:40.440
+ and how you see classifying algorithms by worst case running time
+
+33:40.440 --> 33:42.440
+ in relation to their input.
+
+33:42.440 --> 33:45.440
+ So that way of thinking about functions.
+
+33:45.440 --> 33:47.440
+ Do you think P equals NP?
+
+33:47.440 --> 33:49.440
+ Do you think that's an interesting question?
+
+33:49.440 --> 33:51.440
+ Yeah, it is an interesting question.
+
+33:51.440 --> 34:01.440
+ But let me talk about complexity and about worst case scenario.
+
+34:01.440 --> 34:03.440
+ There is a mathematical setting.
+
+34:03.440 --> 34:07.440
+ When I came to the United States in 1990,
+
+34:07.440 --> 34:09.440
+ people did not know this theory.
+
+34:09.440 --> 34:12.440
+ They did not know statistical learning theory.
+
+34:12.440 --> 34:17.440
+ So in Russia it was published to monographs or monographs,
+
+34:17.440 --> 34:19.440
+ but in America they didn't know.
+
+34:19.440 --> 34:22.440
+ Then they learned.
+
+34:22.440 --> 34:25.440
+ And somebody told me that if it's worst case theory,
+
+34:25.440 --> 34:27.440
+ and they will create real case theory,
+
+34:27.440 --> 34:30.440
+ but till now it did not.
+
+34:30.440 --> 34:33.440
+ Because it is a mathematical tool.
+
+34:33.440 --> 34:38.440
+ You can do only what you can do using mathematics,
+
+34:38.440 --> 34:45.440
+ which has a clear understanding and clear description.
+
+34:45.440 --> 34:52.440
+ And for this reason we introduced complexity.
+
+34:52.440 --> 34:54.440
+ And we need this.
+
+34:54.440 --> 35:01.440
+ Because actually it is diverse, I like this one more.
+
+35:01.440 --> 35:04.440
+ This dimension you can prove some theorems.
+
+35:04.440 --> 35:12.440
+ But we also create theory for case when you know probability measure.
+
+35:12.440 --> 35:14.440
+ And that is the best case which can happen.
+
+35:14.440 --> 35:17.440
+ It is entropy theory.
+
+35:17.440 --> 35:20.440
+ So from a mathematical point of view,
+
+35:20.440 --> 35:25.440
+ you know the best possible case and the worst possible case.
+
+35:25.440 --> 35:28.440
+ You can derive different model in medium.
+
+35:28.440 --> 35:30.440
+ But it's not so interesting.
+
+35:30.440 --> 35:33.440
+ You think the edges are interesting?
+
+35:33.440 --> 35:35.440
+ The edges are interesting.
+
+35:35.440 --> 35:44.440
+ Because it is not so easy to get a good bound, exact bound.
+
+35:44.440 --> 35:47.440
+ It's not many cases where you have.
+
+35:47.440 --> 35:49.440
+ The bound is not exact.
+
+35:49.440 --> 35:54.440
+ But interesting principles which discover the mass.
+
+35:54.440 --> 35:57.440
+ Do you think it's interesting because it's challenging
+
+35:57.440 --> 36:02.440
+ and reveals interesting principles that allow you to get those bounds?
+
+36:02.440 --> 36:05.440
+ Or do you think it's interesting because it's actually very useful
+
+36:05.440 --> 36:10.440
+ for understanding the essence of a function of an algorithm?
+
+36:10.440 --> 36:15.440
+ So it's like me judging your life as a human being
+
+36:15.440 --> 36:19.440
+ by the worst thing you did and the best thing you did
+
+36:19.440 --> 36:21.440
+ versus all the stuff in the middle.
+
+36:21.440 --> 36:25.440
+ It seems not productive.
+
+36:25.440 --> 36:31.440
+ I don't think so because you cannot describe situation in the middle.
+
+36:31.440 --> 36:34.440
+ Or it will be not general.
+
+36:34.440 --> 36:38.440
+ So you can describe edges cases.
+
+36:38.440 --> 36:41.440
+ And it is clear it has some model.
+
+36:41.440 --> 36:47.440
+ But you cannot describe model for every new case.
+
+36:47.440 --> 36:53.440
+ So you will be never accurate when you're using model.
+
+36:53.440 --> 36:55.440
+ But from a statistical point of view,
+
+36:55.440 --> 37:00.440
+ the way you've studied functions and the nature of learning
+
+37:00.440 --> 37:07.440
+ and the world, don't you think that the real world has a very long tail
+
+37:07.440 --> 37:13.440
+ that the edge cases are very far away from the mean,
+
+37:13.440 --> 37:19.440
+ the stuff in the middle, or no?
+
+37:19.440 --> 37:21.440
+ I don't know that.
+
+37:21.440 --> 37:29.440
+ I think that from my point of view,
+
+37:29.440 --> 37:39.440
+ if you will use formal statistic, uniform law of large numbers,
+
+37:39.440 --> 37:47.440
+ if you will use this invariance business,
+
+37:47.440 --> 37:51.440
+ you will need just law of large numbers.
+
+37:51.440 --> 37:55.440
+ And there's a huge difference between uniform law of large numbers
+
+37:55.440 --> 37:57.440
+ and large numbers.
+
+37:57.440 --> 37:59.440
+ Can you describe that a little more?
+
+37:59.440 --> 38:01.440
+ Or should we just take it to...
+
+38:01.440 --> 38:05.440
+ No, for example, when I'm talking about duck,
+
+38:05.440 --> 38:09.440
+ I gave three predicates and it was enough.
+
+38:09.440 --> 38:14.440
+ But if you will try to do formal distinguish,
+
+38:14.440 --> 38:17.440
+ you will need a lot of observations.
+
+38:17.440 --> 38:19.440
+ I got you.
+
+38:19.440 --> 38:24.440
+ And so that means that information about looks like a duck
+
+38:24.440 --> 38:27.440
+ contain a lot of bits of information,
+
+38:27.440 --> 38:29.440
+ formal bits of information.
+
+38:29.440 --> 38:35.440
+ So we don't know that how much bit of information
+
+38:35.440 --> 38:39.440
+ contain things from artificial intelligence.
+
+38:39.440 --> 38:42.440
+ And that is the subject of analysis.
+
+38:42.440 --> 38:47.440
+ Till now, old business,
+
+38:47.440 --> 38:54.440
+ I don't like how people consider artificial intelligence.
+
+38:54.440 --> 39:00.440
+ They consider us some codes which imitate activity of human being.
+
+39:00.440 --> 39:02.440
+ It is not science.
+
+39:02.440 --> 39:04.440
+ It is applications.
+
+39:04.440 --> 39:06.440
+ You would like to imitate God.
+
+39:06.440 --> 39:09.440
+ It is very useful and we have good problem.
+
+39:09.440 --> 39:15.440
+ But you need to learn something more.
+
+39:15.440 --> 39:23.440
+ How people can to develop predicates,
+
+39:23.440 --> 39:25.440
+ swims like a duck,
+
+39:25.440 --> 39:28.440
+ or play like butterfly or something like that.
+
+39:28.440 --> 39:33.440
+ Not the teacher tells you how it came in his mind.
+
+39:33.440 --> 39:36.440
+ How he choose this image.
+
+39:36.440 --> 39:39.440
+ That is problem of intelligence.
+
+39:39.440 --> 39:41.440
+ That is the problem of intelligence.
+
+39:41.440 --> 39:44.440
+ And you see that connected to the problem of learning?
+
+39:44.440 --> 39:45.440
+ Absolutely.
+
+39:45.440 --> 39:48.440
+ Because you immediately give this predicate
+
+39:48.440 --> 39:52.440
+ like specific predicate, swims like a duck,
+
+39:52.440 --> 39:54.440
+ or quack like a duck.
+
+39:54.440 --> 39:57.440
+ It was chosen somehow.
+
+39:57.440 --> 40:00.440
+ So what is the line of work, would you say?
+
+40:00.440 --> 40:05.440
+ If you were to formulate as a set of open problems,
+
+40:05.440 --> 40:07.440
+ that will take us there.
+
+40:07.440 --> 40:09.440
+ Play like a butterfly.
+
+40:09.440 --> 40:11.440
+ We will get a system to be able to...
+
+40:11.440 --> 40:13.440
+ Let's separate two stories.
+
+40:13.440 --> 40:15.440
+ One mathematical story.
+
+40:15.440 --> 40:19.440
+ That if you have predicate, you can do something.
+
+40:19.440 --> 40:22.440
+ And another story you have to get predicate.
+
+40:22.440 --> 40:26.440
+ It is intelligence problem.
+
+40:26.440 --> 40:31.440
+ And people even did not start understanding intelligence.
+
+40:31.440 --> 40:34.440
+ Because to understand intelligence, first of all,
+
+40:34.440 --> 40:37.440
+ try to understand what doing teachers.
+
+40:37.440 --> 40:40.440
+ How teacher teach.
+
+40:40.440 --> 40:43.440
+ Why one teacher better than another one?
+
+40:43.440 --> 40:44.440
+ Yeah.
+
+40:44.440 --> 40:48.440
+ So you think we really even haven't started on the journey
+
+40:48.440 --> 40:50.440
+ of generating the predicate?
+
+40:50.440 --> 40:51.440
+ No.
+
+40:51.440 --> 40:52.440
+ We don't understand.
+
+40:52.440 --> 40:56.440
+ We even don't understand that this problem exists.
+
+40:56.440 --> 40:58.440
+ Because did you hear?
+
+40:58.440 --> 40:59.440
+ You do.
+
+40:59.440 --> 41:02.440
+ No, I just know name.
+
+41:02.440 --> 41:07.440
+ I want to understand why one teacher better than another.
+
+41:07.440 --> 41:12.440
+ And how affect teacher student.
+
+41:12.440 --> 41:17.440
+ It is not because he repeating the problem which is in textbook.
+
+41:17.440 --> 41:18.440
+ Yes.
+
+41:18.440 --> 41:20.440
+ He make some remarks.
+
+41:20.440 --> 41:23.440
+ He make some philosophy of reasoning.
+
+41:23.440 --> 41:24.440
+ Yeah, that's a beautiful...
+
+41:24.440 --> 41:31.440
+ So it is a formulation of a question that is the open problem.
+
+41:31.440 --> 41:33.440
+ Why is one teacher better than another?
+
+41:33.440 --> 41:34.440
+ Right.
+
+41:34.440 --> 41:37.440
+ What he does better.
+
+41:37.440 --> 41:38.440
+ Yeah.
+
+41:38.440 --> 41:39.440
+ What...
+
+41:39.440 --> 41:42.440
+ Why at every level?
+
+41:42.440 --> 41:44.440
+ How do they get better?
+
+41:44.440 --> 41:47.440
+ What does it mean to be better?
+
+41:47.440 --> 41:49.440
+ The whole...
+
+41:49.440 --> 41:50.440
+ Yeah.
+
+41:50.440 --> 41:53.440
+ From whatever model I have.
+
+41:53.440 --> 41:56.440
+ One teacher can give a very good predicate.
+
+41:56.440 --> 42:00.440
+ One teacher can say swims like a dog.
+
+42:00.440 --> 42:03.440
+ And another can say jump like a dog.
+
+42:03.440 --> 42:05.440
+ And jump like a dog.
+
+42:05.440 --> 42:07.440
+ Car is zero information.
+
+42:07.440 --> 42:08.440
+ Yeah.
+
+42:08.440 --> 42:13.440
+ So what is the most exciting problem in statistical learning you've ever worked on?
+
+42:13.440 --> 42:16.440
+ Or are working on now?
+
+42:16.440 --> 42:22.440
+ I just finished this invariant story.
+
+42:22.440 --> 42:24.440
+ And I'm happy that...
+
+42:24.440 --> 42:30.440
+ I believe that it is ultimate learning story.
+
+42:30.440 --> 42:37.440
+ At least I can show that there are no another mechanism, only two mechanisms.
+
+42:37.440 --> 42:44.440
+ But they separate statistical part from intelligent part.
+
+42:44.440 --> 42:48.440
+ And I know nothing about intelligent part.
+
+42:48.440 --> 42:52.440
+ And if we will know this intelligent part,
+
+42:52.440 --> 42:59.440
+ so it will help us a lot in teaching, in learning.
+
+42:59.440 --> 43:02.440
+ You don't know it when we see it?
+
+43:02.440 --> 43:06.440
+ So for example, in my talk, the last slide was the challenge.
+
+43:06.440 --> 43:11.440
+ So you have, say, NIST digital recognition problem.
+
+43:11.440 --> 43:16.440
+ And deep learning claims that they did it very well.
+
+43:16.440 --> 43:21.440
+ Say 99.5% of correct answers.
+
+43:21.440 --> 43:24.440
+ But they use 60,000 observations.
+
+43:24.440 --> 43:26.440
+ Can you do the same?
+
+43:26.440 --> 43:29.440
+ 100 times less.
+
+43:29.440 --> 43:31.440
+ But incorporating invariants.
+
+43:31.440 --> 43:34.440
+ What it means, you know, digit one, two, three.
+
+43:34.440 --> 43:35.440
+ Yeah.
+
+43:35.440 --> 43:37.440
+ Just looking at that.
+
+43:37.440 --> 43:40.440
+ Explain to me which invariant I should keep.
+
+43:40.440 --> 43:43.440
+ To use 100 examples.
+
+43:43.440 --> 43:48.440
+ Or say 100 times less examples to do the same job.
+
+43:48.440 --> 43:49.440
+ Yeah.
+
+43:49.440 --> 43:55.440
+ That last slide, unfortunately, you talk ended quickly.
+
+43:55.440 --> 43:59.440
+ The last slide was a powerful open challenge
+
+43:59.440 --> 44:02.440
+ and a formulation of the essence here.
+
+44:02.440 --> 44:06.440
+ That is the exact problem of intelligence.
+
+44:06.440 --> 44:12.440
+ Because everybody, when machine learning started,
+
+44:12.440 --> 44:15.440
+ it was developed by mathematicians,
+
+44:15.440 --> 44:19.440
+ they immediately recognized that we use much more
+
+44:19.440 --> 44:22.440
+ training data than humans needed.
+
+44:22.440 --> 44:25.440
+ But now again, we came to the same story.
+
+44:25.440 --> 44:27.440
+ Have to decrease.
+
+44:27.440 --> 44:30.440
+ That is the problem of learning.
+
+44:30.440 --> 44:32.440
+ It is not like in deep learning,
+
+44:32.440 --> 44:35.440
+ they use zealons of training data.
+
+44:35.440 --> 44:38.440
+ Because maybe zealons are not enough
+
+44:38.440 --> 44:44.440
+ if you have a good invariance.
+
+44:44.440 --> 44:49.440
+ Maybe you'll never collect some number of observations.
+
+44:49.440 --> 44:53.440
+ But now it is a question to intelligence.
+
+44:53.440 --> 44:55.440
+ Have to do that.
+
+44:55.440 --> 44:58.440
+ Because statistical part is ready.
+
+44:58.440 --> 45:02.440
+ As soon as you supply us with predicate,
+
+45:02.440 --> 45:06.440
+ we can do good job with small amount of observations.
+
+45:06.440 --> 45:10.440
+ And the very first challenges will know digit recognition.
+
+45:10.440 --> 45:12.440
+ And you know digits.
+
+45:12.440 --> 45:15.440
+ And please tell me invariance.
+
+45:15.440 --> 45:16.440
+ I think about that.
+
+45:16.440 --> 45:20.440
+ I can say for digit 3, I would introduce
+
+45:20.440 --> 45:24.440
+ concept of horizontal symmetry.
+
+45:24.440 --> 45:29.440
+ So the digit 3 has horizontal symmetry
+
+45:29.440 --> 45:33.440
+ more than say digit 2 or something like that.
+
+45:33.440 --> 45:37.440
+ But as soon as I get the idea of horizontal symmetry,
+
+45:37.440 --> 45:40.440
+ I can mathematically invent a lot of
+
+45:40.440 --> 45:43.440
+ measure of horizontal symmetry
+
+45:43.440 --> 45:46.440
+ on vertical symmetry or diagonal symmetry,
+
+45:46.440 --> 45:49.440
+ whatever, if I have a day of symmetry.
+
+45:49.440 --> 45:52.440
+ But what else?
+
+45:52.440 --> 46:00.440
+ Looking on digit, I see that it is metapredicate,
+
+46:00.440 --> 46:04.440
+ which is not shape.
+
+46:04.440 --> 46:07.440
+ It is something like symmetry,
+
+46:07.440 --> 46:12.440
+ like how dark is whole picture, something like that.
+
+46:12.440 --> 46:15.440
+ Which can self rise up predicate.
+
+46:15.440 --> 46:18.440
+ You think such a predicate could rise
+
+46:18.440 --> 46:26.440
+ out of something that is not general.
+
+46:26.440 --> 46:31.440
+ Meaning it feels like for me to be able to
+
+46:31.440 --> 46:34.440
+ understand the difference between a 2 and a 3,
+
+46:34.440 --> 46:39.440
+ I would need to have had a childhood
+
+46:39.440 --> 46:45.440
+ of 10 to 15 years playing with kids,
+
+46:45.440 --> 46:50.440
+ going to school, being yelled by parents.
+
+46:50.440 --> 46:55.440
+ All of that, walking, jumping, looking at ducks.
+
+46:55.440 --> 46:58.440
+ And now then I would be able to generate
+
+46:58.440 --> 47:01.440
+ the right predicate for telling the difference
+
+47:01.440 --> 47:03.440
+ between 2 and a 3.
+
+47:03.440 --> 47:06.440
+ Or do you think there is a more efficient way?
+
+47:06.440 --> 47:10.440
+ I know for sure that you must know
+
+47:10.440 --> 47:12.440
+ something more than digits.
+
+47:12.440 --> 47:15.440
+ That's a powerful statement.
+
+47:15.440 --> 47:19.440
+ But maybe there are several languages
+
+47:19.440 --> 47:24.440
+ of description, these elements of digits.
+
+47:24.440 --> 47:27.440
+ So I'm talking about symmetry,
+
+47:27.440 --> 47:30.440
+ about some properties of geometry,
+
+47:30.440 --> 47:33.440
+ I'm talking about something abstract.
+
+47:33.440 --> 47:38.440
+ But this is a problem of intelligence.
+
+47:38.440 --> 47:42.440
+ So in one of our articles, it is trivial to show
+
+47:42.440 --> 47:46.440
+ that every example can carry
+
+47:46.440 --> 47:49.440
+ not more than one bit of information in real.
+
+47:49.440 --> 47:54.440
+ Because when you show example
+
+47:54.440 --> 47:59.440
+ and you say this is one, you can remove, say,
+
+47:59.440 --> 48:03.440
+ a function which does not tell you one, say,
+
+48:03.440 --> 48:06.440
+ the best strategy, if you can do it perfectly,
+
+48:06.440 --> 48:09.440
+ it's remove half of the functions.
+
+48:09.440 --> 48:14.440
+ But when you use one predicate, which looks like a duck,
+
+48:14.440 --> 48:18.440
+ you can remove much more functions than half.
+
+48:18.440 --> 48:20.440
+ And that means that it contains
+
+48:20.440 --> 48:25.440
+ a lot of bit of information from a formal point of view.
+
+48:25.440 --> 48:31.440
+ But when you have a general picture
+
+48:31.440 --> 48:33.440
+ of what you want to recognize,
+
+48:33.440 --> 48:36.440
+ a general picture of the world,
+
+48:36.440 --> 48:40.440
+ can you invent this predicate?
+
+48:40.440 --> 48:46.440
+ And that predicate carries a lot of information.
+
+48:46.440 --> 48:49.440
+ Beautifully put, maybe just me,
+
+48:49.440 --> 48:53.440
+ but in all the math you show, in your work,
+
+48:53.440 --> 48:57.440
+ which is some of the most profound mathematical work
+
+48:57.440 --> 49:01.440
+ in the field of learning AI and just math in general.
+
+49:01.440 --> 49:04.440
+ I hear a lot of poetry and philosophy.
+
+49:04.440 --> 49:09.440
+ You really kind of talk about philosophy of science.
+
+49:09.440 --> 49:12.440
+ There's a poetry and music to a lot of the work you're doing
+
+49:12.440 --> 49:14.440
+ and the way you're thinking about it.
+
+49:14.440 --> 49:16.440
+ So where does that come from?
+
+49:16.440 --> 49:20.440
+ Do you escape to poetry? Do you escape to music?
+
+49:20.440 --> 49:24.440
+ I think that there exists ground truth.
+
+49:24.440 --> 49:26.440
+ There exists ground truth?
+
+49:26.440 --> 49:30.440
+ Yeah, and that can be seen everywhere.
+
+49:30.440 --> 49:32.440
+ The smart guy, philosopher,
+
+49:32.440 --> 49:38.440
+ sometimes I surprise how they deep see.
+
+49:38.440 --> 49:45.440
+ Sometimes I see that some of them are completely out of subject.
+
+49:45.440 --> 49:50.440
+ But the ground truth I see in music.
+
+49:50.440 --> 49:52.440
+ Music is the ground truth?
+
+49:52.440 --> 49:53.440
+ Yeah.
+
+49:53.440 --> 50:01.440
+ And in poetry, many poets, they believe they take dictation.
+
+50:01.440 --> 50:06.440
+ So what piece of music,
+
+50:06.440 --> 50:08.440
+ as a piece of empirical evidence,
+
+50:08.440 --> 50:14.440
+ gave you a sense that they are touching something in the ground truth?
+
+50:14.440 --> 50:16.440
+ It is structure.
+
+50:16.440 --> 50:18.440
+ The structure with the math of music.
+
+50:18.440 --> 50:20.440
+ Because when you're listening to Bach,
+
+50:20.440 --> 50:22.440
+ you see this structure.
+
+50:22.440 --> 50:25.440
+ Very clear, very classic, very simple.
+
+50:25.440 --> 50:31.440
+ And the same in Bach, when you have axioms in geometry,
+
+50:31.440 --> 50:33.440
+ you have the same feeling.
+
+50:33.440 --> 50:36.440
+ And in poetry, sometimes you see the same.
+
+50:36.440 --> 50:40.440
+ And if you look back at your childhood,
+
+50:40.440 --> 50:42.440
+ you grew up in Russia,
+
+50:42.440 --> 50:46.440
+ you maybe were born as a researcher in Russia,
+
+50:46.440 --> 50:48.440
+ you developed as a researcher in Russia,
+
+50:48.440 --> 50:51.440
+ you came to the United States in a few places.
+
+50:51.440 --> 50:53.440
+ If you look back,
+
+50:53.440 --> 50:59.440
+ what were some of your happiest moments as a researcher?
+
+50:59.440 --> 51:02.440
+ Some of the most profound moments.
+
+51:02.440 --> 51:06.440
+ Not in terms of their impact on society,
+
+51:06.440 --> 51:12.440
+ but in terms of their impact on how damn good you feel that day,
+
+51:12.440 --> 51:15.440
+ and you remember that moment.
+
+51:15.440 --> 51:20.440
+ You know, every time when you found something,
+
+51:20.440 --> 51:22.440
+ it is great.
+
+51:22.440 --> 51:24.440
+ It's a life.
+
+51:24.440 --> 51:26.440
+ Every simple thing.
+
+51:26.440 --> 51:32.440
+ But my general feeling that most of my time was wrong.
+
+51:32.440 --> 51:35.440
+ You should go again and again and again
+
+51:35.440 --> 51:39.440
+ and try to be honest in front of yourself.
+
+51:39.440 --> 51:41.440
+ Not to make interpretation,
+
+51:41.440 --> 51:46.440
+ but try to understand that it's related to ground truth.
+
+51:46.440 --> 51:52.440
+ It is not my blah, blah, blah interpretation or something like that.
+
+51:52.440 --> 51:57.440
+ But you're allowed to get excited at the possibility of discovery.
+
+51:57.440 --> 52:00.440
+ You have to double check it, but...
+
+52:00.440 --> 52:04.440
+ No, but how it's related to the other ground truth
+
+52:04.440 --> 52:10.440
+ is it just temporary or it is forever?
+
+52:10.440 --> 52:13.440
+ You know, you always have a feeling
+
+52:13.440 --> 52:17.440
+ when you found something,
+
+52:17.440 --> 52:19.440
+ how big is that?
+
+52:19.440 --> 52:23.440
+ So, 20 years ago, when we discovered statistical learning,
+
+52:23.440 --> 52:25.440
+ so nobody believed.
+
+52:25.440 --> 52:31.440
+ Except for one guy, Dudley from MIT.
+
+52:31.440 --> 52:36.440
+ And then in 20 years, it became fashion.
+
+52:36.440 --> 52:39.440
+ And the same with support vector machines.
+
+52:39.440 --> 52:41.440
+ That's kernel machines.
+
+52:41.440 --> 52:44.440
+ So with support vector machines and learning theory,
+
+52:44.440 --> 52:48.440
+ when you were working on it,
+
+52:48.440 --> 52:55.440
+ you had a sense that you had a sense of the profundity of it,
+
+52:55.440 --> 52:59.440
+ how this seems to be right.
+
+52:59.440 --> 53:01.440
+ It seems to be powerful.
+
+53:01.440 --> 53:04.440
+ Right, absolutely, immediately.
+
+53:04.440 --> 53:08.440
+ I recognize that it will last forever.
+
+53:08.440 --> 53:17.440
+ And now, when I found this invariance story,
+
+53:17.440 --> 53:21.440
+ I have a feeling that it is completely wrong.
+
+53:21.440 --> 53:25.440
+ Because I have proved that there are no different mechanisms.
+
+53:25.440 --> 53:30.440
+ Some say cosmetic improvement you can do,
+
+53:30.440 --> 53:34.440
+ but in terms of invariance,
+
+53:34.440 --> 53:38.440
+ you need both invariance and statistical learning
+
+53:38.440 --> 53:41.440
+ and they should work together.
+
+53:41.440 --> 53:47.440
+ But also, I'm happy that we can formulate
+
+53:47.440 --> 53:51.440
+ what is intelligence from that
+
+53:51.440 --> 53:54.440
+ and to separate from technical part.
+
+53:54.440 --> 53:56.440
+ And that is completely different.
+
+53:56.440 --> 53:58.440
+ Absolutely.
+
+53:58.440 --> 54:00.440
+ Well, Vladimir, thank you so much for talking today.
+
+54:00.440 --> 54:01.440
+ Thank you.
+
+54:01.440 --> 54:28.440
+ Thank you very much.
+

vtt/episode_006_large.vtt

@@ -0,0 +1,2603 @@
+WEBVTT
+
+00:00.000 --> 00:05.680
+ The following is a conversation with Guido van Rossum, creator of Python, one of the most popular
+
+00:05.680 --> 00:11.120
+ programming languages in the world, used in almost any application that involves computers
+
+00:11.120 --> 00:17.760
+ from web back end development to psychology, neuroscience, computer vision, robotics, deep
+
+00:17.760 --> 00:24.560
+ learning, natural language processing, and almost any subfield of AI. This conversation is part of
+
+00:24.560 --> 00:29.280
+ MIT course on artificial general intelligence and the artificial intelligence podcast.
+
+00:29.280 --> 00:36.080
+ If you enjoy it, subscribe on YouTube, iTunes, or your podcast provider of choice, or simply connect
+
+00:36.080 --> 00:44.720
+ with me on Twitter at Lex Friedman, spelled F R I D. And now, here's my conversation with Guido van
+
+00:44.720 --> 00:53.120
+ Rossum. You were born in the Netherlands in 1956. Your parents and the world around you was deeply
+
+00:53.120 --> 01:00.080
+ deeply impacted by World War Two, as was my family from the Soviet Union. So with that context,
+
+01:02.000 --> 01:07.360
+ what is your view of human nature? Are some humans inherently good,
+
+01:07.360 --> 01:12.240
+ and some inherently evil? Or do we all have both good and evil within us?
+
+01:12.240 --> 01:23.920
+ Guido van Rossum Ouch, I did not expect such a deep one. I, I guess we all have good and evil
+
+01:24.880 --> 01:31.440
+ potential in us. And a lot of it depends on circumstances and context.
+
+01:31.440 --> 01:38.800
+ Peter Bell out of that world, at least on the Soviet Union side in Europe, sort of out of
+
+01:38.800 --> 01:46.480
+ suffering, out of challenge, out of that kind of set of traumatic events, often emerges beautiful
+
+01:46.480 --> 01:53.200
+ art, music, literature. In an interview I read or heard, you said you enjoyed Dutch literature
+
+01:54.320 --> 01:59.760
+ when you were a child. Can you tell me about the books that had an influence on you in your
+
+01:59.760 --> 02:01.520
+ childhood? Guido van Rossum
+
+02:01.520 --> 02:09.120
+ Well, with as a teenager, my favorite writer was my favorite Dutch author was a guy named Willem
+
+02:09.120 --> 02:19.440
+ Frederik Hermans, who's writing, certainly his early novels were all about sort of
+
+02:19.440 --> 02:30.480
+ ambiguous things that happened during World War Two. I think he was a young adult during that time.
+
+02:31.600 --> 02:40.800
+ And he wrote about it a lot, and very interesting, very good books, I thought, I think.
+
+02:40.800 --> 02:42.560
+ Peter Bell In a nonfiction way?
+
+02:42.560 --> 02:46.400
+ Guido van Rossum No, it was all fiction, but it was
+
+02:46.400 --> 02:52.640
+ very much set in the ambiguous world of resistance against the Germans,
+
+02:54.560 --> 03:03.840
+ where often you couldn't tell whether someone was truly in the resistance or really a spy for the
+
+03:03.840 --> 03:11.280
+ Germans. And some of the characters in his novels sort of crossed that line, and you never really
+
+03:11.280 --> 03:13.840
+ find out what exactly happened.
+
+03:13.840 --> 03:16.880
+ Peter Bell And in his novels, there's always a
+
+03:16.880 --> 03:22.160
+ good guy and a bad guy, the nature of good and evil. Is it clear there's a hero?
+
+03:22.160 --> 03:25.120
+ Guido van Rossum No, his heroes are often more,
+
+03:25.120 --> 03:34.000
+ his main characters are often anti heroes. And so they're not very heroic. They're often,
+
+03:36.640 --> 03:40.800
+ they fail at some level to accomplish their lofty goals.
+
+03:40.800 --> 03:43.040
+ Peter Bell And looking at the trajectory
+
+03:43.040 --> 03:48.880
+ through the rest of your life, has literature, Dutch or English or translation had an impact
+
+03:50.560 --> 03:54.160
+ outside the technical world that you existed in?
+
+03:54.160 --> 03:59.920
+ Guido van Rossum I still read novels.
+
+04:00.640 --> 04:05.200
+ I don't think that it impacts me that much directly.
+
+04:05.200 --> 04:07.280
+ Peter Bell It doesn't impact your work.
+
+04:07.280 --> 04:10.080
+ Guido van Rossum It's a separate world.
+
+04:10.080 --> 04:17.440
+ My work is highly technical and sort of the world of art and literature doesn't really
+
+04:17.440 --> 04:19.120
+ directly have any bearing on it.
+
+04:19.120 --> 04:22.400
+ Peter Bell You don't think there's a creative element
+
+04:22.400 --> 04:26.880
+ to the design? You know, some would say design of a language is art.
+
+04:26.880 --> 04:32.160
+ Guido van Rossum I'm not disagreeing with that.
+
+04:32.160 --> 04:39.360
+ I'm just saying that sort of I don't feel direct influences from more traditional art
+
+04:39.360 --> 04:40.880
+ on my own creativity.
+
+04:40.880 --> 04:43.280
+ Peter Bell Right. Of course, you don't feel doesn't mean
+
+04:43.280 --> 04:46.000
+ it's not somehow deeply there in your subconscious.
+
+04:46.000 --> 04:48.240
+ Guido van Rossum Who knows?
+
+04:48.240 --> 04:51.200
+ Peter Bell Who knows? So let's go back to your early
+
+04:51.200 --> 04:57.440
+ teens. Your hobbies were building electronic circuits, building mechanical models.
+
+04:57.440 --> 05:06.080
+ What if you can just put yourself back in the mind of that young Guido 12, 13, 14, was
+
+05:06.080 --> 05:12.240
+ that grounded in a desire to create a system? So to create something? Or was it more just
+
+05:12.240 --> 05:14.720
+ tinkering? Just the joy of puzzle solving?
+
+05:14.720 --> 05:18.720
+ Guido van Rossum I think it was more the latter, actually.
+
+05:18.720 --> 05:29.920
+ I maybe towards the end of my high school period, I felt confident enough that that
+
+05:29.920 --> 05:39.120
+ I designed my own circuits that were sort of interesting somewhat. But a lot of that
+
+05:39.120 --> 05:46.000
+ time, I literally just took a model kit and follow the instructions, putting the things
+
+05:46.000 --> 05:51.680
+ together. I mean, I think the first few years that I built electronics kits, I really did
+
+05:51.680 --> 05:59.760
+ not have enough understanding of sort of electronics to really understand what I was doing. I mean,
+
+05:59.760 --> 06:06.480
+ I could debug it, and I could sort of follow the instructions very carefully, which has
+
+06:06.480 --> 06:14.560
+ always stayed with me. But I had a very naive model of, like, how do I build a circuit?
+
+06:14.560 --> 06:22.800
+ Of, like, how a transistor works? And I don't think that in those days, I had any understanding
+
+06:22.800 --> 06:32.560
+ of coils and capacitors, which actually sort of was a major problem when I started to build
+
+06:32.560 --> 06:39.840
+ more complex digital circuits, because I was unaware of the sort of the analog part of
+
+06:39.840 --> 06:50.080
+ the – how they actually work. And I would have things that – the schematic looked
+
+06:50.080 --> 06:57.440
+ – everything looked fine, and it didn't work. And what I didn't realize was that
+
+06:57.440 --> 07:02.720
+ there was some megahertz level oscillation that was throwing the circuit off, because
+
+07:02.720 --> 07:13.360
+ I had a sort of – two wires were too close, or the switches were kind of poorly built.
+
+07:13.360 --> 07:19.280
+ But through that time, I think it's really interesting and instructive to think about,
+
+07:19.280 --> 07:24.600
+ because echoes of it are in this time now. So in the 1970s, the personal computer was
+
+07:24.600 --> 07:33.200
+ being born. So did you sense, in tinkering with these circuits, did you sense the encroaching
+
+07:33.200 --> 07:39.320
+ revolution in personal computing? So if at that point, we would sit you down and ask
+
+07:39.320 --> 07:46.040
+ you to predict the 80s and the 90s, do you think you would be able to do so successfully
+
+07:46.040 --> 07:55.560
+ to unroll the process that's happening? No, I had no clue. I remember, I think, in
+
+07:55.560 --> 08:03.060
+ the summer after my senior year – or maybe it was the summer after my junior year – well,
+
+08:03.060 --> 08:11.600
+ at some point, I think, when I was 18, I went on a trip to the Math Olympiad in Eastern
+
+08:11.600 --> 08:16.920
+ Europe, and there was like – I was part of the Dutch team, and there were other nerdy
+
+08:16.920 --> 08:23.040
+ kids that sort of had different experiences, and one of them told me about this amazing
+
+08:23.040 --> 08:31.840
+ thing called a computer. And I had never heard that word. My own explorations in electronics
+
+08:31.840 --> 08:40.420
+ were sort of about very simple digital circuits, and I had sort of – I had the idea that
+
+08:40.420 --> 08:49.760
+ I somewhat understood how a digital calculator worked. And so there is maybe some echoes
+
+08:49.760 --> 08:56.440
+ of computers there, but I never made that connection. I didn't know that when my parents
+
+08:56.440 --> 09:03.520
+ were paying for magazine subscriptions using punched cards, that there was something called
+
+09:03.520 --> 09:08.260
+ a computer that was involved that read those cards and transferred the money between accounts.
+
+09:08.260 --> 09:15.880
+ I was also not really interested in those things. It was only when I went to university
+
+09:15.880 --> 09:23.120
+ to study math that I found out that they had a computer, and students were allowed to use
+
+09:23.120 --> 09:24.120
+ it.
+
+09:24.120 --> 09:27.800
+ And there were some – you're supposed to talk to that computer by programming it.
+
+09:27.800 --> 09:29.920
+ What did that feel like, finding –
+
+09:29.920 --> 09:35.440
+ Yeah, that was the only thing you could do with it. The computer wasn't really connected
+
+09:35.440 --> 09:41.400
+ to the real world. The only thing you could do was sort of – you typed your program
+
+09:41.400 --> 09:47.840
+ on a bunch of punched cards. You gave the punched cards to the operator, and an hour
+
+09:47.840 --> 09:55.520
+ later the operator gave you back your printout. And so all you could do was write a program
+
+09:55.520 --> 10:04.080
+ that did something very abstract. And I don't even remember what my first forays into programming
+
+10:04.080 --> 10:13.440
+ were, but they were sort of doing simple math exercises and just to learn how a programming
+
+10:13.440 --> 10:15.560
+ language worked.
+
+10:15.560 --> 10:21.680
+ Did you sense, okay, first year of college, you see this computer, you're able to have
+
+10:21.680 --> 10:29.420
+ a program and it generates some output. Did you start seeing the possibility of this,
+
+10:29.420 --> 10:34.920
+ or was it a continuation of the tinkering with circuits? Did you start to imagine that
+
+10:34.920 --> 10:42.460
+ one, the personal computer, but did you see it as something that is a tool, like a word
+
+10:42.460 --> 10:47.160
+ processing tool, maybe for gaming or something? Or did you start to imagine that it could
+
+10:47.160 --> 10:53.860
+ be going to the world of robotics, like the Frankenstein picture that you could create
+
+10:53.860 --> 10:59.640
+ an artificial being? There's like another entity in front of you. You did not see the
+
+10:59.640 --> 11:00.640
+ computer.
+
+11:00.640 --> 11:05.840
+ I don't think I really saw it that way. I was really more interested in the tinkering.
+
+11:05.840 --> 11:14.920
+ It's maybe not a sort of a complete coincidence that I ended up sort of creating a programming
+
+11:14.920 --> 11:20.360
+ language which is a tool for other programmers. I've always been very focused on the sort
+
+11:20.360 --> 11:28.920
+ of activity of programming itself and not so much what happens with the program you
+
+11:28.920 --> 11:29.920
+ write.
+
+11:29.920 --> 11:30.920
+ Right.
+
+11:30.920 --> 11:37.800
+ I do remember, and I don't remember, maybe in my second or third year, probably my second
+
+11:37.800 --> 11:46.680
+ actually, someone pointed out to me that there was this thing called Conway's Game of Life.
+
+11:46.680 --> 11:50.480
+ You're probably familiar with it. I think –
+
+11:50.480 --> 11:53.200
+ In the 70s, I think is when they came up with it.
+
+11:53.200 --> 12:00.840
+ So there was a Scientific American column by someone who did a monthly column about
+
+12:00.840 --> 12:06.580
+ mathematical diversions. I'm also blanking out on the guy's name. It was very famous
+
+12:06.580 --> 12:12.440
+ at the time and I think up to the 90s or so. And one of his columns was about Conway's
+
+12:12.440 --> 12:18.160
+ Game of Life and he had some illustrations and he wrote down all the rules and sort of
+
+12:18.160 --> 12:23.720
+ there was the suggestion that this was philosophically interesting, that that was why Conway had
+
+12:23.720 --> 12:31.480
+ called it that. And all I had was like the two pages photocopy of that article. I don't
+
+12:31.480 --> 12:40.200
+ even remember where I got it. But it spoke to me and I remember implementing a version
+
+12:40.200 --> 12:49.000
+ of that game for the batch computer we were using where I had a whole Pascal program that
+
+12:49.000 --> 12:56.480
+ sort of read an initial situation from input and read some numbers that said do so many
+
+12:56.480 --> 13:05.960
+ generations and print every so many generations and then out would come pages and pages of
+
+13:05.960 --> 13:08.480
+ sort of things.
+
+13:08.480 --> 13:18.360
+ I remember much later I've done a similar thing using Python but that original version
+
+13:18.360 --> 13:27.700
+ I wrote at the time I found interesting because I combined it with some trick I had learned
+
+13:27.700 --> 13:36.000
+ during my electronics hobbyist times. I essentially first on paper I designed a simple circuit
+
+13:36.000 --> 13:45.780
+ built out of logic gates that took nine bits of input which is sort of the cell and its
+
+13:45.780 --> 13:54.040
+ neighbors and produced a new value for that cell and it's like a combination of a half
+
+13:54.040 --> 14:01.040
+ adder and some other clipping. It's actually a full adder. And so I had worked that out
+
+14:01.040 --> 14:10.520
+ and then I translated that into a series of Boolean operations on Pascal integers where
+
+14:10.520 --> 14:21.740
+ you could use the integers as bitwise values. And so I could basically generate 60 bits
+
+14:21.740 --> 14:28.800
+ of a generation in like eight instructions or so.
+
+14:28.800 --> 14:29.800
+ Nice.
+
+14:29.800 --> 14:32.560
+ So I was proud of that.
+
+14:32.560 --> 14:38.120
+ It's funny that you mentioned, so for people who don't know Conway's Game of Life, it's
+
+14:38.120 --> 14:44.840
+ a cellular automata where there's single compute units that kind of look at their neighbors
+
+14:44.840 --> 14:50.080
+ and figure out what they look like in the next generation based on the state of their
+
+14:50.080 --> 14:57.840
+ neighbors and this is deeply distributed system in concept at least. And then there's simple
+
+14:57.840 --> 15:04.400
+ rules that all of them follow and somehow out of this simple rule when you step back
+
+15:04.400 --> 15:13.160
+ and look at what occurs, it's beautiful. There's an emergent complexity. Even though the underlying
+
+15:13.160 --> 15:17.440
+ rules are simple, there's an emergent complexity. Now the funny thing is you've implemented
+
+15:17.440 --> 15:23.660
+ this and the thing you're commenting on is you're proud of a hack you did to make it
+
+15:23.660 --> 15:30.800
+ run efficiently. When you're not commenting on, it's a beautiful implementation, you're
+
+15:30.800 --> 15:36.780
+ not commenting on the fact that there's an emergent complexity that you've coded a simple
+
+15:36.780 --> 15:42.960
+ program and when you step back and you print out the following generation after generation,
+
+15:42.960 --> 15:48.400
+ that's stuff that you may have not predicted would happen is happening.
+
+15:48.400 --> 15:53.600
+ And is that magic? I mean, that's the magic that all of us feel when we program. When
+
+15:53.600 --> 15:59.240
+ you create a program and then you run it and whether it's Hello World or it shows something
+
+15:59.240 --> 16:03.840
+ on screen, if there's a graphical component, are you seeing the magic in the mechanism
+
+16:03.840 --> 16:05.200
+ of creating that?
+
+16:05.200 --> 16:14.440
+ I think I went back and forth. As a student, we had an incredibly small budget of computer
+
+16:14.440 --> 16:20.280
+ time that we could use. It was actually measured. I once got in trouble with one of my professors
+
+16:20.280 --> 16:29.640
+ because I had overspent the department's budget. It's a different story.
+
+16:29.640 --> 16:36.900
+ I actually wanted the efficient implementation because I also wanted to explore what would
+
+16:36.900 --> 16:48.560
+ happen with a larger number of generations and a larger size of the board. Once the implementation
+
+16:48.560 --> 16:57.000
+ was flawless, I would feed it different patterns and then I think maybe there was a follow
+
+16:57.000 --> 17:03.620
+ up article where there were patterns that were like gliders, patterns that repeated
+
+17:03.620 --> 17:13.200
+ themselves after a number of generations but translated one or two positions to the right
+
+17:13.200 --> 17:21.720
+ or up or something like that. I remember things like glider guns. Well, you can Google Conway's
+
+17:21.720 --> 17:27.560
+ Game of Life. People still go aww and ooh over it.
+
+17:27.560 --> 17:32.680
+ For a reason because it's not really well understood why. I mean, this is what Stephen
+
+17:32.680 --> 17:40.240
+ Wolfram is obsessed about. We don't have the mathematical tools to describe the kind of
+
+17:40.240 --> 17:45.120
+ complexity that emerges in these kinds of systems. The only way you can do is to run
+
+17:45.120 --> 17:47.120
+ it.
+
+17:47.120 --> 17:55.720
+ I'm not convinced that it's sort of a problem that lends itself to classic mathematical
+
+17:55.720 --> 17:59.920
+ analysis.
+
+17:59.920 --> 18:05.120
+ One theory of how you create an artificial intelligence or artificial being is you kind
+
+18:05.120 --> 18:10.120
+ of have to, same with the Game of Life, you kind of have to create a universe and let
+
+18:10.120 --> 18:17.520
+ it run. That creating it from scratch in a design way, coding up a Python program that
+
+18:17.520 --> 18:22.760
+ creates a fully intelligent system may be quite challenging. You might need to create
+
+18:22.760 --> 18:27.120
+ a universe just like the Game of Life.
+
+18:27.120 --> 18:33.200
+ You might have to experiment with a lot of different universes before there is a set
+
+18:33.200 --> 18:41.480
+ of rules that doesn't essentially always just end up repeating itself in a trivial
+
+18:41.480 --> 18:42.480
+ way.
+
+18:42.480 --> 18:49.840
+ Yeah, and Stephen Wolfram works with these simple rules, says that it's kind of surprising
+
+18:49.840 --> 18:55.280
+ how quickly you find rules that create interesting things. You shouldn't be able to, but somehow
+
+18:55.280 --> 19:02.120
+ you do. And so maybe our universe is laden with rules that will create interesting things
+
+19:02.120 --> 19:07.440
+ that might not look like humans, but emergent phenomena that's interesting may not be as
+
+19:07.440 --> 19:09.440
+ difficult to create as we think.
+
+19:09.440 --> 19:10.440
+ Sure.
+
+19:10.440 --> 19:17.440
+ But let me sort of ask, at that time, some of the world, at least in popular press, was
+
+19:17.440 --> 19:25.680
+ kind of captivated, perhaps at least in America, by the idea of artificial intelligence, that
+
+19:25.680 --> 19:33.240
+ these computers would be able to think pretty soon. And did that touch you at all? In science
+
+19:33.240 --> 19:37.800
+ fiction or in reality in any way?
+
+19:37.800 --> 19:49.000
+ I didn't really start reading science fiction until much, much later. I think as a teenager
+
+19:49.000 --> 19:54.520
+ I read maybe one bundle of science fiction stories.
+
+19:54.520 --> 19:57.960
+ Was it in the background somewhere, like in your thoughts?
+
+19:57.960 --> 20:04.720
+ That sort of the using computers to build something intelligent always felt to me, because
+
+20:04.720 --> 20:12.920
+ I felt I had so much understanding of what actually goes on inside a computer. I knew
+
+20:12.920 --> 20:22.880
+ how many bits of memory it had and how difficult it was to program. And sort of, I didn't believe
+
+20:22.880 --> 20:30.560
+ at all that you could just build something intelligent out of that, that would really
+
+20:30.560 --> 20:40.600
+ sort of satisfy my definition of intelligence. I think the most influential thing that I
+
+20:40.600 --> 20:48.680
+ read in my early twenties was Gödel Escherbach. That was about consciousness, and that was
+
+20:48.680 --> 20:54.040
+ a big eye opener in some sense.
+
+20:54.040 --> 21:00.760
+ In what sense? So, on your own brain, did you at the time or do you now see your own
+
+21:00.760 --> 21:07.720
+ brain as a computer? Or is there a total separation of the way? So yeah, you're very pragmatically
+
+21:07.720 --> 21:14.600
+ practically know the limits of memory, the limits of this sequential computing or weakly
+
+21:14.600 --> 21:21.000
+ paralyzed computing, and you just know what we have now, and it's hard to see how it creates.
+
+21:21.000 --> 21:29.920
+ But it's also easy to see, it was in the 40s, 50s, 60s, and now at least similarities between
+
+21:29.920 --> 21:31.680
+ the brain and our computers.
+
+21:31.680 --> 21:43.200
+ Oh yeah, I mean, I totally believe that brains are computers in some sense. I mean, the rules
+
+21:43.200 --> 21:51.200
+ they use to play by are pretty different from the rules we can sort of implement in our
+
+21:51.200 --> 22:02.960
+ current hardware, but I don't believe in, like, a separate thing that infuses us with
+
+22:02.960 --> 22:10.480
+ intelligence or consciousness or any of that. There's no soul, I've been an atheist
+
+22:10.480 --> 22:18.800
+ probably from when I was 10 years old, just by thinking a bit about math and the universe,
+
+22:18.800 --> 22:26.640
+ and well, my parents were atheists. Now, I know that you could be an atheist and still
+
+22:26.640 --> 22:34.080
+ believe that there is something sort of about intelligence or consciousness that cannot
+
+22:34.080 --> 22:44.560
+ possibly emerge from a fixed set of rules. I am not in that camp. I totally see that,
+
+22:44.560 --> 22:53.680
+ sort of, given how many millions of years evolution took its time, DNA is a particular
+
+22:53.680 --> 23:07.040
+ machine that sort of encodes information and an unlimited amount of information in chemical
+
+23:07.040 --> 23:12.320
+ form and has figured out a way to replicate itself.
+
+23:12.320 --> 23:16.880
+ I thought that that was, maybe it's 300 million years ago, but I thought it was closer
+
+23:16.880 --> 23:25.120
+ to half a billion years ago, that that's sort of originated and it hasn't really changed,
+
+23:25.120 --> 23:32.040
+ that the sort of the structure of DNA hasn't changed ever since. That is like our binary
+
+23:32.040 --> 23:35.200
+ code that we have in hardware. I mean...
+
+23:35.200 --> 23:39.760
+ The basic programming language hasn't changed, but maybe the programming itself...
+
+23:39.760 --> 23:48.320
+ Obviously, it did sort of, it happened to be a set of rules that was good enough to
+
+23:48.320 --> 23:59.520
+ sort of develop endless variability and sort of the idea of self replicating molecules
+
+23:59.520 --> 24:05.360
+ competing with each other for resources and one type eventually sort of always taking
+
+24:05.360 --> 24:12.320
+ over. That happened before there were any fossils, so we don't know how that exactly
+
+24:12.320 --> 24:17.920
+ happened, but I believe it's clear that that did happen.
+
+24:17.920 --> 24:25.360
+ Can you comment on consciousness and how you see it? Because I think we'll talk about
+
+24:25.360 --> 24:30.080
+ programming quite a bit. We'll talk about, you know, intelligence connecting to programming
+
+24:30.080 --> 24:38.080
+ fundamentally, but consciousness is this whole other thing. Do you think about it often as
+
+24:38.080 --> 24:45.440
+ a developer of a programming language and as a human?
+
+24:45.440 --> 24:55.000
+ Those are pretty sort of separate topics. Sort of my line of work working with programming
+
+24:55.000 --> 25:02.720
+ does not involve anything that goes in the direction of developing intelligence or consciousness,
+
+25:02.720 --> 25:13.880
+ but sort of privately as an avid reader of popular science writing, I have some thoughts
+
+25:13.880 --> 25:25.680
+ which is mostly that I don't actually believe that consciousness is an all or nothing thing.
+
+25:25.680 --> 25:35.960
+ I have a feeling that, and I forget what I read that influenced this, but I feel that
+
+25:35.960 --> 25:41.400
+ if you look at a cat or a dog or a mouse, they have some form of intelligence. If you
+
+25:41.400 --> 25:54.040
+ look at a fish, it has some form of intelligence, and that evolution just took a long time,
+
+25:54.040 --> 26:01.320
+ but I feel that the sort of evolution of more and more intelligence that led to sort of
+
+26:01.320 --> 26:12.920
+ the human form of intelligence followed the evolution of the senses, especially the visual
+
+26:12.920 --> 26:20.480
+ sense. I mean, there is an enormous amount of processing that's needed to interpret
+
+26:20.480 --> 26:28.240
+ a scene, and humans are still better at that than computers are.
+
+26:28.240 --> 26:39.660
+ And I have a feeling that there is a sort of, the reason that like mammals in particular
+
+26:39.660 --> 26:47.960
+ developed the levels of consciousness that they have and that eventually sort of going
+
+26:47.960 --> 26:55.360
+ from intelligence to self awareness and consciousness has to do with sort of being a robot that
+
+26:55.360 --> 26:58.920
+ has very highly developed senses.
+
+26:58.920 --> 27:04.760
+ Has a lot of rich sensory information coming in, so that's a really interesting thought
+
+27:04.760 --> 27:14.200
+ that whatever that basic mechanism of DNA, whatever that basic building blocks of programming,
+
+27:14.200 --> 27:21.080
+ if you just add more abilities, more high resolution sensors, more sensors, you just
+
+27:21.080 --> 27:26.760
+ keep stacking those things on top that this basic programming in trying to survive develops
+
+27:26.760 --> 27:35.000
+ very interesting things that start to us humans to appear like intelligence and consciousness.
+
+27:35.000 --> 27:42.280
+ As far as robots go, I think that the self driving cars have that sort of the greatest
+
+27:42.280 --> 27:50.400
+ opportunity of developing something like that, because when I drive myself, I don't just
+
+27:50.400 --> 27:53.800
+ pay attention to the rules of the road.
+
+27:53.800 --> 28:01.220
+ I also look around and I get clues from that, oh, this is a shopping district, oh, here's
+
+28:01.220 --> 28:08.960
+ an old lady crossing the street, oh, here is someone carrying a pile of mail, there's
+
+28:08.960 --> 28:14.040
+ a mailbox, I bet you they're going to cross the street to reach that mailbox.
+
+28:14.040 --> 28:17.520
+ And I slow down, and I don't even think about that.
+
+28:17.520 --> 28:25.780
+ And so, there is so much where you turn your observations into an understanding of what
+
+28:25.780 --> 28:32.680
+ other consciousnesses are going to do, or what other systems in the world are going
+
+28:32.680 --> 28:37.400
+ to be, oh, that tree is going to fall.
+
+28:37.400 --> 28:46.800
+ I see sort of, I see much more of, I expect somehow that if anything is going to become
+
+28:46.800 --> 28:55.520
+ unconscious, it's going to be the self driving car and not the network of a bazillion computers
+
+28:55.520 --> 29:03.160
+ in a Google or Amazon data center that are all networked together to do whatever they
+
+29:03.160 --> 29:04.160
+ do.
+
+29:04.160 --> 29:09.640
+ So, in that sense, so you actually highlight, because that's what I work in Thomas Vehicles,
+
+29:09.640 --> 29:15.600
+ you highlight the big gap between what we currently can't do and what we truly need
+
+29:15.600 --> 29:18.500
+ to be able to do to solve the problem.
+
+29:18.500 --> 29:24.600
+ Under that formulation, then consciousness and intelligence is something that basically
+
+29:24.600 --> 29:30.020
+ a system should have in order to interact with us humans, as opposed to some kind of
+
+29:30.020 --> 29:35.280
+ abstract notion of a consciousness.
+
+29:35.280 --> 29:39.200
+ Consciousness is something that you need to have to be able to empathize, to be able to
+
+29:39.200 --> 29:47.440
+ fear, understand what the fear of death is, all these aspects that are important for interacting
+
+29:47.440 --> 29:56.160
+ with pedestrians, you need to be able to do basic computation based on our human desires
+
+29:56.160 --> 29:57.160
+ and thoughts.
+
+29:57.160 --> 30:02.280
+ And if you sort of, yeah, if you look at the dog, the dog clearly knows, I mean, I'm
+
+30:02.280 --> 30:07.340
+ not the dog owner, but I have friends who have dogs, the dogs clearly know what the
+
+30:07.340 --> 30:11.400
+ humans around them are going to do, or at least they have a model of what those humans
+
+30:11.400 --> 30:14.160
+ are going to do and they learn.
+
+30:14.160 --> 30:19.060
+ Some dogs know when you're going out and they want to go out with you, they're sad when
+
+30:19.060 --> 30:26.080
+ you leave them alone, they cry, they're afraid because they were mistreated when they were
+
+30:26.080 --> 30:31.040
+ younger.
+
+30:31.040 --> 30:39.280
+ We don't assign sort of consciousness to dogs, or at least not all that much, but I also
+
+30:39.280 --> 30:42.500
+ don't think they have none of that.
+
+30:42.500 --> 30:50.360
+ So I think it's consciousness and intelligence are not all or nothing.
+
+30:50.360 --> 30:52.780
+ The spectrum is really interesting.
+
+30:52.780 --> 30:58.760
+ But in returning to programming languages and the way we think about building these
+
+30:58.760 --> 31:03.260
+ kinds of things, about building intelligence, building consciousness, building artificial
+
+31:03.260 --> 31:04.260
+ beings.
+
+31:04.260 --> 31:10.920
+ So I think one of the exciting ideas came in the 17th century and with Leibniz, Hobbes,
+
+31:10.920 --> 31:18.520
+ Descartes, where there's this feeling that you can convert all thought, all reasoning,
+
+31:18.520 --> 31:24.480
+ all the thing that we find very special in our brains, you can convert all of that into
+
+31:24.480 --> 31:25.480
+ logic.
+
+31:25.480 --> 31:30.400
+ So you can formalize it, formal reasoning, and then once you formalize everything, all
+
+31:30.400 --> 31:34.400
+ of knowledge, then you can just calculate and that's what we're doing with our brains
+
+31:34.400 --> 31:35.400
+ is we're calculating.
+
+31:35.400 --> 31:40.240
+ So there's this whole idea that this is possible, that this we can actually program.
+
+31:40.240 --> 31:46.520
+ But they weren't aware of the concept of pattern matching in the sense that we are aware of
+
+31:46.520 --> 31:47.640
+ it now.
+
+31:47.640 --> 31:57.640
+ They sort of thought they had discovered incredible bits of mathematics like Newton's calculus
+
+31:57.640 --> 32:06.840
+ and their sort of idealism, their sort of extension of what they could do with logic
+
+32:06.840 --> 32:18.000
+ and math sort of went along those lines and they thought there's like, yeah, logic.
+
+32:18.000 --> 32:22.020
+ There's like a bunch of rules and a bunch of input.
+
+32:22.020 --> 32:28.600
+ They didn't realize that how you recognize a face is not just a bunch of rules but is
+
+32:28.600 --> 32:39.160
+ a shit ton of data plus a circuit that sort of interprets the visual clues and the context
+
+32:39.160 --> 32:49.400
+ and everything else and somehow can massively parallel pattern match against stored rules.
+
+32:49.400 --> 32:56.320
+ I mean, if I see you tomorrow here in front of the Dropbox office, I might recognize you.
+
+32:56.320 --> 33:01.320
+ Even if I'm wearing a different shirt, yeah, but if I see you tomorrow in a coffee shop
+
+33:01.320 --> 33:06.640
+ in Belmont, I might have no idea that it was you or on the beach or whatever.
+
+33:06.640 --> 33:10.160
+ I make those kind of mistakes myself all the time.
+
+33:10.160 --> 33:16.320
+ I see someone that I only know as like, oh, this person is a colleague of my wife's and
+
+33:16.320 --> 33:20.860
+ then I see them at the movies and I didn't recognize them.
+
+33:20.860 --> 33:29.320
+ But do you see those, you call it pattern matching, do you see that rules is unable
+
+33:29.320 --> 33:32.380
+ to encode that?
+
+33:32.380 --> 33:36.320
+ Everything you see, all the pieces of information you look around this room, I'm wearing a black
+
+33:36.320 --> 33:41.720
+ shirt, I have a certain height, I'm a human, all these, there's probably tens of thousands
+
+33:41.720 --> 33:45.680
+ of facts you pick up moment by moment about this scene.
+
+33:45.680 --> 33:50.000
+ You take them for granted and you aggregate them together to understand the scene.
+
+33:50.000 --> 33:53.800
+ You don't think all of that could be encoded to where at the end of the day, you can just
+
+33:53.800 --> 33:57.440
+ put it all on the table and calculate?
+
+33:57.440 --> 33:58.840
+ I don't know what that means.
+
+33:58.840 --> 34:08.680
+ I mean, yes, in the sense that there is no actual magic there, but there are enough layers
+
+34:08.680 --> 34:17.640
+ of abstraction from the facts as they enter my eyes and my ears to the understanding of
+
+34:17.640 --> 34:29.240
+ the scene that I don't think that AI has really covered enough of that distance.
+
+34:29.240 --> 34:37.800
+ It's like if you take a human body and you realize it's built out of atoms, well, that
+
+34:37.800 --> 34:41.960
+ is a uselessly reductionist view, right?
+
+34:41.960 --> 34:46.380
+ The body is built out of organs, the organs are built out of cells, the cells are built
+
+34:46.380 --> 34:53.240
+ out of proteins, the proteins are built out of amino acids, the amino acids are built
+
+34:53.240 --> 34:58.040
+ out of atoms and then you get to quantum mechanics.
+
+34:58.040 --> 34:59.920
+ So that's a very pragmatic view.
+
+34:59.920 --> 35:03.720
+ I mean, obviously as an engineer, I agree with that kind of view, but you also have
+
+35:03.720 --> 35:13.160
+ to consider the Sam Harris view of, well, intelligence is just information processing.
+
+35:13.160 --> 35:17.320
+ Like you said, you take in sensory information, you do some stuff with it and you come up
+
+35:17.320 --> 35:20.760
+ with actions that are intelligent.
+
+35:20.760 --> 35:22.480
+ That makes it sound so easy.
+
+35:22.480 --> 35:24.280
+ I don't know who Sam Harris is.
+
+35:24.280 --> 35:26.400
+ Oh, well, it's a philosopher.
+
+35:26.400 --> 35:29.680
+ So like this is how philosophers often think, right?
+
+35:29.680 --> 35:33.760
+ And essentially that's what Descartes was, is wait a minute, if there is, like you said,
+
+35:33.760 --> 35:39.320
+ no magic, so he basically says it doesn't appear like there's any magic, but we know
+
+35:39.320 --> 35:44.280
+ so little about it that it might as well be magic.
+
+35:44.280 --> 35:47.800
+ So just because we know that we're made of atoms, just because we know we're made
+
+35:47.800 --> 35:53.280
+ of organs, the fact that we know very little how to get from the atoms to organs in a way
+
+35:53.280 --> 36:00.400
+ that's recreatable means that you shouldn't get too excited just yet about the fact that
+
+36:00.400 --> 36:02.240
+ you figured out that we're made of atoms.
+
+36:02.240 --> 36:11.920
+ Right, and the same about taking facts as our sensory organs take them in and turning
+
+36:11.920 --> 36:19.820
+ that into reasons and actions, that sort of, there are a lot of abstractions that we haven't
+
+36:19.820 --> 36:23.960
+ quite figured out how to deal with those.
+
+36:23.960 --> 36:37.440
+ I mean, sometimes, I don't know if I can go on a tangent or not, so if I take a simple
+
+36:37.440 --> 36:45.640
+ program that parses, say I have a compiler that parses a program, in a sense the input
+
+36:45.640 --> 36:55.640
+ routine of that compiler, of that parser, is a sensing organ, and it builds up a mighty
+
+36:55.640 --> 37:01.960
+ complicated internal representation of the program it just saw, it doesn't just have
+
+37:01.960 --> 37:08.200
+ a linear sequence of bytes representing the text of the program anymore, it has an abstract
+
+37:08.200 --> 37:15.480
+ syntax tree, and I don't know how many of your viewers or listeners are familiar with
+
+37:15.480 --> 37:18.680
+ compiler technology, but there's…
+
+37:18.680 --> 37:21.880
+ Fewer and fewer these days, right?
+
+37:21.880 --> 37:24.920
+ That's also true, probably.
+
+37:24.920 --> 37:30.360
+ People want to take a shortcut, but there's sort of, this abstraction is a data structure
+
+37:30.360 --> 37:37.480
+ that the compiler then uses to produce outputs that is relevant, like a translation of that
+
+37:37.480 --> 37:47.880
+ program to machine code that can be executed by hardware, and then that data structure
+
+37:47.880 --> 37:50.600
+ gets thrown away.
+
+37:50.600 --> 38:02.560
+ When a fish or a fly sees, sort of gets visual impulses, I'm sure it also builds up some
+
+38:02.560 --> 38:10.000
+ data structure, and for the fly that may be very minimal, a fly may have only a few, I
+
+38:10.000 --> 38:17.680
+ mean, in the case of a fly's brain, I could imagine that there are few enough layers of
+
+38:17.680 --> 38:24.040
+ abstraction that it's not much more than when it's darker here than it is here, well
+
+38:24.040 --> 38:29.880
+ it can sense motion, because a fly sort of responds when you move your arm towards it,
+
+38:29.880 --> 38:39.240
+ so clearly its visual processing is intelligent, well, not intelligent, but it has an abstraction
+
+38:39.240 --> 38:46.440
+ for motion, and we still have similar things in, but much more complicated in our brains,
+
+38:46.440 --> 38:50.400
+ I mean, otherwise you couldn't drive a car if you couldn't, if you didn't have an
+
+38:50.400 --> 38:53.480
+ incredibly good abstraction for motion.
+
+38:53.480 --> 38:59.160
+ Yeah, in some sense, the same abstraction for motion is probably one of the primary
+
+38:59.160 --> 39:05.080
+ sources of our, of information for us, we just know what to do, I think we know what
+
+39:05.080 --> 39:08.280
+ to do with that, we've built up other abstractions on top.
+
+39:08.280 --> 39:14.120
+ We build much more complicated data structures based on that, and we build more persistent
+
+39:14.120 --> 39:20.320
+ data structures, sort of after some processing, some information sort of gets stored in our
+
+39:20.320 --> 39:27.240
+ memory pretty much permanently, and is available on recall, I mean, there are some things that
+
+39:27.240 --> 39:34.040
+ you sort of, you're conscious that you're remembering it, like, you give me your phone
+
+39:34.040 --> 39:39.560
+ number, I, well, at my age I have to write it down, but I could imagine, I could remember
+
+39:39.560 --> 39:46.240
+ those seven numbers, or ten digits, and reproduce them in a while, if I sort of repeat them
+
+39:46.240 --> 39:53.320
+ to myself a few times, so that's a fairly conscious form of memorization.
+
+39:53.320 --> 39:57.800
+ On the other hand, how do I recognize your face, I have no idea.
+
+39:57.800 --> 40:04.080
+ My brain has a whole bunch of specialized hardware that knows how to recognize faces,
+
+40:04.080 --> 40:10.200
+ I don't know how much of that is sort of coded in our DNA, and how much of that is
+
+40:10.200 --> 40:17.960
+ trained over and over between the ages of zero and three, but somehow our brains know
+
+40:17.960 --> 40:26.000
+ how to do lots of things like that, that are useful in our interactions with other humans,
+
+40:26.000 --> 40:29.880
+ without really being conscious of how it's done anymore.
+
+40:29.880 --> 40:36.200
+ Right, so our actual day to day lives, we're operating at the very highest level of abstraction,
+
+40:36.200 --> 40:39.760
+ we're just not even conscious of all the little details underlying it.
+
+40:39.760 --> 40:43.360
+ There's compilers on top of, it's like turtles on top of turtles, or turtles all the way
+
+40:43.360 --> 40:48.200
+ down, there's compilers all the way down, but that's essentially, you say that there's
+
+40:48.200 --> 40:54.920
+ no magic, that's what I, what I was trying to get at, I think, is with Descartes started
+
+40:54.920 --> 40:59.600
+ this whole train of saying that there's no magic, I mean, there's all this beforehand.
+
+40:59.600 --> 41:06.120
+ Well didn't Descartes also have the notion though that the soul and the body were fundamentally
+
+41:06.120 --> 41:07.120
+ separate?
+
+41:07.120 --> 41:11.800
+ Separate, yeah, I think he had to write in God in there for political reasons, so I don't
+
+41:11.800 --> 41:17.880
+ know actually, I'm not a historian, but there's notions in there that all of reasoning, all
+
+41:17.880 --> 41:20.120
+ of human thought can be formalized.
+
+41:20.120 --> 41:28.480
+ I think that continued in the 20th century with Russell and with Gadot's incompleteness
+
+41:28.480 --> 41:33.120
+ theorem, this debate of what are the limits of the things that could be formalized, that's
+
+41:33.120 --> 41:37.960
+ where the Turing machine came along, and this exciting idea, I mean, underlying a lot of
+
+41:37.960 --> 41:43.160
+ computing that you can do quite a lot with a computer.
+
+41:43.160 --> 41:47.640
+ You can encode a lot of the stuff we're talking about in terms of recognizing faces and so
+
+41:47.640 --> 41:53.960
+ on, theoretically, in an algorithm that can then run on a computer.
+
+41:53.960 --> 42:05.040
+ And in that context, I'd like to ask programming in a philosophical way, what does it mean
+
+42:05.040 --> 42:06.480
+ to program a computer?
+
+42:06.480 --> 42:13.360
+ So you said you write a Python program or compiled a C++ program that compiles to some
+
+42:13.360 --> 42:21.200
+ byte code, it's forming layers, you're programming a layer of abstraction that's higher, how
+
+42:21.200 --> 42:24.920
+ do you see programming in that context?
+
+42:24.920 --> 42:29.800
+ Can it keep getting higher and higher levels of abstraction?
+
+42:29.800 --> 42:35.960
+ I think at some point the higher levels of abstraction will not be called programming
+
+42:35.960 --> 42:44.720
+ and they will not resemble what we call programming at the moment.
+
+42:44.720 --> 42:52.080
+ There will not be source code, I mean, there will still be source code sort of at a lower
+
+42:52.080 --> 42:59.320
+ level of the machine, just like there are still molecules and electrons and sort of
+
+42:59.320 --> 43:09.120
+ proteins in our brains, but, and so there's still programming and system administration
+
+43:09.120 --> 43:15.960
+ and who knows what, to keep the machine running, but what the machine does is a different level
+
+43:15.960 --> 43:23.060
+ of abstraction in a sense, and as far as I understand the way that for the last decade
+
+43:23.060 --> 43:28.440
+ or more people have made progress with things like facial recognition or the self driving
+
+43:28.440 --> 43:38.200
+ cars is all by endless, endless amounts of training data where at least as a lay person,
+
+43:38.200 --> 43:47.420
+ and I feel myself totally as a lay person in that field, it looks like the researchers
+
+43:47.420 --> 43:57.400
+ who publish the results don't necessarily know exactly how their algorithms work, and
+
+43:57.400 --> 44:04.840
+ I often get upset when I sort of read a sort of a fluff piece about Facebook in the newspaper
+
+44:04.840 --> 44:12.680
+ or social networks and they say, well, algorithms, and that's like a totally different interpretation
+
+44:12.680 --> 44:19.240
+ of the word algorithm, because for me, the way I was trained or what I learned when I
+
+44:19.240 --> 44:25.920
+ was eight or ten years old, an algorithm is a set of rules that you completely understand
+
+44:25.920 --> 44:30.720
+ that can be mathematically analyzed and you can prove things.
+
+44:30.720 --> 44:37.840
+ You can like prove that Aristotelian sieve produces all prime numbers and only prime
+
+44:37.840 --> 44:38.840
+ numbers.
+
+44:38.840 --> 44:39.840
+ Yeah.
+
+44:39.840 --> 44:44.360
+ So I don't know if you know who Andrej Karpathy is, I'm afraid not.
+
+44:44.360 --> 44:51.980
+ So he's a head of AI at Tesla now, but he was at Stanford before and he has this cheeky
+
+44:51.980 --> 44:56.480
+ way of calling this concept software 2.0.
+
+44:56.480 --> 45:00.120
+ So let me disentangle that for a second.
+
+45:00.120 --> 45:06.080
+ So kind of what you're referring to is the traditional, the algorithm, the concept of
+
+45:06.080 --> 45:09.560
+ an algorithm, something that's there, it's clear, you can read it, you understand it,
+
+45:09.560 --> 45:14.800
+ you can prove it's functioning as kind of software 1.0.
+
+45:14.800 --> 45:21.920
+ And what software 2.0 is, is exactly what you described, which is you have neural networks,
+
+45:21.920 --> 45:26.600
+ which is a type of machine learning that you feed a bunch of data and that neural network
+
+45:26.600 --> 45:30.200
+ learns to do a function.
+
+45:30.200 --> 45:35.220
+ All you specify is the inputs and the outputs you want and you can't look inside.
+
+45:35.220 --> 45:37.040
+ You can't analyze it.
+
+45:37.040 --> 45:41.920
+ All you can do is train this function to map the inputs to the outputs by giving a lot
+
+45:41.920 --> 45:42.920
+ of data.
+
+45:42.920 --> 45:47.040
+ And that's as programming becomes getting a lot of data.
+
+45:47.040 --> 45:48.920
+ That's what programming is.
+
+45:48.920 --> 45:52.120
+ Well, that would be programming 2.0.
+
+45:52.120 --> 45:53.800
+ To programming 2.0.
+
+45:53.800 --> 45:55.600
+ I wouldn't call that programming.
+
+45:55.600 --> 45:57.480
+ It's just a different activity.
+
+45:57.480 --> 46:02.640
+ Just like building organs out of cells is not called chemistry.
+
+46:02.640 --> 46:09.680
+ Well, so let's just step back and think sort of more generally, of course.
+
+46:09.680 --> 46:18.080
+ But you know, it's like as a parent teaching your kids, things can be called programming.
+
+46:18.080 --> 46:22.720
+ In that same sense, that's how programming is being used.
+
+46:22.720 --> 46:27.080
+ You're providing them data, examples, use cases.
+
+46:27.080 --> 46:36.680
+ So imagine writing a function not by, not with for loops and clearly readable text,
+
+46:36.680 --> 46:42.760
+ but more saying, well, here's a lot of examples of what this function should take.
+
+46:42.760 --> 46:47.860
+ And here's a lot of examples of when it takes those functions, it should do this.
+
+46:47.860 --> 46:50.280
+ And then figure out the rest.
+
+46:50.280 --> 46:52.640
+ So that's the 2.0 concept.
+
+46:52.640 --> 46:58.560
+ And so the question I have for you is like, it's a very fuzzy way.
+
+46:58.560 --> 47:01.680
+ This is the reality of a lot of these pattern recognition systems and so on.
+
+47:01.680 --> 47:05.400
+ It's a fuzzy way of quote unquote programming.
+
+47:05.400 --> 47:09.160
+ What do you think about this kind of world?
+
+47:09.160 --> 47:13.640
+ Should it be called something totally different than programming?
+
+47:13.640 --> 47:21.000
+ If you're a software engineer, does that mean you're designing systems that are very, can
+
+47:21.000 --> 47:28.140
+ be systematically tested, evaluated, they have a very specific specification and then this
+
+47:28.140 --> 47:33.520
+ other fuzzy software 2.0 world, machine learning world, that's something else totally?
+
+47:33.520 --> 47:41.000
+ Or is there some intermixing that's possible?
+
+47:41.000 --> 47:48.600
+ Well the question is probably only being asked because we don't quite know what that software
+
+47:48.600 --> 47:51.400
+ 2.0 actually is.
+
+47:51.400 --> 48:02.960
+ And I think there is a truism that every task that AI has tackled in the past, at some point
+
+48:02.960 --> 48:09.160
+ we realized how it was done and then it was no longer considered part of artificial intelligence
+
+48:09.160 --> 48:15.200
+ because it was no longer necessary to use that term.
+
+48:15.200 --> 48:21.600
+ It was just, oh now we know how to do this.
+
+48:21.600 --> 48:30.320
+ And a new field of science or engineering has been developed and I don't know if sort
+
+48:30.320 --> 48:39.000
+ of every form of learning or sort of controlling computer systems should always be called programming.
+
+48:39.000 --> 48:43.720
+ So I don't know, maybe I'm focused too much on the terminology.
+
+48:43.720 --> 48:56.200
+ But I expect that there just will be different concepts where people with sort of different
+
+48:56.200 --> 49:07.920
+ education and a different model of what they're trying to do will develop those concepts.
+
+49:07.920 --> 49:17.240
+ I guess if you could comment on another way to put this concept is, I think the kind of
+
+49:17.240 --> 49:23.480
+ functions that neural networks provide is things as opposed to being able to upfront
+
+49:23.480 --> 49:28.720
+ prove that this should work for all cases you throw at it.
+
+49:28.720 --> 49:32.320
+ All you're able, it's the worst case analysis versus average case analysis.
+
+49:32.320 --> 49:39.800
+ All you're able to say is it seems on everything we've tested to work 99.9% of the time, but
+
+49:39.800 --> 49:44.160
+ we can't guarantee it and it fails in unexpected ways.
+
+49:44.160 --> 49:48.080
+ We can't even give you examples of how it fails in unexpected ways, but it's like really
+
+49:48.080 --> 49:50.120
+ good most of the time.
+
+49:50.120 --> 50:00.720
+ Is there no room for that in current ways we think about programming?
+
+50:00.720 --> 50:11.080
+ programming 1.0 is actually sort of getting to that point too, where the sort of the ideal
+
+50:11.080 --> 50:21.120
+ of a bug free program has been abandoned long ago by most software developers.
+
+50:21.120 --> 50:30.120
+ We only care about bugs that manifest themselves often enough to be annoying.
+
+50:30.120 --> 50:40.680
+ And we're willing to take the occasional crash or outage or incorrect result for granted
+
+50:40.680 --> 50:47.600
+ because we can't possibly, we don't have enough programmers to make all the code bug free
+
+50:47.600 --> 50:50.200
+ and it would be an incredibly tedious business.
+
+50:50.200 --> 50:56.320
+ And if you try to throw formal methods at it, it becomes even more tedious.
+
+50:56.320 --> 51:05.520
+ So every once in a while the user clicks on a link and somehow they get an error and the
+
+51:05.520 --> 51:07.360
+ average user doesn't panic.
+
+51:07.360 --> 51:14.840
+ They just click again and see if it works better the second time, which often magically
+
+51:14.840 --> 51:21.600
+ it does, or they go up and they try some other way of performing their tasks.
+
+51:21.600 --> 51:29.880
+ So that's sort of an end to end recovery mechanism and inside systems there is all
+
+51:29.880 --> 51:39.120
+ sorts of retries and timeouts and fallbacks and I imagine that that sort of biological
+
+51:39.120 --> 51:46.320
+ systems are even more full of that because otherwise they wouldn't survive.
+
+51:46.320 --> 51:54.160
+ Do you think programming should be taught and thought of as exactly what you just said?
+
+51:54.160 --> 52:01.560
+ I come from this kind of, you're always denying that fact always.
+
+52:01.560 --> 52:12.680
+ In sort of basic programming education, the sort of the programs you're having students
+
+52:12.680 --> 52:23.480
+ write are so small and simple that if there is a bug you can always find it and fix it.
+
+52:23.480 --> 52:29.720
+ Because the sort of programming as it's being taught in some, even elementary, middle schools,
+
+52:29.720 --> 52:36.680
+ in high school, introduction to programming classes in college typically, it's programming
+
+52:36.680 --> 52:38.920
+ in the small.
+
+52:38.920 --> 52:47.560
+ Very few classes sort of actually teach software engineering, building large systems.
+
+52:47.560 --> 52:51.360
+ Every summer here at Dropbox we have a large number of interns.
+
+52:51.360 --> 52:56.720
+ Every tech company on the West Coast has the same thing.
+
+52:56.720 --> 53:02.520
+ These interns are always amazed because this is the first time in their life that they
+
+53:02.520 --> 53:12.920
+ see what goes on in a really large software development environment.
+
+53:12.920 --> 53:20.280
+ Everything they've learned in college was almost always about a much smaller scale and
+
+53:20.280 --> 53:27.840
+ somehow that difference in scale makes a qualitative difference in how you do things and how you
+
+53:27.840 --> 53:29.600
+ think about it.
+
+53:29.600 --> 53:36.300
+ If you then take a few steps back into decades, 70s and 80s, when you were first thinking
+
+53:36.300 --> 53:41.840
+ about Python or just that world of programming languages, did you ever think that there would
+
+53:41.840 --> 53:46.720
+ be systems as large as underlying Google, Facebook, and Dropbox?
+
+53:46.720 --> 53:51.440
+ Did you, when you were thinking about Python?
+
+53:51.440 --> 53:57.520
+ I was actually always caught by surprise by sort of this, yeah, pretty much every stage
+
+53:57.520 --> 53:59.680
+ of computing.
+
+53:59.680 --> 54:07.280
+ So maybe just because you've spoken in other interviews, but I think the evolution of programming
+
+54:07.280 --> 54:13.080
+ languages are fascinating and it's especially because it leads from my perspective towards
+
+54:13.080 --> 54:15.640
+ greater and greater degrees of intelligence.
+
+54:15.640 --> 54:21.880
+ I learned the first programming language I played with in Russia was with the Turtle
+
+54:21.880 --> 54:22.880
+ logo.
+
+54:22.880 --> 54:24.840
+ Logo, yeah.
+
+54:24.840 --> 54:29.960
+ And if you look, I just have a list of programming languages, all of which I've now played with
+
+54:29.960 --> 54:30.960
+ a little bit.
+
+54:30.960 --> 54:36.640
+ I mean, they're all beautiful in different ways from Fortran, Cobalt, Lisp, Algol 60,
+
+54:36.640 --> 54:46.160
+ Basic, Logo again, C, as a few, the object oriented came along in the 60s, Simula, Pascal,
+
+54:46.160 --> 54:47.560
+ Smalltalk.
+
+54:47.560 --> 54:48.560
+ All of that leads.
+
+54:48.560 --> 54:49.560
+ They're all the classics.
+
+54:49.560 --> 54:50.560
+ The classics.
+
+54:50.560 --> 54:51.560
+ Yeah.
+
+54:51.560 --> 54:52.560
+ The classic hits, right?
+
+54:52.560 --> 54:58.280
+ Steam, that's built on top of Lisp.
+
+54:58.280 --> 55:05.900
+ On the database side, SQL, C++, and all of that leads up to Python, Pascal too, and that's
+
+55:05.900 --> 55:10.960
+ before Python, MATLAB, these kind of different communities, different languages.
+
+55:10.960 --> 55:13.240
+ So can you talk about that world?
+
+55:13.240 --> 55:18.680
+ I know that sort of Python came out of ABC, which I actually never knew that language.
+
+55:18.680 --> 55:24.400
+ I just, having researched this conversation, went back to ABC and it looks remarkably,
+
+55:24.400 --> 55:31.240
+ it has a lot of annoying qualities, but underneath those, like all caps and so on, but underneath
+
+55:31.240 --> 55:35.720
+ that, there's elements of Python that are quite, they're already there.
+
+55:35.720 --> 55:37.540
+ That's where I got all the good stuff.
+
+55:37.540 --> 55:38.540
+ All the good stuff.
+
+55:38.540 --> 55:41.580
+ So, but in that world, you're swimming these programming languages, were you focused on
+
+55:41.580 --> 55:48.080
+ just the good stuff in your specific circle, or did you have a sense of what is everyone
+
+55:48.080 --> 55:49.080
+ chasing?
+
+55:49.080 --> 55:57.000
+ You said that every programming language is built to scratch an itch.
+
+55:57.000 --> 55:59.920
+ Were you aware of all the itches in the community?
+
+55:59.920 --> 56:05.080
+ And if not, or if yes, I mean, what itch were you trying to scratch with Python?
+
+56:05.080 --> 56:12.040
+ Well, I'm glad I wasn't aware of all the itches because I would probably not have been able
+
+56:12.040 --> 56:14.040
+ to do anything.
+
+56:14.040 --> 56:19.760
+ I mean, if you're trying to solve every problem at once, you'll solve nothing.
+
+56:19.760 --> 56:23.880
+ Well, yeah, it's too overwhelming.
+
+56:23.880 --> 56:28.360
+ And so I had a very, very focused problem.
+
+56:28.360 --> 56:41.880
+ I wanted a programming language that sat somewhere in between shell scripting and C. And now,
+
+56:41.880 --> 56:48.720
+ arguably, there is like, one is higher level, one is lower level.
+
+56:48.720 --> 56:56.760
+ And Python is sort of a language of an intermediate level, although it's still pretty much at
+
+56:56.760 --> 57:00.560
+ the high level end.
+
+57:00.560 --> 57:11.200
+ I was thinking about much more about, I want a tool that I can use to be more productive
+
+57:11.200 --> 57:16.640
+ as a programmer in a very specific environment.
+
+57:16.640 --> 57:22.280
+ And I also had given myself a time budget for the development of the tool.
+
+57:22.280 --> 57:29.340
+ And that was sort of about three months for both the design, like thinking through what
+
+57:29.340 --> 57:38.900
+ are all the features of the language syntactically and semantically, and how do I implement the
+
+57:38.900 --> 57:43.680
+ whole pipeline from parsing the source code to executing it.
+
+57:43.680 --> 57:51.440
+ So I think both with the timeline and the goals, it seems like productivity was at the
+
+57:51.440 --> 57:54.040
+ core of it as a goal.
+
+57:54.040 --> 58:01.280
+ So like, for me in the 90s, and the first decade of the 21st century, I was always doing
+
+58:01.280 --> 58:07.620
+ machine learning, AI programming for my research was always in C++.
+
+58:07.620 --> 58:14.240
+ And then the other people who are a little more mechanical engineering, electrical engineering,
+
+58:14.240 --> 58:15.240
+ are MATLABby.
+
+58:15.240 --> 58:18.520
+ They're a little bit more MATLAB focused.
+
+58:18.520 --> 58:21.200
+ Those are the world, and maybe a little bit Java too.
+
+58:21.200 --> 58:29.160
+ But people who are more interested in emphasizing the object oriented nature of things.
+
+58:29.160 --> 58:34.920
+ So within the last 10 years or so, especially with the oncoming of neural networks and these
+
+58:34.920 --> 58:41.360
+ packages that are built on Python to interface with neural networks, I switched to Python
+
+58:41.360 --> 58:47.120
+ and it's just, I've noticed a significant boost that I can't exactly, because I don't
+
+58:47.120 --> 58:52.840
+ think about it, but I can't exactly put into words why I'm just much, much more productive.
+
+58:52.840 --> 58:56.400
+ Just being able to get the job done much, much faster.
+
+58:56.400 --> 59:01.880
+ So how do you think, whatever that qualitative difference is, I don't know if it's quantitative,
+
+59:01.880 --> 59:07.280
+ it could be just a feeling, I don't know if I'm actually more productive, but how
+
+59:07.280 --> 59:08.280
+ do you think about...
+
+59:08.280 --> 59:09.280
+ You probably are.
+
+59:09.280 --> 59:10.280
+ Yeah.
+
+59:10.280 --> 59:11.880
+ Well, that's right.
+
+59:11.880 --> 59:15.400
+ I think there's elements, let me just speak to one aspect that I think that was affecting
+
+59:15.400 --> 59:26.160
+ my productivity is C++ was, I really enjoyed creating performant code and creating a beautiful
+
+59:26.160 --> 59:31.000
+ structure where everything that, you know, this kind of going into this, especially with
+
+59:31.000 --> 59:37.080
+ the newer and newer standards of templated programming of just really creating this beautiful
+
+59:37.080 --> 59:42.000
+ formal structure that I found myself spending most of my time doing that as opposed to getting
+
+59:42.000 --> 59:47.520
+ it, parsing a file and extracting a few keywords or whatever the task was trying to do.
+
+59:47.520 --> 59:49.980
+ So what is it about Python?
+
+59:49.980 --> 59:54.520
+ How do you think of productivity in general as you were designing it now, sort of through
+
+59:54.520 --> 1:00:00.120
+ the decades, last three decades, what do you think it means to be a productive programmer?
+
+1:00:00.120 --> 1:00:03.560
+ And how did you try to design it into the language?
+
+1:00:03.560 --> 1:00:10.400
+ There are different tasks and as a programmer, it's useful to have different tools available
+
+1:00:10.400 --> 1:00:13.940
+ that sort of are suitable for different tasks.
+
+1:00:13.940 --> 1:00:25.600
+ So I still write C code, I still write shell code, but I write most of my things in Python.
+
+1:00:25.600 --> 1:00:33.000
+ Why do I still use those other languages, because sometimes the task just demands it.
+
+1:00:33.000 --> 1:00:39.000
+ And well, I would say most of the time the task actually demands a certain language because
+
+1:00:39.000 --> 1:00:45.600
+ the task is not write a program that solves problem X from scratch, but it's more like
+
+1:00:45.600 --> 1:00:56.680
+ fix a bug in existing program X or add a small feature to an existing large program.
+
+1:00:56.680 --> 1:01:10.160
+ But even if you're not constrained in your choice of language by context like that, there
+
+1:01:10.160 --> 1:01:21.360
+ is still the fact that if you write it in a certain language, then you have this balance
+
+1:01:21.360 --> 1:01:31.840
+ between how long does it take you to write the code and how long does the code run?
+
+1:01:31.840 --> 1:01:42.760
+ And when you're in the phase of exploring solutions, you often spend much more time
+
+1:01:42.760 --> 1:01:50.720
+ writing the code than running it because every time you've run it, you see that the output
+
+1:01:50.720 --> 1:01:58.480
+ is not quite what you wanted and you spend some more time coding.
+
+1:01:58.480 --> 1:02:06.760
+ And a language like Python just makes that iteration much faster because there are fewer
+
+1:02:06.760 --> 1:02:19.480
+ details that you have to get right before your program compiles and runs.
+
+1:02:19.480 --> 1:02:26.400
+ There are libraries that do all sorts of stuff for you, so you can sort of very quickly take
+
+1:02:26.400 --> 1:02:36.320
+ a bunch of existing components, put them together, and get your prototype application running.
+
+1:02:36.320 --> 1:02:42.860
+ Just like when I was building electronics, I was using a breadboard most of the time,
+
+1:02:42.860 --> 1:02:51.320
+ so I had this sprawl out circuit that if you shook it, it would stop working because it
+
+1:02:51.320 --> 1:02:58.800
+ was not put together very well, but it functioned and all I wanted was to see that it worked
+
+1:02:58.800 --> 1:03:05.000
+ and then move on to the next schematic or design or add something to it.
+
+1:03:05.000 --> 1:03:10.500
+ Once you've sort of figured out, oh, this is the perfect design for my radio or light
+
+1:03:10.500 --> 1:03:15.800
+ sensor or whatever, then you can say, okay, how do we design a PCB for this?
+
+1:03:15.800 --> 1:03:19.920
+ How do we solder the components in a small space?
+
+1:03:19.920 --> 1:03:32.840
+ How do we make it so that it is robust against, say, voltage fluctuations or mechanical disruption?
+
+1:03:32.840 --> 1:03:37.320
+ I know nothing about that when it comes to designing electronics, but I know a lot about
+
+1:03:37.320 --> 1:03:40.400
+ that when it comes to writing code.
+
+1:03:40.400 --> 1:03:46.080
+ So the initial steps are efficient, fast, and there's not much stuff that gets in the
+
+1:03:46.080 --> 1:03:56.680
+ way, but you're kind of describing, like Darwin described the evolution of species, right?
+
+1:03:56.680 --> 1:04:00.520
+ You're observing of what is true about Python.
+
+1:04:00.520 --> 1:04:07.800
+ Now if you take a step back, if the act of creating languages is art and you had three
+
+1:04:07.800 --> 1:04:15.640
+ months to do it, initial steps, so you just specified a bunch of goals, sort of things
+
+1:04:15.640 --> 1:04:19.400
+ that you observe about Python, perhaps you had those goals, but how do you create the
+
+1:04:19.400 --> 1:04:25.600
+ rules, the syntactic structure, the features that result in those?
+
+1:04:25.600 --> 1:04:29.880
+ So I have in the beginning and I have follow up questions about through the evolution of
+
+1:04:29.880 --> 1:04:35.440
+ Python too, but in the very beginning when you were sitting there creating the lexical
+
+1:04:35.440 --> 1:04:37.440
+ analyzer or whatever.
+
+1:04:37.440 --> 1:04:47.240
+ Python was still a big part of it because I sort of, I said to myself, I don't want
+
+1:04:47.240 --> 1:04:53.640
+ to have to design everything from scratch, I'm going to borrow features from other languages
+
+1:04:53.640 --> 1:04:54.640
+ that I like.
+
+1:04:54.640 --> 1:04:55.640
+ Oh, interesting.
+
+1:04:55.640 --> 1:04:58.360
+ So you basically, exactly, you first observe what you like.
+
+1:04:58.360 --> 1:05:05.240
+ Yeah, and so that's why if you're 17 years old and you want to sort of create a programming
+
+1:05:05.240 --> 1:05:11.600
+ language, you're not going to be very successful at it because you have no experience with
+
+1:05:11.600 --> 1:05:24.300
+ other languages, whereas I was in my, let's say mid 30s, I had written parsers before,
+
+1:05:24.300 --> 1:05:30.880
+ so I had worked on the implementation of ABC, I had spent years debating the design of ABC
+
+1:05:30.880 --> 1:05:37.520
+ with its authors, with its designers, I had nothing to do with the design, it was designed
+
+1:05:37.520 --> 1:05:42.080
+ fully as it ended up being implemented when I joined the team.
+
+1:05:42.080 --> 1:05:51.440
+ But so you borrow ideas and concepts and very concrete sort of local rules from different
+
+1:05:51.440 --> 1:05:58.920
+ languages like the indentation and certain other syntactic features from ABC, but I chose
+
+1:05:58.920 --> 1:06:07.960
+ to borrow string literals and how numbers work from C and various other things.
+
+1:06:07.960 --> 1:06:13.800
+ So in then, if you take that further, so yet you've had this funny sounding, but I think
+
+1:06:13.800 --> 1:06:21.000
+ surprisingly accurate and at least practical title of benevolent dictator for life for
+
+1:06:21.000 --> 1:06:25.240
+ quite, you know, for the last three decades or whatever, or no, not the actual title,
+
+1:06:25.240 --> 1:06:27.940
+ but functionally speaking.
+
+1:06:27.940 --> 1:06:34.280
+ So you had to make decisions, design decisions.
+
+1:06:34.280 --> 1:06:41.960
+ Can you maybe, let's take Python 2, so releasing Python 3 as an example.
+
+1:06:41.960 --> 1:06:47.240
+ It's not backward compatible to Python 2 in ways that a lot of people know.
+
+1:06:47.240 --> 1:06:50.640
+ So what was that deliberation, discussion, decision like?
+
+1:06:50.640 --> 1:06:51.640
+ Yeah.
+
+1:06:51.640 --> 1:06:54.520
+ What was the psychology of that experience?
+
+1:06:54.520 --> 1:06:58.520
+ Do you regret any aspects of how that experience undergone that?
+
+1:06:58.520 --> 1:07:03.040
+ Well, yeah, so it was a group process really.
+
+1:07:03.040 --> 1:07:11.880
+ At that point, even though I was BDFL in name and certainly everybody sort of respected
+
+1:07:11.880 --> 1:07:22.160
+ my position as the creator and the current sort of owner of the language design, I was
+
+1:07:22.160 --> 1:07:26.560
+ looking at everyone else for feedback.
+
+1:07:26.560 --> 1:07:35.280
+ Sort of Python 3.0 in some sense was sparked by other people in the community pointing
+
+1:07:35.280 --> 1:07:46.360
+ out, oh, well, there are a few issues that sort of bite users over and over.
+
+1:07:46.360 --> 1:07:48.920
+ Can we do something about that?
+
+1:07:48.920 --> 1:07:56.360
+ And for Python 3, we took a number of those Python words as they were called at the time
+
+1:07:56.360 --> 1:08:04.800
+ and we said, can we try to sort of make small changes to the language that address those
+
+1:08:04.800 --> 1:08:06.560
+ words?
+
+1:08:06.560 --> 1:08:15.360
+ And we had sort of in the past, we had always taken backwards compatibility very seriously.
+
+1:08:15.360 --> 1:08:20.420
+ And so many Python words in earlier versions had already been resolved because they could
+
+1:08:20.420 --> 1:08:29.740
+ be resolved while maintaining backwards compatibility or sort of using a very gradual path of evolution
+
+1:08:29.740 --> 1:08:31.960
+ of the language in a certain area.
+
+1:08:31.960 --> 1:08:39.760
+ And so we were stuck with a number of words that were widely recognized as problems, not
+
+1:08:39.760 --> 1:08:47.680
+ like roadblocks, but nevertheless sort of things that some people trip over and you know that
+
+1:08:47.680 --> 1:08:52.080
+ that's always the same thing that people trip over when they trip.
+
+1:08:52.080 --> 1:08:58.480
+ And we could not think of a backwards compatible way of resolving those issues.
+
+1:08:58.480 --> 1:09:01.920
+ But it's still an option to not resolve the issues, right?
+
+1:09:01.920 --> 1:09:07.920
+ And so yes, for a long time, we had sort of resigned ourselves to, well, okay, the language
+
+1:09:07.920 --> 1:09:13.400
+ is not going to be perfect in this way and that way and that way.
+
+1:09:13.400 --> 1:09:19.440
+ And we sort of, certain of these, I mean, there are still plenty of things where you
+
+1:09:19.440 --> 1:09:32.680
+ can say, well, that particular detail is better in Java or in R or in Visual Basic or whatever.
+
+1:09:32.680 --> 1:09:37.960
+ And we're okay with that because, well, we can't easily change it.
+
+1:09:37.960 --> 1:09:38.960
+ It's not too bad.
+
+1:09:38.960 --> 1:09:47.180
+ We can do a little bit with user education or we can have a static analyzer or warnings
+
+1:09:47.180 --> 1:09:49.440
+ in the parse or something.
+
+1:09:49.440 --> 1:09:54.880
+ But there were things where we thought, well, these are really problems that are not going
+
+1:09:54.880 --> 1:09:55.880
+ away.
+
+1:09:55.880 --> 1:10:00.840
+ They are getting worse in the future.
+
+1:10:00.840 --> 1:10:03.040
+ We should do something about that.
+
+1:10:03.040 --> 1:10:05.640
+ But ultimately there is a decision to be made, right?
+
+1:10:05.640 --> 1:10:13.320
+ So was that the toughest decision in the history of Python you had to make as the benevolent
+
+1:10:13.320 --> 1:10:15.180
+ dictator for life?
+
+1:10:15.180 --> 1:10:20.160
+ Or if not, what are there, maybe even on the smaller scale, what was the decision where
+
+1:10:20.160 --> 1:10:22.040
+ you were really torn up about?
+
+1:10:22.040 --> 1:10:25.800
+ Well, the toughest decision was probably to resign.
+
+1:10:25.800 --> 1:10:28.120
+ All right, let's go there.
+
+1:10:28.120 --> 1:10:29.360
+ Hold on a second then.
+
+1:10:29.360 --> 1:10:33.200
+ Let me just, because in the interest of time too, because I have a few cool questions for
+
+1:10:33.200 --> 1:10:38.160
+ you and let's touch a really important one because it was quite dramatic and beautiful
+
+1:10:38.160 --> 1:10:40.400
+ in certain kinds of ways.
+
+1:10:40.400 --> 1:10:47.320
+ In July this year, three months ago, you wrote, now that PEP 572 is done, I don't ever want
+
+1:10:47.320 --> 1:10:52.680
+ to have to fight so hard for a PEP and find that so many people despise my decisions.
+
+1:10:52.680 --> 1:10:56.240
+ I would like to remove myself entirely from the decision process.
+
+1:10:56.240 --> 1:11:01.520
+ I'll still be there for a while as an ordinary core developer and I'll still be available
+
+1:11:01.520 --> 1:11:05.440
+ to mentor people, possibly more available.
+
+1:11:05.440 --> 1:11:11.000
+ But I'm basically giving myself a permanent vacation from being BDFL, benevolent dictator
+
+1:11:11.000 --> 1:11:12.000
+ for life.
+
+1:11:12.000 --> 1:11:14.240
+ And you all will be on your own.
+
+1:11:14.240 --> 1:11:19.720
+ First of all, it's almost Shakespearean.
+
+1:11:19.720 --> 1:11:22.300
+ I'm not going to appoint a successor.
+
+1:11:22.300 --> 1:11:24.640
+ So what are you all going to do?
+
+1:11:24.640 --> 1:11:29.240
+ Create a democracy, anarchy, a dictatorship, a federation?
+
+1:11:29.240 --> 1:11:34.560
+ So that was a very dramatic and beautiful set of statements.
+
+1:11:34.560 --> 1:11:40.080
+ It's almost, it's open ended nature called the community to create a future for Python.
+
+1:11:40.080 --> 1:11:43.280
+ It's just kind of a beautiful aspect to it.
+
+1:11:43.280 --> 1:11:48.320
+ So what, and dramatic, you know, what was making that decision like?
+
+1:11:48.320 --> 1:11:54.560
+ What was on your heart, on your mind, stepping back now a few months later?
+
+1:11:54.560 --> 1:12:02.940
+ I'm glad you liked the writing because it was actually written pretty quickly.
+
+1:12:02.940 --> 1:12:14.240
+ It was literally something like after months and months of going around in circles, I had
+
+1:12:14.240 --> 1:12:26.240
+ finally approved PEP572, which I had a big hand in its design, although I didn't initiate
+
+1:12:26.240 --> 1:12:27.760
+ it originally.
+
+1:12:27.760 --> 1:12:36.320
+ I sort of gave it a bunch of nudges in a direction that would be better for the language.
+
+1:12:36.320 --> 1:12:40.320
+ So sorry, just to ask, is async IO, that's the one or no?
+
+1:12:40.320 --> 1:12:49.320
+ PEP572 was actually a small feature, which is assignment expressions.
+
+1:12:49.320 --> 1:12:58.200
+ That had been, there was just a lot of debate where a lot of people claimed that they knew
+
+1:12:58.200 --> 1:13:04.800
+ what was Pythonic and what was not Pythonic, and they knew that this was going to destroy
+
+1:13:04.800 --> 1:13:06.080
+ the language.
+
+1:13:06.080 --> 1:13:11.800
+ This was like a violation of Python's most fundamental design philosophy, and I thought
+
+1:13:11.800 --> 1:13:17.200
+ that was all bullshit because I was in favor of it, and I would think I know something
+
+1:13:17.200 --> 1:13:19.120
+ about Python's design philosophy.
+
+1:13:19.120 --> 1:13:26.340
+ So I was really tired and also stressed of that thing, and literally after sort of announcing
+
+1:13:26.340 --> 1:13:34.560
+ I was going to accept it, a certain Wednesday evening I had finally sent the email, it's
+
+1:13:34.560 --> 1:13:35.560
+ accepted.
+
+1:13:35.560 --> 1:13:38.920
+ I can just go implement it.
+
+1:13:38.920 --> 1:13:44.120
+ So I went to bed feeling really relieved, that's behind me.
+
+1:13:44.120 --> 1:13:54.320
+ And I wake up Thursday morning, 7 a.m., and I think, well, that was the last one that's
+
+1:13:54.320 --> 1:14:03.880
+ going to be such a terrible debate, and that's the last time that I let myself be so stressed
+
+1:14:03.880 --> 1:14:06.520
+ out about a pep decision.
+
+1:14:06.520 --> 1:14:07.920
+ I should just resign.
+
+1:14:07.920 --> 1:14:15.520
+ I've been sort of thinking about retirement for half a decade, I've been joking and sort
+
+1:14:15.520 --> 1:14:22.460
+ of mentioning retirement, sort of telling the community at some point in the future
+
+1:14:22.460 --> 1:14:29.400
+ I'm going to retire, don't take that FL part of my title too literally.
+
+1:14:29.400 --> 1:14:32.080
+ And I thought, okay, this is it.
+
+1:14:32.080 --> 1:14:39.200
+ I'm done, I had the day off, I wanted to have a good time with my wife, we were going to
+
+1:14:39.200 --> 1:14:48.480
+ a little beach town nearby, and in I think maybe 15, 20 minutes I wrote that thing that
+
+1:14:48.480 --> 1:14:51.320
+ you just called Shakespearean.
+
+1:14:51.320 --> 1:15:01.560
+ The funny thing is I didn't even realize what a monumental decision it was, because
+
+1:15:01.560 --> 1:15:09.200
+ five minutes later I read that link to my message back on Twitter, where people were
+
+1:15:09.200 --> 1:15:15.280
+ already discussing on Twitter, Guido resigned as the BDFL.
+
+1:15:15.280 --> 1:15:22.440
+ And I had posted it on an internal forum that I thought was only read by core developers,
+
+1:15:22.440 --> 1:15:28.520
+ so I thought I would at least have one day before the news would sort of get out.
+
+1:15:28.520 --> 1:15:36.200
+ The on your own aspects had also an element of quite, it was quite a powerful element
+
+1:15:36.200 --> 1:15:43.080
+ of the uncertainty that lies ahead, but can you also just briefly talk about, for example
+
+1:15:43.080 --> 1:15:49.920
+ I play guitar as a hobby for fun, and whenever I play people are super positive, super friendly,
+
+1:15:49.920 --> 1:15:52.680
+ they're like, this is awesome, this is great.
+
+1:15:52.680 --> 1:15:57.520
+ But sometimes I enter as an outside observer, I enter the programming community and there
+
+1:15:57.520 --> 1:16:05.560
+ seems to sometimes be camps on whatever the topic, and the two camps, the two or plus
+
+1:16:05.560 --> 1:16:11.700
+ camps, are often pretty harsh at criticizing the opposing camps.
+
+1:16:11.700 --> 1:16:14.880
+ As an onlooker, I may be totally wrong on this, but what do you think of this?
+
+1:16:14.880 --> 1:16:19.760
+ Yeah, holy wars are sort of a favorite activity in the programming community.
+
+1:16:19.760 --> 1:16:22.120
+ And what is the psychology behind that?
+
+1:16:22.120 --> 1:16:25.120
+ Is that okay for a healthy community to have?
+
+1:16:25.120 --> 1:16:29.760
+ Is that a productive force ultimately for the evolution of a language?
+
+1:16:29.760 --> 1:16:39.080
+ Well, if everybody is patting each other on the back and never telling the truth, it would
+
+1:16:39.080 --> 1:16:40.840
+ not be a good thing.
+
+1:16:40.840 --> 1:16:52.760
+ I think there is a middle ground where sort of being nasty to each other is not okay,
+
+1:16:52.760 --> 1:17:01.760
+ but there is a middle ground where there is healthy ongoing criticism and feedback that
+
+1:17:01.760 --> 1:17:04.780
+ is very productive.
+
+1:17:04.780 --> 1:17:07.760
+ And you mean at every level you see that.
+
+1:17:07.760 --> 1:17:17.760
+ I mean, someone proposes to fix a very small issue in a code base, chances are that some
+
+1:17:17.760 --> 1:17:27.080
+ reviewer will sort of respond by saying, well, actually, you can do it better the other way.
+
+1:17:27.080 --> 1:17:34.360
+ When it comes to deciding on the future of the Python core developer community, we now
+
+1:17:34.360 --> 1:17:41.160
+ have, I think, five or six competing proposals for a constitution.
+
+1:17:41.160 --> 1:17:48.040
+ So that future, do you have a fear of that future, do you have a hope for that future?
+
+1:17:48.040 --> 1:17:51.280
+ I'm very confident about that future.
+
+1:17:51.280 --> 1:17:58.920
+ By and large, I think that the debate has been very healthy and productive.
+
+1:17:58.920 --> 1:18:07.680
+ And I actually, when I wrote that resignation email, I knew that Python was in a very good
+
+1:18:07.680 --> 1:18:16.840
+ spot and that the Python core developer community, the group of 50 or 100 people who sort of
+
+1:18:16.840 --> 1:18:24.720
+ write or review most of the code that goes into Python, those people get along very well
+
+1:18:24.720 --> 1:18:27.680
+ most of the time.
+
+1:18:27.680 --> 1:18:40.120
+ A large number of different areas of expertise are represented, different levels of experience
+
+1:18:40.120 --> 1:18:45.440
+ in the Python core dev community, different levels of experience completely outside it
+
+1:18:45.440 --> 1:18:53.040
+ in software development in general, large systems, small systems, embedded systems.
+
+1:18:53.040 --> 1:19:03.880
+ So I felt okay resigning because I knew that the community can really take care of itself.
+
+1:19:03.880 --> 1:19:12.360
+ And out of a grab bag of future feature developments, let me ask if you can comment, maybe on all
+
+1:19:12.360 --> 1:19:19.120
+ very quickly, concurrent programming, parallel computing, async IO.
+
+1:19:19.120 --> 1:19:24.880
+ These are things that people have expressed hope, complained about, whatever, have discussed
+
+1:19:24.880 --> 1:19:25.880
+ on Reddit.
+
+1:19:25.880 --> 1:19:32.200
+ Async IO, so the parallelization in general, packaging, I was totally clueless on this.
+
+1:19:32.200 --> 1:19:38.600
+ I just used pip to install stuff, but apparently there's pipenv, poetry, there's these dependency
+
+1:19:38.600 --> 1:19:41.300
+ packaging systems that manage dependencies and so on.
+
+1:19:41.300 --> 1:19:45.520
+ They're emerging and there's a lot of confusion about what's the right thing to use.
+
+1:19:45.520 --> 1:19:56.360
+ Then also functional programming, are we going to get more functional programming or not,
+
+1:19:56.360 --> 1:19:59.040
+ this kind of idea.
+
+1:19:59.040 --> 1:20:08.280
+ And of course the GIL connected to the parallelization, I suppose, the global interpreter lock problem.
+
+1:20:08.280 --> 1:20:12.800
+ Can you just comment on whichever you want to comment on?
+
+1:20:12.800 --> 1:20:25.440
+ Well, let's take the GIL and parallelization and async IO as one topic.
+
+1:20:25.440 --> 1:20:35.820
+ I'm not that hopeful that Python will develop into a sort of high concurrency, high parallelism
+
+1:20:35.820 --> 1:20:37.960
+ language.
+
+1:20:37.960 --> 1:20:44.800
+ That's sort of the way the language is designed, the way most users use the language, the way
+
+1:20:44.800 --> 1:20:50.280
+ the language is implemented, all make that a pretty unlikely future.
+
+1:20:50.280 --> 1:20:56.040
+ So you think it might not even need to, really the way people use it, it might not be something
+
+1:20:56.040 --> 1:20:58.160
+ that should be of great concern.
+
+1:20:58.160 --> 1:21:05.620
+ I think async IO is a special case because it sort of allows overlapping IO and only
+
+1:21:05.620 --> 1:21:18.160
+ IO and that is a sort of best practice of supporting very high throughput IO, many connections
+
+1:21:18.160 --> 1:21:21.680
+ per second.
+
+1:21:21.680 --> 1:21:22.780
+ I'm not worried about that.
+
+1:21:22.780 --> 1:21:25.280
+ I think async IO will evolve.
+
+1:21:25.280 --> 1:21:27.440
+ There are a couple of competing packages.
+
+1:21:27.440 --> 1:21:36.800
+ We have some very smart people who are sort of pushing us to make async IO better.
+
+1:21:36.800 --> 1:21:43.800
+ Parallel computing, I think that Python is not the language for that.
+
+1:21:43.800 --> 1:21:53.560
+ There are ways to work around it, but you can't expect to write an algorithm in Python
+
+1:21:53.560 --> 1:21:57.440
+ and have a compiler automatically parallelize that.
+
+1:21:57.440 --> 1:22:03.520
+ What you can do is use a package like NumPy and there are a bunch of other very powerful
+
+1:22:03.520 --> 1:22:12.480
+ packages that sort of use all the CPUs available because you tell the package, here's the data,
+
+1:22:12.480 --> 1:22:19.040
+ here's the abstract operation to apply over it, go at it, and then we're back in the C++
+
+1:22:19.040 --> 1:22:20.040
+ world.
+
+1:22:20.040 --> 1:22:24.600
+ Those packages are themselves implemented usually in C++.
+
+1:22:24.600 --> 1:22:28.000
+ That's where TensorFlow and all these packages come in, where they parallelize across GPUs,
+
+1:22:28.000 --> 1:22:30.480
+ for example, they take care of that for you.
+
+1:22:30.480 --> 1:22:36.600
+ In terms of packaging, can you comment on the future of packaging in Python?
+
+1:22:36.600 --> 1:22:42.640
+ Packaging has always been my least favorite topic.
+
+1:22:42.640 --> 1:22:55.600
+ It's a really tough problem because the OS and the platform want to own packaging, but
+
+1:22:55.600 --> 1:23:01.000
+ their packaging solution is not specific to a language.
+
+1:23:01.000 --> 1:23:07.480
+ If you take Linux, there are two competing packaging solutions for Linux or for Unix
+
+1:23:07.480 --> 1:23:15.000
+ in general, but they all work across all languages.
+
+1:23:15.000 --> 1:23:24.760
+ Several languages like Node, JavaScript, Ruby, and Python all have their own packaging solutions
+
+1:23:24.760 --> 1:23:29.480
+ that only work within the ecosystem of that language.
+
+1:23:29.480 --> 1:23:31.920
+ What should you use?
+
+1:23:31.920 --> 1:23:34.560
+ That is a tough problem.
+
+1:23:34.560 --> 1:23:43.520
+ My own approach is I use the system packaging system to install Python, and I use the Python
+
+1:23:43.520 --> 1:23:49.280
+ packaging system then to install third party Python packages.
+
+1:23:49.280 --> 1:23:51.480
+ That's what most people do.
+
+1:23:51.480 --> 1:23:56.400
+ Ten years ago, Python packaging was really a terrible situation.
+
+1:23:56.400 --> 1:24:05.360
+ Nowadays, pip is the future, there is a separate ecosystem for numerical and scientific Python
+
+1:24:05.360 --> 1:24:08.200
+ based on Anaconda.
+
+1:24:08.200 --> 1:24:09.760
+ Those two can live together.
+
+1:24:09.760 --> 1:24:13.600
+ I don't think there is a need for more than that.
+
+1:24:13.600 --> 1:24:14.600
+ That's packaging.
+
+1:24:14.600 --> 1:24:18.720
+ Well, at least for me, that's where I've been extremely happy.
+
+1:24:18.720 --> 1:24:22.320
+ I didn't even know this was an issue until it was brought up.
+
+1:24:22.320 --> 1:24:27.600
+ In the interest of time, let me sort of skip through a million other questions I have.
+
+1:24:27.600 --> 1:24:32.880
+ So I watched the five and a half hour oral history that you've done with the Computer
+
+1:24:32.880 --> 1:24:37.600
+ History Museum, and the nice thing about it, it gave this, because of the linear progression
+
+1:24:37.600 --> 1:24:44.480
+ of the interview, it gave this feeling of a life, you know, a life well lived with interesting
+
+1:24:44.480 --> 1:24:52.160
+ things in it, sort of a pretty, I would say a good spend of this little existence we have
+
+1:24:52.160 --> 1:24:53.160
+ on Earth.
+
+1:24:53.160 --> 1:24:59.840
+ So, outside of your family, looking back, what about this journey are you really proud
+
+1:24:59.840 --> 1:25:00.840
+ of?
+
+1:25:00.840 --> 1:25:07.040
+ Are there moments that stand out, accomplishments, ideas?
+
+1:25:07.040 --> 1:25:14.040
+ Is it the creation of Python itself that stands out as a thing that you look back and say,
+
+1:25:14.040 --> 1:25:16.480
+ damn, I did pretty good there?
+
+1:25:16.480 --> 1:25:25.520
+ Well, I would say that Python is definitely the best thing I've ever done, and I wouldn't
+
+1:25:25.520 --> 1:25:36.560
+ sort of say just the creation of Python, but the way I sort of raised Python, like a baby.
+
+1:25:36.560 --> 1:25:42.480
+ I didn't just conceive a child, but I raised a child, and now I'm setting the child free
+
+1:25:42.480 --> 1:25:50.200
+ in the world, and I've set up the child to sort of be able to take care of himself, and
+
+1:25:50.200 --> 1:25:52.640
+ I'm very proud of that.
+
+1:25:52.640 --> 1:25:56.740
+ And as the announcer of Monty Python's Flying Circus used to say, and now for something
+
+1:25:56.740 --> 1:26:02.280
+ completely different, do you have a favorite Monty Python moment, or a moment in Hitchhiker's
+
+1:26:02.280 --> 1:26:07.720
+ Guide, or any other literature show or movie that cracks you up when you think about it?
+
+1:26:07.720 --> 1:26:11.320
+ You can always play me the dead parrot sketch.
+
+1:26:11.320 --> 1:26:13.680
+ Oh, that's brilliant.
+
+1:26:13.680 --> 1:26:14.680
+ That's my favorite as well.
+
+1:26:14.680 --> 1:26:15.680
+ It's pushing up the daisies.
+
+1:26:15.680 --> 1:26:20.680
+ Okay, Greta, thank you so much for talking with me today.
+
+1:26:20.680 --> 1:26:44.080
+ Lex, this has been a great conversation.
+

vtt/episode_006_small.vtt

@@ -0,0 +1,2051 @@
+WEBVTT
+
+00:00.000 --> 00:04.640
+ The following is a conversation with Guido Van Rossum, creator of Python,
+
+00:04.640 --> 00:09.520
+ one of the most popular programming languages in the world, used in almost any application
+
+00:09.520 --> 00:16.000
+ that involves computers, from web backend development to psychology, neuroscience,
+
+00:16.000 --> 00:21.040
+ computer vision, robotics, deep learning, natural language processing, and almost any
+
+00:21.040 --> 00:27.360
+ subfield of AI. This conversation is part of MIT course on artificial general intelligence
+
+00:27.360 --> 00:33.760
+ and the artificial intelligence podcast. If you enjoy it, subscribe on YouTube, iTunes,
+
+00:33.760 --> 00:39.040
+ or your podcast provider of choice, or simply connect with me on Twitter at Lex Friedman,
+
+00:39.040 --> 00:45.520
+ spelled F R I D. And now here's my conversation with Guido Van Rossum.
+
+00:46.320 --> 00:53.600
+ You were born in the Netherlands in 1956. Your parents and the world around you was deeply
+
+00:53.600 --> 01:00.080
+ impacted by World War Two, as was my family from the Soviet Union. So with that context,
+
+01:01.920 --> 01:08.240
+ what is your view of human nature? Are some humans inherently good and some inherently
+
+01:08.240 --> 01:12.240
+ evil, or do we all have both good and evil within us?
+
+01:12.240 --> 01:26.320
+ Ouch, I did not expect such a deep one. I guess we all have good and evil potential in us,
+
+01:26.320 --> 01:31.440
+ and a lot of it depends on circumstances and context.
+
+01:32.960 --> 01:39.360
+ Out of that world, at least on the Soviet Union side in Europe, sort of out of suffering,
+
+01:39.360 --> 01:46.800
+ out of challenge, out of that kind of set of traumatic events, often emerges beautiful art,
+
+01:46.800 --> 01:53.200
+ music, literature. In an interview I read or heard, you said you enjoyed Dutch literature
+
+01:54.320 --> 01:59.680
+ when you were a child. Can you tell me about the books that had an influence on you in your
+
+01:59.680 --> 02:06.960
+ childhood? Well, as a teenager, my favorite writer was, my favorite Dutch author was
+
+02:06.960 --> 02:17.920
+ a guy named Willem Friedrich Hermans, whose writing, certainly his early novels, were all about
+
+02:19.040 --> 02:29.520
+ sort of ambiguous things that happened during World War Two. I think he was a young adult
+
+02:29.520 --> 02:40.400
+ during that time, and he wrote about it a lot and very interesting, very good books, I thought,
+
+02:40.400 --> 02:50.000
+ I think. In a nonfiction way? No, it was all fiction, but it was very much set in the ambiguous
+
+02:50.000 --> 02:57.680
+ world of resistance against the Germans, where often you couldn't tell whether someone
+
+02:57.680 --> 03:07.280
+ was truly in the resistance or really a spy for the Germans, and some of the characters in his
+
+03:07.280 --> 03:13.840
+ novels sort of crossed that line, and you never really find out what exactly happened.
+
+03:14.800 --> 03:20.080
+ And in his novels, there's always a good guy and a bad guy, in the nature of good and evil,
+
+03:20.080 --> 03:30.320
+ is it clear there's a hero? No, his main characters are often antiheroes, and so they're
+
+03:30.320 --> 03:40.800
+ not very heroic. They fail at some level to accomplish their lofty goals.
+
+03:41.680 --> 03:45.120
+ And looking at the trajectory through the rest of your life, has literature,
+
+03:45.120 --> 03:54.240
+ Dutch or English or translation had an impact outside the technical world that you existed in?
+
+03:58.160 --> 04:05.200
+ I still read novels. I don't think that it impacts me that much directly.
+
+04:06.240 --> 04:14.320
+ It doesn't impact your work. It's a separate world. My work is highly technical and sort of
+
+04:14.320 --> 04:19.200
+ the world of art and literature doesn't really directly have any bearing on it.
+
+04:20.320 --> 04:26.880
+ You don't think there's a creative element to the design of a language's art?
+
+04:30.560 --> 04:36.080
+ I'm not disagreeing with that. I'm just saying that I don't feel
+
+04:36.800 --> 04:41.840
+ direct influences from more traditional art on my own creativity.
+
+04:41.840 --> 04:46.560
+ All right, of course, you don't feel doesn't mean it's not somehow deeply there in your subconscious.
+
+04:50.240 --> 04:56.880
+ So let's go back to your early teens. Your hobbies were building electronic circuits,
+
+04:56.880 --> 05:03.920
+ building mechanical models. If you can just put yourself back in the mind of that
+
+05:03.920 --> 05:13.200
+ young widow, 12, 13, 14, was that grounded in a desire to create a system? So to create
+
+05:13.200 --> 05:17.600
+ something? Or was it more just tinkering? Just the joy of puzzle solving?
+
+05:19.280 --> 05:27.200
+ I think it was more the latter, actually. Maybe towards the end of my high school
+
+05:27.200 --> 05:36.720
+ period, I felt confident enough that I designed my own circuits that were sort of interesting.
+
+05:38.720 --> 05:48.000
+ Somewhat. But a lot of that time, I literally just took a model kit and followed the instructions,
+
+05:48.000 --> 05:53.840
+ putting the things together. I mean, I think the first few years that I built electronics kits,
+
+05:53.840 --> 06:01.520
+ I really did not have enough understanding of electronics to really understand what I was
+
+06:01.520 --> 06:09.600
+ doing. I could debug it and I could follow the instructions very carefully, which has always
+
+06:09.600 --> 06:20.400
+ stayed with me. But I had a very naive model of how a transistor works. I don't think that in
+
+06:20.400 --> 06:31.360
+ those days, I had any understanding of coils and capacitors, which actually was a major problem
+
+06:31.360 --> 06:39.120
+ when I started to build more complex digital circuits, because I was unaware of the analog
+
+06:39.120 --> 06:50.960
+ part of how they actually work. And I would have things that the schematic looked, everything
+
+06:50.960 --> 06:57.920
+ looked fine, and it didn't work. And what I didn't realize was that there was some
+
+06:58.640 --> 07:04.880
+ megahertz level oscillation that was throwing the circuit off, because I had a sort of,
+
+07:04.880 --> 07:12.080
+ two wires were too close or the switches were kind of poorly built.
+
+07:13.040 --> 07:18.960
+ But through that time, I think it's really interesting and instructive to think about,
+
+07:18.960 --> 07:24.880
+ because there's echoes of it in this time now. So in the 1970s, the personal computer was being
+
+07:24.880 --> 07:33.920
+ born. So did you sense in tinkering with these circuits, did you sense the encroaching revolution
+
+07:33.920 --> 07:40.000
+ in personal computing? So if at that point, you're sitting, we'll sit you down and ask you to predict
+
+07:40.000 --> 07:47.920
+ the 80s and the 90s, do you think you would be able to do so successfully to unroll this,
+
+07:47.920 --> 07:57.840
+ the process? No, I had no clue. I, I remember, I think in the summer after my senior year,
+
+07:57.840 --> 08:04.240
+ or maybe it was the summer after my junior year. Well, at some point, I think when I was 18,
+
+08:04.240 --> 08:13.200
+ I went on a trip to the math Olympiad in Eastern Europe. And there was like, I was part of the
+
+08:13.200 --> 08:20.080
+ Dutch team. And there were other nerdy kids that sort of had different experiences. And one of
+
+08:20.080 --> 08:26.240
+ them told me about this amazing thing called a computer. And I had never heard that word.
+
+08:26.240 --> 08:35.680
+ My own explorations in electronics were sort of about very simple digital circuits. And I,
+
+08:35.680 --> 08:43.680
+ I had sort of, I had the idea that I somewhat understood how a digital calculator worked. And
+
+08:43.680 --> 08:51.440
+ so there is maybe some echoes of computers there, but I didn't, didn't, I never made that connection.
+
+08:51.440 --> 08:59.360
+ I didn't know that when my parents were paying for magazine subscriptions using punched cards,
+
+08:59.360 --> 09:04.480
+ that there was something called a computer that was involved that read those cards and
+
+09:04.480 --> 09:09.600
+ transferred the money between accounts. I was actually also not really interested in those
+
+09:09.600 --> 09:18.640
+ things. It was only when I went to university to study math that I found out that they had a
+
+09:18.640 --> 09:24.560
+ computer and students were allowed to use it. And there were some, you're supposed to talk to that
+
+09:24.560 --> 09:30.080
+ computer by programming it. What did that feel like? Yeah, that was the only thing you could do
+
+09:30.080 --> 09:36.560
+ with it. The computer wasn't really connected to the real world. The only thing you could do was
+
+09:36.560 --> 09:43.840
+ sort of, you typed your program on a bunch of punched cards. You gave the punched cards to
+
+09:43.840 --> 09:52.000
+ the operator. And an hour later, the operator gave you back your printout. And so all you could do
+
+09:52.000 --> 10:00.080
+ was write a program that did something very abstract. And I don't even remember what my
+
+10:00.080 --> 10:10.400
+ first forays into programming were, but they were sort of doing simple math exercises and just to
+
+10:10.400 --> 10:18.320
+ learn how a programming language worked. Did you sense, okay, first year of college, you see this
+
+10:18.320 --> 10:25.360
+ computer, you're able to have a program and it generates some output. Did you start seeing the
+
+10:25.360 --> 10:32.560
+ possibility of this? Or was it a continuation of the tinkering with circuits? Did you start to
+
+10:32.560 --> 10:39.040
+ imagine that one, the personal computer, but did you see it as something that is a tool
+
+10:39.040 --> 10:44.880
+ to get tools like a word processing tool, maybe maybe for gaming or something? Or did you start
+
+10:44.880 --> 10:50.320
+ to imagine that it could be, you know, going to the world of robotics, like you, you know, the
+
+10:50.320 --> 10:55.280
+ Frankenstein picture that you could create an artificial being. There's like another entity
+
+10:55.280 --> 11:02.720
+ in front of you. You did not see it. I don't think I really saw it that way. I was really more
+
+11:02.720 --> 11:09.440
+ interested in the tinkering. It's maybe not a sort of a complete coincidence that I ended up
+
+11:10.720 --> 11:17.120
+ sort of creating a programming language, which is a tool for other programmers. I've always been
+
+11:17.120 --> 11:24.560
+ very focused on the sort of activity of programming itself and not so much what happens with
+
+11:24.560 --> 11:34.800
+ what happens with the program you write. I do remember, and I don't remember, maybe in my second
+
+11:34.800 --> 11:42.240
+ or third year, probably my second, actually, someone pointed out to me that there was this
+
+11:42.240 --> 11:51.120
+ thing called Conway's Game of Life. You're probably familiar with it. I think in the 70s,
+
+11:51.120 --> 11:55.760
+ I think, as long as you came up with it. So there was a Scientific American column by
+
+11:57.680 --> 12:04.880
+ someone who did a monthly column about mathematical diversions and also blinking out on the guy's
+
+12:04.880 --> 12:11.360
+ name. It was very famous at the time and I think up to the 90s or so. And one of his columns was
+
+12:11.360 --> 12:16.160
+ about Conway's Game of Life and he had some illustrations and he wrote down all the rules
+
+12:16.160 --> 12:23.040
+ and sort of there was the suggestion that this was philosophically interesting, that that was why
+
+12:23.040 --> 12:30.400
+ Conway had called it that. And all I had was like the two pages photocopy of that article.
+
+12:31.040 --> 12:37.520
+ I don't even remember where I got it. But it spoke to me and I remember implementing
+
+12:37.520 --> 12:48.800
+ a version of that game for the batch computer we were using where I had a whole Pascal program
+
+12:48.800 --> 12:55.200
+ that sort of read an initial situation from input and read some numbers that said,
+
+12:55.760 --> 13:03.120
+ do so many generations and print every so many generations and then out would come pages and
+
+13:03.120 --> 13:12.560
+ pages of sort of things. Patterns of different kinds and yeah. Yeah. And I remember much later
+
+13:13.120 --> 13:19.520
+ I've done a similar thing using Python, but I sort of that original version I wrote at the time
+
+13:20.560 --> 13:28.960
+ I found interesting because I combined it with some trick I had learned during my electronics
+
+13:28.960 --> 13:38.880
+ hobbyist times. I essentially first on paper I designed a simple circuit built out of logic gates
+
+13:39.840 --> 13:46.320
+ that took nine bits of input, which is the sort of the cell and its neighbors
+
+13:47.680 --> 13:55.360
+ and produce the new value for that cell. And it's like a combination of a half adder and some
+
+13:55.360 --> 14:02.400
+ other clipping. No, it's actually a full adder. And so I had worked that out and then I translated
+
+14:02.400 --> 14:12.400
+ that into a series of Boolean operations on Pascal integers where you could use the integers as
+
+14:12.400 --> 14:27.920
+ bitwise values. And so I could basically generate 60 bits of a generation in like eight instructions
+
+14:27.920 --> 14:35.360
+ or so. Nice. So I was proud of that. It's funny that you mentioned so for people who don't know
+
+14:35.360 --> 14:42.160
+ Conway's Game of Life is a cellular automata where there's single compute units that kind of
+
+14:42.160 --> 14:49.520
+ look at their neighbors and figure out what they look like in the next generation based on the
+
+14:49.520 --> 14:57.040
+ state of their neighbors and this is deeply distributed system in concept at least. And then
+
+14:57.040 --> 15:04.400
+ there's simple rules that all of them follow and somehow out of the simple rule when you step back
+
+15:04.400 --> 15:13.120
+ and look at what occurs, it's beautiful. There's an emergent complexity, even though the underlying
+
+15:13.120 --> 15:17.600
+ rules are simple, there's an emergent complexity. Now the funny thing is you've implemented this
+
+15:17.600 --> 15:24.480
+ and the thing you're commenting on is you're proud of a hack you did to make it run efficiently.
+
+15:25.280 --> 15:29.360
+ When you're not commenting on what like this is a beautiful implementation.
+
+15:29.360 --> 15:33.680
+ You're not commenting on the fact that there's an emergent complexity
+
+15:34.480 --> 15:40.240
+ that you've quoted a simple program and when you step back and you print out the
+
+15:40.240 --> 15:45.360
+ following generation after generation, that's stuff that you may have not predicted what
+
+15:45.360 --> 15:52.480
+ happened is happening. And is that magic? I mean that's the magic that all of us feel when we
+
+15:52.480 --> 15:58.880
+ program. When you create a program and then you run it and whether it's Hello World or it shows
+
+15:58.880 --> 16:03.200
+ something on screen if there's a graphical component, are you seeing the magic and the
+
+16:03.200 --> 16:11.120
+ mechanism of creating that? I think I went back and forth. As a student, we had an incredibly
+
+16:11.120 --> 16:19.120
+ small budget of computer time that we could use. It was actually measured. I once got in trouble with
+
+16:19.120 --> 16:24.560
+ one of my professors because I had overspent the department's budget. It's a different story.
+
+16:24.560 --> 16:35.520
+ But so I actually wanted the efficient implementation because I also wanted to explore
+
+16:36.400 --> 16:44.080
+ what would happen with a larger number of generations and a larger sort of size of the
+
+16:44.080 --> 16:55.280
+ board. And so once the implementation was flawless, I would feed it different patterns and then I
+
+16:55.280 --> 17:01.520
+ think maybe there was a follow up article where there were patterns that were like gliders,
+
+17:02.400 --> 17:12.320
+ patterns that repeated themselves after a number of generations but translated one or two positions
+
+17:12.320 --> 17:19.840
+ to the right or up or something like that. And there were, I remember things like glider guns.
+
+17:19.840 --> 17:28.240
+ Well, you can Google Conway's Game of Life. People still go on over it. For a reason because
+
+17:28.240 --> 17:33.760
+ it's not really well understood. I mean this is what Stephen Wolfram is obsessed about.
+
+17:37.440 --> 17:41.520
+ We don't have the mathematical tools to describe the kind of complexity that emerges
+
+17:41.520 --> 17:45.120
+ in these kinds of systems. And the only way you can do is to run it.
+
+17:46.960 --> 17:56.160
+ I'm not convinced that it's sort of a problem that lends itself to classic mathematical analysis.
+
+17:56.720 --> 18:04.560
+ No. And so one theory of how you create an artificial intelligence or an artificial being
+
+18:04.560 --> 18:08.960
+ is you kind of have to, same with the game of life, you kind of have to create a universe
+
+18:08.960 --> 18:16.400
+ and let it run. That creating it from scratch in a design way in the, you know, coding up a
+
+18:16.400 --> 18:22.080
+ Python program that creates a fully intelligent system may be quite challenging that you might
+
+18:22.080 --> 18:28.640
+ need to create a universe just like the game of life is. Well, you might have to experiment with
+
+18:28.640 --> 18:36.560
+ a lot of different universes before. There is a set of rules that doesn't essentially always just
+
+18:36.560 --> 18:46.320
+ and repeating itself in a trivial way. Yeah. And Steve Wolfram, Stephen Wolfram works with
+
+18:46.320 --> 18:51.520
+ these simple rules, says that it's kind of surprising how quickly you find rules that
+
+18:51.520 --> 18:58.240
+ create interesting things. You shouldn't be able to, but somehow you do. And so maybe our universe
+
+18:58.240 --> 19:03.440
+ is laden with rules that will create interesting things that might not look like humans, but
+
+19:03.440 --> 19:08.640
+ you know, emergent phenomena that's interesting may not be as difficult to create as we think.
+
+19:08.640 --> 19:15.120
+ Sure. But let me sort of ask, at that time, you know, some of the world, at least in popular press,
+
+19:17.120 --> 19:23.360
+ was kind of captivated, perhaps at least in America, by the idea of artificial intelligence,
+
+19:24.000 --> 19:31.520
+ that these computers would be able to think pretty soon. And did that touch you at all? Did
+
+19:31.520 --> 19:40.560
+ that in science fiction or in reality, in any way? I didn't really start reading science fiction
+
+19:40.560 --> 19:52.560
+ until much, much later. I think as a teenager, I read maybe one bundle of science fiction stories.
+
+19:54.160 --> 19:56.960
+ Was it in the background somewhere, like in your thoughts?
+
+19:56.960 --> 20:04.160
+ That sort of the using computers to build something intelligent always fell to me,
+
+20:04.160 --> 20:10.320
+ because I felt I had so much understanding of what actually goes on inside a computer.
+
+20:11.600 --> 20:19.280
+ I knew how many bits of memory it had and how difficult it was to program and sort of
+
+20:19.280 --> 20:29.440
+ I didn't believe at all that that you could just build something intelligent out of that,
+
+20:29.440 --> 20:38.080
+ that that would really sort of satisfy my definition of intelligence. I think the most
+
+20:38.080 --> 20:44.960
+ the most influential thing that I read in my early 20s was Gödel Escherbach.
+
+20:44.960 --> 20:51.760
+ That was about consciousness and that was a big eye opener, in some sense.
+
+20:53.600 --> 21:00.560
+ In what sense? So on your own brain, did you at the time or do you now see your
+
+21:00.560 --> 21:06.880
+ own brain as a computer? Or is there a total separation of the way? So yeah, you're very
+
+21:06.880 --> 21:13.840
+ pragmatically, practically know the limits of memory, the limits of this sequential computing,
+
+21:13.840 --> 21:19.120
+ or weekly paralyzed computing, and you just know what we have now and it's hard to see
+
+21:19.120 --> 21:26.160
+ how it creates, but it's also easy to see it was in the in the 40s, 50s, 60s, and now
+
+21:27.120 --> 21:32.480
+ at least similarities between the brain and our computers. Oh yeah, I mean, I
+
+21:32.480 --> 21:44.400
+ I totally believe that brains are computers in some sense. I mean, the rules they they use to
+
+21:44.400 --> 21:51.680
+ play by are pretty different from the rules we we can sort of implement in in our current
+
+21:51.680 --> 22:04.160
+ hardware. But I don't believe in like a separate thing that infuses us with intelligence or
+
+22:06.160 --> 22:10.880
+ consciousness or any of that. There's no soul. I've been an atheist probably
+
+22:11.840 --> 22:18.720
+ from when I was 10 years old, just by thinking a bit about math and the universe.
+
+22:18.720 --> 22:26.640
+ And well, my parents were atheists. Now, I know that you you you could be an atheist and still
+
+22:26.640 --> 22:34.720
+ believe that there is something sort of about intelligence or consciousness that cannot possibly
+
+22:34.720 --> 22:42.560
+ emerge from a fixed set of rules. I am not in that camp. I totally see that
+
+22:42.560 --> 22:53.840
+ that sort of given how many millions of years evolution took its time. DNA is is a particular
+
+22:53.840 --> 23:04.560
+ machine that that sort of encodes information and an unlimited amount of information in in
+
+23:04.560 --> 23:14.000
+ chemical form and has figured out a way to replicate itself. I thought that that was maybe
+
+23:14.000 --> 23:19.520
+ it's 300 million years ago, but I thought it was closer to half a half a billion years ago that that's
+
+23:20.480 --> 23:27.200
+ sort of originated and it hasn't really changed that the sort of the structure of DNA hasn't
+
+23:27.200 --> 23:35.520
+ changed ever since that is like our binary code that we have in hardware. I mean, the basic
+
+23:35.520 --> 23:43.360
+ programming language hasn't changed, but maybe the programming itself, obviously did sort of it.
+
+23:43.360 --> 23:49.120
+ It happened to be a set of rules that was good enough to to sort of develop
+
+23:49.120 --> 23:58.560
+ of endless variability and and sort of the the idea of self replicating molecules
+
+23:59.440 --> 24:05.680
+ competing with each other for resources and and one type eventually sort of always taking over
+
+24:07.120 --> 24:12.560
+ that happened before there were any fossils. So we don't know how that exactly happened, but
+
+24:12.560 --> 24:21.440
+ I believe it it's it's clear that that did happen and can you comment on consciousness and how you
+
+24:22.320 --> 24:27.760
+ see it? Because I think we'll talk about programming quite a bit. We'll talk about,
+
+24:27.760 --> 24:33.600
+ you know, intelligence connecting to programming fundamentally, but consciousness consciousness
+
+24:33.600 --> 24:39.680
+ is this whole other other thing. Do you think about it often as a developer of a programming
+
+24:39.680 --> 24:48.000
+ language and and as a human? Those those are pretty sort of separate topics.
+
+24:49.440 --> 24:58.800
+ Sort of my line of work working with programming does not involve anything that that goes in the
+
+24:58.800 --> 25:06.320
+ direction of developing intelligence or consciousness, but sort of privately as an avid reader of
+
+25:06.320 --> 25:16.880
+ popular science writing. I have some thoughts which which is mostly that
+
+25:18.400 --> 25:27.840
+ I don't actually believe that consciousness is an all or nothing thing. I have a feeling that and
+
+25:27.840 --> 25:37.840
+ and I forget what I read that influenced this, but I feel that if you look at a cat or a dog or a
+
+25:37.840 --> 25:47.280
+ mouse, they have some form of intelligence. If you look at a fish, it has some form of intelligence
+
+25:47.280 --> 25:56.560
+ and that evolution just took a long time. But I feel that the the sort of evolution of
+
+25:58.240 --> 26:02.880
+ more and more intelligence that led to to sort of the human form of intelligence
+
+26:04.160 --> 26:12.880
+ follow the the evolution of the senses, especially the visual sense.
+
+26:12.880 --> 26:21.120
+ I mean, there is an enormous amount of processing that's needed to interpret a scene. And humans are
+
+26:21.120 --> 26:28.480
+ still better at that than than computers are. Yeah, and so and and I have a feeling that
+
+26:29.680 --> 26:41.680
+ there is a sort of the reason that that like mammals is in particular developed the levels of
+
+26:41.680 --> 26:49.280
+ consciousness that they have and that eventually sort of going from intelligence to to self
+
+26:49.280 --> 26:56.880
+ awareness and consciousness has to do with sort of being a robot that has very highly developed
+
+26:56.880 --> 27:03.840
+ senses. Has a lot of rich sensory information coming in. So the that's a really interesting
+
+27:03.840 --> 27:12.240
+ thought that that whatever that basic mechanism of DNA, whatever that basic building blocks of
+
+27:12.240 --> 27:19.840
+ programming is you if you just add more abilities, more more high resolution sensors, more sensors,
+
+27:20.400 --> 27:25.760
+ you just keep stacking those things on top that this basic programming in trying to survive
+
+27:25.760 --> 27:31.360
+ develops very interesting things that start to us humans to appear like intelligence and
+
+27:31.360 --> 27:39.200
+ consciousness. Yeah, so in in as far as robots go, I think that the self driving cars have that sort
+
+27:39.200 --> 27:49.520
+ of the greatest opportunity of developing something like that because when I drive myself, I don't
+
+27:49.520 --> 27:57.040
+ just pay attention to the rules of the road. I also look around and I get clues from that Oh,
+
+27:57.040 --> 28:04.800
+ this is a shopping district. Oh, here's an old lady crossing the street. Oh, here is someone
+
+28:04.800 --> 28:12.400
+ carrying a pile of mail. There's a mailbox. I bet you they're gonna cross the street to reach
+
+28:12.400 --> 28:18.640
+ that mailbox. And I slow down. And I don't even think about that. Yeah. And so there is there is
+
+28:18.640 --> 28:27.760
+ so much where you turn your observations into an understanding of what other consciousnesses
+
+28:28.480 --> 28:34.960
+ are going to do or what what other systems in the world are going to be Oh, that tree is going to
+
+28:34.960 --> 28:46.320
+ fall. Yeah, I see sort of I see much more of I expect somehow that if anything is going to
+
+28:46.320 --> 28:52.640
+ become conscious, it's going to be the self driving car and not the network of a bazillion
+
+28:54.080 --> 29:00.240
+ computers at in a Google or Amazon data center that are all networked together to
+
+29:02.080 --> 29:08.320
+ to do whatever they do. So in that sense, so you actually highlight because that's what I work in
+
+29:08.320 --> 29:14.480
+ is an autonomous vehicles, you highlight the big gap between what we currently can't do and
+
+29:14.480 --> 29:20.400
+ what we truly need to be able to do to solve the problem. Under that formulation, then consciousness
+
+29:20.400 --> 29:27.280
+ and intelligence is something that basically a system should have in order to interact with us
+
+29:27.280 --> 29:35.440
+ humans, as opposed to some kind of abstract notion of a consciousness consciousness is
+
+29:35.440 --> 29:39.200
+ something that you need to have to be able to empathize to be able to
+
+29:39.200 --> 29:46.960
+ to fear the understand what the fear of death is. All these aspects that are important for
+
+29:46.960 --> 29:54.080
+ interacting with pedestrians need to be able to do basic computation based on our human
+
+29:55.600 --> 30:02.080
+ desires and if you sort of Yeah, if you if you look at the dog, the dog clearly knows, I mean,
+
+30:02.080 --> 30:06.320
+ I'm not the dog owner, my brother, I have friends who have dogs, the dogs clearly know
+
+30:06.320 --> 30:11.440
+ what the humans around them are going to do or at least they have a model of what those humans
+
+30:11.440 --> 30:17.360
+ are going to do when they learn the dog some dogs know when you're going out and they want to go
+
+30:17.360 --> 30:23.920
+ out with you, they're sad when you leave them alone, they cry. They're afraid because they were
+
+30:24.480 --> 30:35.680
+ mistreated when they were younger. We don't assign sort of consciousness to dogs or at least
+
+30:35.680 --> 30:43.520
+ not not all that much but I also don't think they have none of that. So I think it's it's
+
+30:45.280 --> 30:48.960
+ consciousness and intelligence are not all or nothing.
+
+30:50.160 --> 30:54.320
+ The spectrum is really interesting. But in returning to
+
+30:56.000 --> 31:00.560
+ programming languages and the way we think about building these kinds of things about building
+
+31:00.560 --> 31:05.440
+ intelligence, building consciousness, building artificial beings. So I think one of the exciting
+
+31:05.440 --> 31:13.360
+ ideas came in the 17th century. And with liveness, Hobbes, Descartes, where there's this feeling that
+
+31:13.360 --> 31:23.120
+ you can convert all thought all reasoning, all the thing that we find very special in our brains,
+
+31:23.120 --> 31:28.800
+ you can convert all of that into logic. You can formalize it, former reasoning. And then once
+
+31:28.800 --> 31:33.280
+ you formalize everything, all of knowledge, then you can just calculate. And that's what
+
+31:33.280 --> 31:39.120
+ we're doing with our brains is we're calculating. So there's this whole idea that we that this is
+
+31:39.120 --> 31:45.920
+ possible that this but they weren't aware of the concept of pattern matching in the sense that we
+
+31:45.920 --> 31:53.840
+ are aware of it now. They sort of thought you they had discovered incredible bits of mathematics
+
+31:53.840 --> 32:04.720
+ like Newton's calculus. And their sort of idealism there, their sort of extension of what they could
+
+32:04.720 --> 32:16.480
+ do with logic and math sort of went along those lines. And they thought there's there's like,
+
+32:16.480 --> 32:23.920
+ yeah, logic, there's there's like a bunch of rules, and a bunch of input, they didn't realize that how
+
+32:23.920 --> 32:33.520
+ you recognize a face is not just a bunch of rules, but is a shit ton of data, plus a circuit that
+
+32:34.560 --> 32:42.800
+ that sort of interprets the visual clues and the context and everything else. And somehow
+
+32:42.800 --> 32:53.120
+ how can massively parallel pattern match against stored rules? I mean, if I see you tomorrow here
+
+32:53.120 --> 32:58.320
+ in front of the Dropbox office, I might recognize you even if I'm wearing a different shirt. Yeah,
+
+32:58.320 --> 33:04.240
+ but if I if I see you tomorrow in a coffee shop in Belmont, I might have no idea that it was you
+
+33:04.240 --> 33:11.920
+ or on the beach or whatever. I make those mistakes myself all the time. I see someone that I only
+
+33:11.920 --> 33:17.840
+ know as like, Oh, this person is a colleague of my wife's. And then I see them at the movies and
+
+33:18.640 --> 33:26.400
+ I don't recognize them. But do you see those you call it pattern matching? Do you see that rules is
+
+33:28.880 --> 33:34.880
+ unable to encode that to you? Everything you see all the piece of information you look around
+
+33:34.880 --> 33:39.520
+ this room, I'm wearing a black shirt, I have a certain height, I'm a human all these you can
+
+33:39.520 --> 33:45.440
+ there's probably tens of thousands of facts you pick up moment by moment about this scene,
+
+33:45.440 --> 33:49.760
+ you take them for granted and you accumulate aggregate them together to understand the scene.
+
+33:49.760 --> 33:53.760
+ You don't think all of that could be encoded to weren't at the end of the day, you just put
+
+33:53.760 --> 34:02.160
+ it on the table and calculate. Oh, I don't know what that means. I mean, yes, in the sense that
+
+34:02.160 --> 34:10.880
+ there is no, there is no actual magic there, but there are enough layers of abstraction from sort
+
+34:10.880 --> 34:19.440
+ of from the facts as they enter my eyes and my ears to the understanding of the scene that I don't
+
+34:19.440 --> 34:31.200
+ think that that AI has really covered enough of that distance. It's like if you take a human body
+
+34:31.200 --> 34:40.960
+ and you realize it's built out of atoms, well, that that is a uselessly reductionist view, right?
+
+34:41.760 --> 34:46.640
+ The body is built out of organs, the organs are built out of cells, the cells are built out of
+
+34:46.640 --> 34:54.240
+ proteins, the proteins are built out of amino acids, the amino acids are built out of atoms,
+
+34:54.240 --> 35:00.800
+ and then you get to quantum mechanics. So that's a very pragmatic view. I mean, obviously as an
+
+35:00.800 --> 35:06.720
+ engineer, I agree with that kind of view, but I also you also have to consider the the with the
+
+35:06.720 --> 35:13.120
+ Sam Harris view of well, well, intelligence is just information processing. Do you just like
+
+35:13.120 --> 35:17.840
+ you said you take in sensory information, you do some stuff with it and you come up with actions
+
+35:17.840 --> 35:25.680
+ that are intelligent. That makes it sound so easy. I don't know who Sam Harris is. Oh, it's
+
+35:25.680 --> 35:30.240
+ philosopher. So like this is how philosophers often think, right? And essentially, that's what
+
+35:30.240 --> 35:37.040
+ Descartes was is, wait a minute, if there is, like you said, no magic. So you basically says it
+
+35:37.040 --> 35:43.280
+ doesn't appear like there's any magic, but we know so little about it that it might as well be magic.
+
+35:43.280 --> 35:48.240
+ So just because we know that we're made of atoms, just because we know we're made of organs,
+
+35:48.240 --> 35:54.320
+ the fact that we know very little how to get from the atoms to organs in a way that's recreatable
+
+35:54.320 --> 36:01.280
+ means it that you shouldn't get too excited just yet about the fact that you figured out that we're
+
+36:01.280 --> 36:09.600
+ made of atoms. Right. And and and the same about taking facts as our our sensory organs take them
+
+36:09.600 --> 36:19.440
+ in and turning that into reasons and actions that sort of there are a lot of abstractions that we
+
+36:19.440 --> 36:30.000
+ haven't quite figured out how to how to deal with those. I mean, I sometimes I don't know if I can
+
+36:30.000 --> 36:38.880
+ go on a tangent or not. Please. Drag you back in. Sure. So if I take a simple program that parses,
+
+36:40.880 --> 36:47.760
+ say I have a compiler, it parses a program. In a sense, the input routine of that compiler
+
+36:48.320 --> 36:57.120
+ of that parser is a sense, a sensing organ. And it builds up a mighty complicated internal
+
+36:57.120 --> 37:03.920
+ representation of the program it just saw it doesn't just have a linear sequence of bytes
+
+37:03.920 --> 37:10.640
+ representing the text of the program anymore, it has an abstract syntax tree. And I don't know how
+
+37:10.640 --> 37:18.800
+ many of your viewers or listeners are familiar with compiler technology, but there is fewer and
+
+37:18.800 --> 37:26.720
+ fewer these days, right? That's also true, probably. People want to take a shortcut, but there's sort
+
+37:26.720 --> 37:35.200
+ of this abstraction is a data structure that the compiler then uses to produce outputs that is
+
+37:35.200 --> 37:41.280
+ relevant like a translation of that program to machine code that can be executed by by hardware.
+
+37:45.360 --> 37:53.360
+ And then that data structure gets thrown away. When a fish or a fly sees
+
+37:53.360 --> 38:03.920
+ these sort of gets visual impulses. I'm sure it also builds up some data structure and for
+
+38:03.920 --> 38:11.920
+ the fly that may be very minimal, a fly may may have only a few. I mean, in the case of a fly's
+
+38:11.920 --> 38:20.720
+ brain, I could imagine that there are few enough layers of abstraction that it's not much more
+
+38:20.720 --> 38:28.000
+ than when it's darker here than it is here. Well, it can sense motion, because a fly sort of responds
+
+38:28.000 --> 38:35.440
+ when you move your arm towards it. So clearly, it's visual processing is intelligent, or well,
+
+38:35.440 --> 38:43.600
+ not intelligent, but is has an abstraction for motion. And we still have similar things in in
+
+38:43.600 --> 38:48.880
+ but much more complicated in our brains. I mean, otherwise, you couldn't drive a car if you,
+
+38:48.880 --> 38:52.880
+ you couldn't sort if you didn't have an incredibly good abstraction for motion.
+
+38:54.560 --> 39:00.160
+ Yeah, in some sense, the same abstraction for motion is probably one of the primary sources of
+
+39:00.160 --> 39:06.160
+ our of information for us, we just know what to do. I think we know what to do with that.
+
+39:06.160 --> 39:11.200
+ We've built up other abstractions on top. We build much more complicated data structures
+
+39:11.200 --> 39:17.280
+ based on that. And we build more persistent data structures, sort of after some processing,
+
+39:17.280 --> 39:24.240
+ some information sort of gets stored in our memory, pretty much permanently, and is available on
+
+39:24.240 --> 39:31.680
+ recall. I mean, there are some things that you sort of, you're conscious that you're remembering it,
+
+39:31.680 --> 39:37.840
+ like you give me your phone number, I, well, at my age, I have to write it down, but I could
+
+39:37.840 --> 39:44.480
+ imagine I could remember those seven numbers or 10, 10 digits, and reproduce them in a while.
+
+39:44.480 --> 39:53.120
+ If I sort of repeat them to myself a few times. So that's a fairly conscious form of memorization.
+
+39:53.120 --> 40:00.880
+ On the other hand, how do I recognize your face? I have no idea. My brain has a whole bunch of
+
+40:00.880 --> 40:07.120
+ specialized hardware that knows how to recognize faces. I don't know how much of that is sort of
+
+40:07.120 --> 40:15.200
+ coded in our DNA and how much of that is trained over and over between the ages of zero and three.
+
+40:16.240 --> 40:23.120
+ But somehow our brains know how to do lots of things like that that are useful in our interactions
+
+40:23.120 --> 40:30.080
+ with other humans without really being conscious of how it's done anymore.
+
+40:30.080 --> 40:35.920
+ Right. So our actual day to day lives, we're operating at the very highest level of abstraction.
+
+40:35.920 --> 40:40.720
+ We're just not even conscious of all the little details underlying it. There's compilers on top
+
+40:40.720 --> 40:45.040
+ of, it's like turtles on top of turtles or turtles all the way down. It's compilers all the way down.
+
+40:46.160 --> 40:52.800
+ But that's essentially, you say that there's no magic. That's what I, what I was trying to get at,
+
+40:52.800 --> 40:58.640
+ I think, is with Descartes started this whole train of saying that there's no magic. I mean,
+
+40:58.640 --> 41:03.280
+ there's others beforehand. Well, didn't Descartes also have the notion, though, that the soul and
+
+41:03.280 --> 41:09.520
+ the body were fundamentally separate? Yeah, I think he had to write in God in there for
+
+41:10.320 --> 41:16.480
+ political reasons. So I don't actually, I'm not historian, but there's notions in there that
+
+41:16.480 --> 41:22.720
+ all of reasoning, all of human thought can be formalized. I think that continued in the 20th
+
+41:22.720 --> 41:30.880
+ century with the Russell and with Gato's incompleteness theorem, this debate of what are
+
+41:30.880 --> 41:35.280
+ the limits of the things that could be formalized? That's where the Turing machine came along.
+
+41:35.280 --> 41:40.800
+ And this exciting idea, I mean, underlying a lot of computing, that you can do quite a lot
+
+41:40.800 --> 41:46.320
+ with a computer. You can, you can encode a lot of the stuff we're talking about in terms of
+
+41:46.320 --> 41:52.400
+ recognizing faces and so on, theoretically, in an algorithm that can then run on the computer.
+
+41:52.400 --> 42:01.120
+ And in that context, I'd like to ask programming in a philosophical way.
+
+42:02.960 --> 42:08.160
+ What, so what does it mean to program a computer? So you said you write a Python program
+
+42:08.880 --> 42:16.800
+ or compiled a C++ program that compiles to somebody code. It's forming layers.
+
+42:16.800 --> 42:22.400
+ You're, you're, you're programming in a layer of abstraction that's higher. How do you see programming
+
+42:22.960 --> 42:27.760
+ in that context? Can it keep getting higher and higher levels of abstraction?
+
+42:29.680 --> 42:35.120
+ I think at some, at some point, the higher level of levels of abstraction will not be called
+
+42:35.120 --> 42:44.800
+ programming and they will not resemble what we, we call programming at the moment. There will
+
+42:44.800 --> 42:53.600
+ not be source code. I mean, there will still be source code sort of at a lower level of the machine,
+
+42:53.600 --> 43:04.480
+ just like there's still molecules and electrons and sort of proteins in our brains. But, and so
+
+43:04.480 --> 43:11.520
+ there's still programming and system administration and who knows what keeping to keep the machine
+
+43:11.520 --> 43:17.600
+ running. But what the machine does is, is a different level of abstraction in a sense. And
+
+43:18.160 --> 43:25.120
+ as far as I understand the way that for last decade or more people have made progress with
+
+43:25.120 --> 43:32.000
+ things like facial recognition or the self driving cars is all by endless, endless amounts of
+
+43:32.000 --> 43:42.240
+ training data where at least as, as, as a layperson and I feel myself totally as a layperson in that
+
+43:42.240 --> 43:52.240
+ field, it looks like the researchers who publish the results don't necessarily know exactly how,
+
+43:52.240 --> 44:02.480
+ how their algorithms work. And I often get upset when I sort of read a sort of a fluff piece about
+
+44:02.480 --> 44:10.000
+ Facebook in the newspaper or social networks and they say, well, algorithms. And that's like a totally
+
+44:10.000 --> 44:18.640
+ different interpretation of the word algorithm. Because for me, the way I was trained or what I
+
+44:18.640 --> 44:25.200
+ learned when I was eight or 10 years old, an algorithm is a set of rules that you completely
+
+44:25.200 --> 44:31.600
+ understand that can be mathematically analyzed. And, and, and you can prove things, you can like
+
+44:31.600 --> 44:37.920
+ prove that Aristotle's Civ produces all prime numbers and only prime numbers.
+
+44:39.120 --> 44:45.360
+ Yeah. So the, I don't know if you know who Andre Capati is. I'm afraid not. So he's a
+
+44:45.360 --> 44:52.880
+ head of AI at Tesla now, but he was at Stanford before, and he has this cheeky way of calling
+
+44:53.520 --> 45:01.200
+ this concept software 2.0. So let me disentangle that for a second. So kind of what you're
+
+45:01.200 --> 45:06.560
+ referring to is the traditional, traditional, the algorithm, the concept of an algorithm,
+
+45:06.560 --> 45:10.320
+ something that's there, it's clear, you can read it, you understand it, you can prove it's
+
+45:10.320 --> 45:19.280
+ functioning as kind of software 1.0. And what software 2.0 is, is exactly what you describe,
+
+45:19.280 --> 45:24.560
+ which is you have neural networks, which is a type of machine learning that you feed a bunch
+
+45:24.560 --> 45:31.600
+ of data, and that neural network learns to do a function. All you specify is the inputs and
+
+45:31.600 --> 45:38.560
+ the outputs you want, and you can't look inside. You can't analyze it. All you can do is train
+
+45:38.560 --> 45:43.840
+ this function to map the inputs, the outputs by giving a lot of data. In that sense, programming
+
+45:43.840 --> 45:49.360
+ becomes getting a lot of cleaning, getting a lot of data. That's what programming is in this.
+
+45:49.360 --> 45:52.880
+ Well, that would be programming 2.0. 2.0 to programming 2.0.
+
+45:53.680 --> 45:57.680
+ I wouldn't call that programming. It's just a different activity, just like
+
+45:58.400 --> 46:01.600
+ building organs out of cells is not called chemistry.
+
+46:01.600 --> 46:10.560
+ Well, so let's just step back and think sort of more generally, of course, but it's like
+
+46:12.560 --> 46:20.000
+ as a parent teaching your kids, things can be called programming. In that same sense,
+
+46:20.000 --> 46:26.320
+ that's how programming is being used. You're providing them data, examples, use cases.
+
+46:26.320 --> 46:37.200
+ So imagine writing a function not with for loops and clearly readable text, but more saying,
+
+46:37.760 --> 46:43.840
+ well, here's a lot of examples of what this function should take, and here's a lot of
+
+46:43.840 --> 46:48.880
+ examples of when it takes those functions, it should do this, and then figure out the rest.
+
+46:48.880 --> 46:57.600
+ So that's the 2.0 concept. And the question I have for you is like, it's a very fuzzy way.
+
+46:58.320 --> 47:02.560
+ This is the reality of a lot of these pattern recognition systems and so on. It's a fuzzy way
+
+47:02.560 --> 47:09.520
+ of quote unquote programming. What do you think about this kind of world? Should it be called
+
+47:09.520 --> 47:17.840
+ something totally different than programming? If you're a software engineer, does that mean
+
+47:17.840 --> 47:24.800
+ you're designing systems that are very can be systematically tested, evaluated, they have a
+
+47:24.800 --> 47:31.440
+ very specific specification, and then this other fuzzy software 2.0 world machine learning world,
+
+47:31.440 --> 47:35.920
+ that's that's something else totally? Or is there some intermixing that's possible?
+
+47:35.920 --> 47:47.200
+ Well, the question is probably only being asked because we we don't quite know what
+
+47:47.200 --> 47:57.920
+ that software 2.0 actually is. And it sort of I think there is a truism that every task that
+
+47:58.960 --> 48:05.440
+ AI has has tackled in the past. At some point, we realized how it was done. And then it was no
+
+48:05.440 --> 48:14.160
+ longer considered part of artificial intelligence because it was no longer necessary to to use
+
+48:14.160 --> 48:25.120
+ that term. It was just, oh, now we know how to do this. And a new field of science or engineering
+
+48:25.120 --> 48:36.320
+ has been developed. And I don't know if sort of every form of learning or sort of controlling
+
+48:36.320 --> 48:41.920
+ computer systems should always be called programming. So I don't know, maybe I'm focused too much on
+
+48:41.920 --> 48:53.120
+ the terminology. I but I expect that that there just will be different concepts where people with
+
+48:53.120 --> 49:05.600
+ sort of different education and a different model of what they're trying to do will will develop those
+
+49:05.600 --> 49:14.800
+ concepts. Yeah, and I guess, if you could comment on another way to put this concept is, I think,
+
+49:16.480 --> 49:22.720
+ I think the kind of functions that neural networks provide is things as opposed to being able to
+
+49:22.720 --> 49:29.760
+ upfront prove that this should work for all cases you throw at it. All you're able, it's the worst
+
+49:29.760 --> 49:36.400
+ case analysis versus average case analysis, all you're able to say is it seems on everything
+
+49:36.400 --> 49:43.840
+ we've tested to work 99.9% of the time, but we can't guarantee it and it fails in unexpected ways.
+
+49:43.840 --> 49:48.400
+ We can even give you examples of how it fails in unexpected ways. But it's like really good
+
+49:48.400 --> 49:55.200
+ most of the time. Yeah, but there's no room for that in current ways we think about programming.
+
+50:00.160 --> 50:03.600
+ Programming 1.0 is actually sort of
+
+50:06.160 --> 50:14.080
+ getting to that point to where the sort of the ideal of a bug free program
+
+50:14.080 --> 50:25.440
+ has been abandoned long ago by most software developers. We only care about bugs that manifest
+
+50:25.440 --> 50:33.840
+ themselves often enough to be annoying. And we're willing to take the occasional crash or
+
+50:33.840 --> 50:45.760
+ outage or incorrect result for granted, because we can't possibly we don't have enough programmers
+
+50:45.760 --> 50:50.880
+ to make all the code bug free and it would be an incredibly tedious business. And if you try to
+
+50:50.880 --> 50:59.120
+ throw formal methods at it, it gets it becomes even more tedious. So every once in a while,
+
+50:59.120 --> 51:07.200
+ the user clicks on a link in and somehow they get an error. And the average user doesn't panic,
+
+51:07.200 --> 51:15.360
+ they just click again and see if it works better the second time, which often magically it does.
+
+51:16.320 --> 51:24.240
+ Or they go up and they try some other way of performing their tasks. So that's sort of an
+
+51:24.240 --> 51:33.440
+ end to end recovery mechanism and inside systems, there is all sorts of retries and timeouts and
+
+51:34.720 --> 51:41.520
+ fallbacks. And I imagine that that sort of biological systems are even more full of that
+
+51:41.520 --> 51:50.320
+ because otherwise they wouldn't survive. Do you think programming should be taught and thought of
+
+51:50.320 --> 51:59.440
+ as exactly what you just said before I come from is kind of you're you're always denying that fact
+
+51:59.440 --> 52:12.240
+ always in in sort of basic programming education, the sort of the programs you're, you're having
+
+52:12.240 --> 52:22.240
+ students write are so small and simple that if there is a bug, you can always find it and fix it.
+
+52:22.960 --> 52:29.520
+ Because the sort of programming as it's being taught in some even elementary middle schools
+
+52:29.520 --> 52:36.880
+ in high school, introduction to programming classes in college, typically, it's programming in the
+
+52:36.880 --> 52:45.280
+ small. Very few classes sort of actually teach software engineering building large systems. I
+
+52:45.280 --> 52:52.480
+ mean, every summer here at Dropbox, we have a large number of interns, every tech company
+
+52:53.440 --> 53:00.880
+ on the West Coast has the same thing. These interns are always amazed because this is the
+
+53:00.880 --> 53:09.120
+ first time in their life that they see what goes on in a really large software development environment.
+
+53:10.480 --> 53:20.320
+ And everything they've learned in college was almost always about a much smaller scale and
+
+53:20.320 --> 53:26.880
+ somehow that difference in scale makes a qualitative difference in how you how you
+
+53:26.880 --> 53:33.120
+ do things and how you think about it. If you then take a few steps back into decades,
+
+53:34.000 --> 53:39.040
+ 70s and 80s, when you're first thinking about Python or just that world of programming languages,
+
+53:39.840 --> 53:45.680
+ did you ever think that there would be systems as large as underlying Google, Facebook and Dropbox?
+
+53:46.480 --> 53:54.000
+ Did you when you were thinking about Python? I was actually always caught by surprise by
+
+53:54.000 --> 53:58.240
+ every sort of this. Yeah, pretty much every stage of computing.
+
+53:59.440 --> 54:06.800
+ So maybe just because you spoke in other interviews, but I think the evolution of
+
+54:06.800 --> 54:11.920
+ programming languages are fascinating. And it's especially because it leads from my
+
+54:11.920 --> 54:17.440
+ perspective towards greater and greater degrees of intelligence. I learned the first programming
+
+54:17.440 --> 54:26.720
+ language I played with in Russia was with the turtle logo logo. Yeah. And if you look, I just
+
+54:26.720 --> 54:31.520
+ have a list of programming languages, all of which I've played with a little bit. And they're all
+
+54:31.520 --> 54:38.320
+ beautiful in different ways from Fortran, Cobalt, Lisp, Algal 60, basic logo again, C
+
+54:38.320 --> 54:48.240
+ as a few object oriented came along in the 60s, Simula, Pascal, small talk, all of that leads
+
+54:48.240 --> 54:55.360
+ all the classics, the classics. Yeah, the classic hits, right? Scheme built that's built on top of
+
+54:55.360 --> 55:03.680
+ Lisp on the database side SQL C plus plus and all that leads up to Python, Pascal to
+
+55:03.680 --> 55:10.720
+ and all that's before Python, MATLAB, these kind of different communities, different languages.
+
+55:10.720 --> 55:17.040
+ So can you talk about that world? I know that sort of Python came out of ABC, which actually
+
+55:17.040 --> 55:22.720
+ never knew that language. I just having researched this conversation went back to ABC and it looks
+
+55:22.720 --> 55:29.680
+ remarkably, it has a lot of annoying qualities. But underneath those like all caps and so on.
+
+55:29.680 --> 55:34.880
+ But underneath that, there's elements of Python that are quite they're already there.
+
+55:35.440 --> 55:39.280
+ That's where I got all the good stuff, all the good stuff. So but in that world,
+
+55:39.280 --> 55:43.680
+ you're swimming in these programming languages, were you focused on just the good stuff in your
+
+55:43.680 --> 55:50.240
+ specific circle? Or did you have a sense of what, what is everyone chasing? You said that every
+
+55:50.240 --> 55:59.520
+ programming language is built to scratch an itch. Were you aware of all the itches in the community
+
+55:59.520 --> 56:04.880
+ and if not, or if yes, I mean, what itch we try to scratch with Python?
+
+56:05.600 --> 56:12.240
+ Well, I'm glad I wasn't aware of all the itches because I would probably not have been able to
+
+56:12.880 --> 56:17.040
+ do anything. I mean, if you're trying to solve every problem at once,
+
+56:18.000 --> 56:27.760
+ you saw nothing. Well, yeah, it's, it's too overwhelming. And so I had a very, very focused
+
+56:27.760 --> 56:36.480
+ problem. I wanted a programming language that set somewhere in between shell scripting and C.
+
+56:38.480 --> 56:48.480
+ And now, arguably, there is like, one is higher level, one is lower level. And
+
+56:49.680 --> 56:56.800
+ Python is sort of a language of an intermediate level, although it's still pretty much at the
+
+56:56.800 --> 57:10.160
+ high level. And I was I was thinking about much more about I want a tool that I can use to be
+
+57:10.160 --> 57:19.040
+ more productive as a programmer in a very specific environment. And I also had given myself a time
+
+57:19.040 --> 57:27.600
+ budget for the development of the tool. And that was sort of about three months for both the design
+
+57:27.600 --> 57:31.920
+ like thinking through what are all the features of the language syntactically.
+
+57:33.760 --> 57:42.080
+ And semantically, and how do I implement the whole pipeline from parsing the source code to
+
+57:42.080 --> 57:51.200
+ executing it. So I think both with the timeline and the goals, it seems like productivity was
+
+57:51.200 --> 57:59.520
+ at the core of it as a goal. So, like for me, in the 90s, and the first decade of the 21st
+
+57:59.520 --> 58:06.400
+ century, I was always doing machine learning AI, programming for my research was always in C++.
+
+58:06.400 --> 58:12.160
+ Wow. And then the other people who are a little more mechanical engineering,
+
+58:12.160 --> 58:18.640
+ electrical engineering, are Matlabby. They're a little bit more Matlab focused. Those are the
+
+58:18.640 --> 58:26.480
+ world and maybe a little bit Java too, but people who are more interested in emphasizing
+
+58:26.480 --> 58:33.440
+ the object oriented nature of things. So within the last 10 years or so, especially with the
+
+58:33.440 --> 58:38.400
+ oncoming of neural networks and these packages that are built on Python to interface with
+
+58:39.200 --> 58:45.760
+ neural networks, I switched to Python. And it's just, I've noticed a significant boost that I
+
+58:45.760 --> 58:50.400
+ can't exactly, because I don't think about it, but I can't exactly put into words why I'm just
+
+58:50.400 --> 58:57.520
+ much, much more productive, just being able to get the job done much, much faster. So how do you
+
+58:57.520 --> 59:02.640
+ think whatever that qualitative difference is, I don't know if it's quantitative, it could be just
+
+59:02.640 --> 59:08.000
+ a feeling. I don't know if I'm actually more productive, but how do you think about? You probably
+
+59:08.000 --> 59:14.480
+ are. Yeah, well, that's right. I think there's elements. Let me just speak to one aspect that
+
+59:14.480 --> 59:23.920
+ I think that was affecting our productivity is C++ was, I really enjoyed creating performant code
+
+59:24.800 --> 59:29.840
+ and creating a beautiful structure where everything that, you know, this kind of going
+
+59:29.840 --> 59:34.560
+ into this, especially with the newer and newer standards of templated programming of just really
+
+59:34.560 --> 59:41.280
+ creating this beautiful, formal structure that I found myself spending most of my time doing that
+
+59:41.280 --> 59:46.160
+ as opposed to getting it parsing a file and extracting a few keywords or whatever the task
+
+59:46.160 --> 59:51.680
+ goes trying to do. So what is it about Python? How do you think of productivity in general as
+
+59:51.680 --> 59:57.440
+ you were designing it now? So through the decades, last three decades, what do you think it means
+
+59:57.440 --> 1:00:01.920
+ to be a productive programmer? And how did you try to design it into the language?
+
+1:00:03.200 --> 1:00:10.240
+ There are different tasks. And as a programmer, it's, it's useful to have different tools available
+
+1:00:10.240 --> 1:00:17.680
+ that sort of are suitable for different tasks. So I still write C code. I still write shell code.
+
+1:00:18.720 --> 1:00:22.080
+ But I write most of my, my things in Python.
+
+1:00:22.080 --> 1:00:30.960
+ Why do I still use those other languages? Because sometimes the task just demands it.
+
+1:00:32.400 --> 1:00:38.880
+ And, well, I would say most of the time, the task actually demands a certain language because
+
+1:00:38.880 --> 1:00:44.640
+ the task is not write a program that solves problem x from scratch, but it's more like
+
+1:00:44.640 --> 1:00:52.400
+ fix a bug in existing program x or add a small feature to an existing large program.
+
+1:00:56.320 --> 1:01:04.560
+ But even if, if you sort of, if you're not constrained in your choice of language
+
+1:01:04.560 --> 1:01:15.360
+ by context like that, there is still the fact that if you write it in a certain language, then you
+
+1:01:15.360 --> 1:01:26.080
+ sort of, you, you have this balance between how long does it time? Does it take you to write the
+
+1:01:26.080 --> 1:01:38.560
+ code? And how long does the code run? And when you're in sort of, in the phase of exploring
+
+1:01:38.560 --> 1:01:45.760
+ solutions, you often spend much more time writing the code than running it, because every time
+
+1:01:46.560 --> 1:01:52.880
+ you've sort of, you've run it, you see that the output is not quite what you wanted. And
+
+1:01:52.880 --> 1:02:05.520
+ you spend some more time coding. And a language like Python just makes that iteration much faster,
+
+1:02:05.520 --> 1:02:13.600
+ because there are fewer details. There is a large library, sort of there are fewer details that,
+
+1:02:13.600 --> 1:02:20.480
+ that you have to get right before your program compiles and runs. There are libraries that
+
+1:02:20.480 --> 1:02:27.040
+ do all sorts of stuff for you. So you can sort of very quickly take a bunch of
+
+1:02:28.000 --> 1:02:36.560
+ existing components, put them together and get your prototype application running just like
+
+1:02:37.120 --> 1:02:44.640
+ when I was building electronics, I was using a breadboard most of the time. So I had this like
+
+1:02:44.640 --> 1:02:52.240
+ sprawl out circuit that if you shook it, it would stop working because it was not put together
+
+1:02:52.800 --> 1:03:00.160
+ very well. But it functioned and all I wanted was to see that it worked and then move on to the next
+
+1:03:01.040 --> 1:03:07.280
+ next schematic or design or add something to it. Once you've sort of figured out, oh, this is the
+
+1:03:07.280 --> 1:03:13.920
+ perfect design for my radio or light sensor or whatever, then you can say, okay, how do we
+
+1:03:13.920 --> 1:03:20.560
+ design a PCB for this? How do we solder the components in a small space? How do we make it
+
+1:03:20.560 --> 1:03:32.800
+ so that it is robust against, say, voltage fluctuations or mechanical disruption? I mean,
+
+1:03:32.800 --> 1:03:37.280
+ I know nothing about that when it comes to designing electronics, but I know a lot about
+
+1:03:37.280 --> 1:03:45.200
+ that when it comes to writing code. So the initial steps are efficient, fast, and there's not much
+
+1:03:45.200 --> 1:03:54.080
+ stuff that gets in the way. But you're kind of describing from like Darwin described the evolution
+
+1:03:54.080 --> 1:04:01.920
+ of species, right? You're observing of what is true about Python. Now, if you take a step back,
+
+1:04:01.920 --> 1:04:09.680
+ if the act of creating languages is art, and you had three months to do it, initial steps,
+
+1:04:12.480 --> 1:04:17.040
+ so you just specified a bunch of goals, sort of things that you observe about Python. Perhaps
+
+1:04:17.040 --> 1:04:23.440
+ you had those goals, but how do you create the rules, the syntactic structure, the features
+
+1:04:23.440 --> 1:04:29.200
+ that result in those? So I have, in the beginning, and I have follow up questions about through the
+
+1:04:29.200 --> 1:04:35.440
+ evolution of Python, too. But in the very beginning, when you're sitting there, creating the lexical
+
+1:04:35.440 --> 1:04:45.680
+ analyze or whatever evolution was still a big part of it, because I sort of I said to myself,
+
+1:04:46.480 --> 1:04:52.800
+ I don't want to have to design everything from scratch. I'm going to borrow features from
+
+1:04:52.800 --> 1:04:57.440
+ other languages that I like. Oh, interesting. So you basically, exactly, you first observe what
+
+1:04:57.440 --> 1:05:04.800
+ you like. Yeah. And so that's why if you're 17 years old, and you want to sort of create a programming
+
+1:05:04.800 --> 1:05:11.840
+ language, you're not going to be very successful at it. Because you have no experience with other
+
+1:05:11.840 --> 1:05:25.280
+ languages. Whereas I was in my, let's say mid 30s. I had written parsers before. So I had worked on
+
+1:05:25.280 --> 1:05:32.000
+ the implementation of ABC, I had spent years debating the design of ABC with its authors,
+
+1:05:32.000 --> 1:05:36.560
+ it's with its designers, I had nothing to do with the design, it was designed
+
+1:05:37.600 --> 1:05:42.400
+ fully as it was ended up being implemented when I joined the team. But so
+
+1:05:44.480 --> 1:05:52.000
+ you borrow ideas and concepts and very concrete sort of local rules from different languages,
+
+1:05:52.000 --> 1:06:00.240
+ like the indentation and certain other syntactic features from ABC. But I chose to borrow string
+
+1:06:00.240 --> 1:06:10.400
+ literals and how numbers work from C and various other things. So in then, if you take that further,
+
+1:06:10.400 --> 1:06:17.280
+ so yet, you've had this funny sounding, but I think surprisingly accurate and at least practical
+
+1:06:17.280 --> 1:06:23.280
+ title of benevolent dictator for life for quite, you know, for the last three decades or whatever,
+
+1:06:23.280 --> 1:06:30.800
+ or no, not the actual title, but functionally speaking. So you had to make decisions, design
+
+1:06:30.800 --> 1:06:40.240
+ decisions. Can you maybe let's take Python two, so Python releasing Python three as an example.
+
+1:06:40.240 --> 1:06:47.680
+ Mm hmm. It's not backward compatible to Python two in ways that a lot of people know. So what was
+
+1:06:47.680 --> 1:06:53.120
+ that deliberation discussion decision like? Yeah, what was the psychology of that experience?
+
+1:06:54.320 --> 1:07:01.360
+ Do you regret any aspects of how that experience undergone that? Well, yeah, so it was a group
+
+1:07:01.360 --> 1:07:10.640
+ process really. At that point, even though I was BDFL in name, and certainly everybody sort of
+
+1:07:11.200 --> 1:07:20.240
+ respected my position as the creator and the current sort of owner of the language design,
+
+1:07:21.680 --> 1:07:24.880
+ I was looking at everyone else for feedback.
+
+1:07:24.880 --> 1:07:35.600
+ Sort of Python 3.0 in some sense was sparked by other people in the community pointing out,
+
+1:07:36.880 --> 1:07:47.360
+ oh, well, there are a few issues that sort of bite users over and over. Can we do something
+
+1:07:47.360 --> 1:07:55.280
+ about that? And for Python three, we took a number of those Python words as they were called at the
+
+1:07:55.280 --> 1:08:04.880
+ time. And we said, can we try to sort of make small changes to the language that address those words?
+
+1:08:06.000 --> 1:08:14.080
+ And we had sort of in the past, we had always taken backwards compatibility very seriously.
+
+1:08:14.080 --> 1:08:20.000
+ And so many Python words in earlier versions had already been resolved, because they could be resolved
+
+1:08:21.040 --> 1:08:28.960
+ while maintaining backwards compatibility or sort of using a very gradual path of evolution of the
+
+1:08:28.960 --> 1:08:36.400
+ language in a certain area. And so we were stuck with a number of words that were widely recognized
+
+1:08:36.400 --> 1:08:45.120
+ as problems, not like roadblocks, but nevertheless, sort of things that some people trip over. And you
+
+1:08:45.120 --> 1:08:53.120
+ know that that's always the same thing that that people trip over when they trip. And we could not
+
+1:08:53.120 --> 1:09:00.640
+ think of a backwards compatible way of resolving those issues. But it's still an option to not
+
+1:09:00.640 --> 1:09:06.960
+ resolve the issues. And so yes, for for a long time, we had sort of resigned ourselves to well,
+
+1:09:06.960 --> 1:09:14.720
+ okay, the language is not going to be perfect in this way, and that way, and that way. And we sort
+
+1:09:14.720 --> 1:09:20.400
+ of certain of these I mean, there are still plenty of things where you can say, well, that's
+
+1:09:20.400 --> 1:09:32.800
+ that particular detail is better in Java or in R or in visual basic or whatever. And we're okay with
+
+1:09:32.800 --> 1:09:39.760
+ that because well, we can't easily change it. It's not too bad, we can do a little bit with user
+
+1:09:39.760 --> 1:09:50.080
+ education, or we can have static analyzer or warnings in in the parse or something. But there
+
+1:09:50.080 --> 1:09:55.200
+ were things where we thought, well, these are really problems that are not going away, they're
+
+1:09:55.200 --> 1:10:03.280
+ getting worse. In the future, we should do something about it. Do something. But ultimately, there is
+
+1:10:03.280 --> 1:10:10.480
+ a decision to be made, right? Yes. So was that the toughest decision in the history of Python you
+
+1:10:10.480 --> 1:10:17.600
+ had to make as the benevolent dictator for life? Or if not, what are other maybe even on a smaller
+
+1:10:17.600 --> 1:10:23.360
+ scale? What was the decision where you were really torn up about? Well, the toughest decision was
+
+1:10:23.360 --> 1:10:30.400
+ probably to resign. All right, let's go there. Hold on a second, then let me just because in the
+
+1:10:30.400 --> 1:10:35.040
+ interest of time too, because I have a few cool questions for you. And let's touch a really
+
+1:10:35.040 --> 1:10:40.480
+ important one because it was quite dramatic and beautiful in certain kinds of ways. In July this
+
+1:10:40.480 --> 1:10:47.760
+ year, three months ago, you wrote, now that PEP 572 is done, I don't ever want to have to fight so
+
+1:10:47.760 --> 1:10:53.360
+ hard for a PEP and find that so many people despise my decisions. I would like to remove myself
+
+1:10:53.360 --> 1:10:59.280
+ entirely from the decision process. I'll still be there for a while as an ordinary core developer.
+
+1:10:59.280 --> 1:11:06.240
+ And I'll still be available to mentor people possibly more available. But I'm basically giving
+
+1:11:06.240 --> 1:11:12.800
+ myself a permanent vacation from being BDFL benevolent dictator for life. And you all will
+
+1:11:12.800 --> 1:11:20.720
+ be on your own. First of all, just this, it's almost Shakespearean. I'm not going to appoint a
+
+1:11:20.720 --> 1:11:27.600
+ successor. So what are you all going to do? Create a democracy, anarchy, a dictatorship,
+
+1:11:27.600 --> 1:11:35.120
+ a federation. So that was a very dramatic and beautiful set of statements. It's almost,
+
+1:11:35.120 --> 1:11:41.120
+ it's open ended nature, called the community to create a future for Python. This is kind of a
+
+1:11:41.120 --> 1:11:48.080
+ beautiful aspect to it. Wow. So what and dramatic, you know, what was making that decision like?
+
+1:11:48.080 --> 1:11:52.400
+ What was on your heart, on your mind, stepping back now, a few months later,
+
+1:11:52.400 --> 1:12:00.800
+ taking it to your mindset? I'm glad you liked the writing because it was actually written pretty
+
+1:12:00.800 --> 1:12:12.320
+ quickly. It was literally something like after months and months of going around in circles,
+
+1:12:12.320 --> 1:12:24.000
+ I had finally approved PEP 572, which I had a big hand in its design, although I didn't
+
+1:12:24.000 --> 1:12:34.000
+ initiate it originally. I sort of gave it a bunch of nudges in a direction that would be
+
+1:12:34.000 --> 1:12:42.160
+ better for the language. So sorry, just to ask, is async IO, is that the one or no? No, PEP 572 was
+
+1:12:42.160 --> 1:12:48.240
+ actually a small feature, which is assignment expressions. Oh, assignment expressions, okay.
+
+1:12:49.120 --> 1:12:55.840
+ That had been thought there was just a lot of debate where a lot of people claimed that
+
+1:12:55.840 --> 1:13:03.440
+ they knew what was Pythonic and what was not Pythonic, and they knew that this was going to
+
+1:13:03.440 --> 1:13:10.480
+ destroy the language. This was like a violation of Python's most fundamental design philosophy.
+
+1:13:10.480 --> 1:13:15.920
+ And I thought that was all bullshit because I was in favor of it. And I would think I know
+
+1:13:15.920 --> 1:13:22.560
+ something about Python's design philosophy. So I was really tired and also stressed of that thing.
+
+1:13:22.560 --> 1:13:31.360
+ And literally, after sort of announcing, I was going to accept it. A certain Wednesday evening,
+
+1:13:31.920 --> 1:13:39.760
+ I had finally send the email, it's accepted. Now let's just go implement it. So I went to bed
+
+1:13:40.320 --> 1:13:48.480
+ feeling really relieved. That's behind me. And I wake up Thursday morning, 7am. And I think,
+
+1:13:48.480 --> 1:13:59.520
+ well, that was the last one. That's going to be such such a terrible debate. And that's
+
+1:13:59.520 --> 1:14:05.600
+ going to be that's the last time that I let myself be so stressed out about a PEP decision.
+
+1:14:06.480 --> 1:14:13.200
+ I should just resign. I've been sort of thinking about retirement for half a decade. I've been
+
+1:14:13.200 --> 1:14:22.080
+ joking and sort of mentioning retirement, sort of telling the community, some point in the
+
+1:14:22.080 --> 1:14:30.240
+ future, I'm going to retire. Don't take that FL part of my title too literally. And I thought,
+
+1:14:30.240 --> 1:14:38.560
+ okay, this is it. I'm done. I had the day off. I wanted to have a good time with my wife. We
+
+1:14:38.560 --> 1:14:48.240
+ were going to a little beach town nearby. And in, I think maybe 15, 20 minutes, I wrote that thing
+
+1:14:48.240 --> 1:14:53.600
+ that you just called Shakespearean. And the funny thing is, I get so much crap for calling you
+
+1:14:53.600 --> 1:15:00.320
+ Shakespearean. I didn't even I didn't even realize what a monumental decision it was.
+
+1:15:00.320 --> 1:15:08.640
+ Because five minutes later, I read that a link to my message back on Twitter, where people were
+
+1:15:08.640 --> 1:15:17.360
+ already discussing on Twitter, Guido resigned as the BDFL. And I had, I had posted it on an internal
+
+1:15:17.360 --> 1:15:22.880
+ forum that I thought was only read by core developers. So I thought I would at least
+
+1:15:22.880 --> 1:15:30.560
+ have one day before the news would sort of get out. The on your own aspects, I had also an
+
+1:15:30.560 --> 1:15:39.120
+ element of quite, it was quite a powerful element of the uncertainty that lies ahead. But can you
+
+1:15:39.120 --> 1:15:45.280
+ also just briefly talk about, you know, like, for example, I play guitar as a hobby for fun.
+
+1:15:45.280 --> 1:15:50.720
+ And whenever I play, people are super positive, super friendly. They're like, this is awesome.
+
+1:15:50.720 --> 1:15:56.320
+ This is great. But sometimes I enter as an outside observer, enter the programming community.
+
+1:15:57.120 --> 1:16:04.000
+ And there seems to some sometimes be camps on whatever the topic. And in the two camps,
+
+1:16:04.000 --> 1:16:08.880
+ the two or plus camps, are often pretty harsh at criticizing the opposing camps.
+
+1:16:11.520 --> 1:16:18.400
+ As an onlooker, I may be totally wrong on this. Yeah, holy wars are sort of a favorite activity
+
+1:16:18.400 --> 1:16:23.200
+ in the programming community. And what is the psychology behind that? Is, is that okay for
+
+1:16:23.200 --> 1:16:28.400
+ a healthy community to have? Is that, is that a productive force ultimately for the evolution
+
+1:16:28.400 --> 1:16:35.840
+ of a language? Well, if everybody is batting each other on the back and never telling the truth,
+
+1:16:35.840 --> 1:16:45.120
+ yes, it would not be a good thing. I think there is a middle ground where sort of
+
+1:16:48.640 --> 1:16:56.960
+ being nasty to each other is not okay. But there there is is is a middle ground where there is
+
+1:16:56.960 --> 1:17:06.880
+ is healthy ongoing criticism and feedback that is very productive. And you mean at every level,
+
+1:17:06.880 --> 1:17:13.840
+ you see that I mean, someone proposes to fix a very small issue in a code base.
+
+1:17:16.240 --> 1:17:22.480
+ Chances are that some reviewer will sort of respond by saying, well, actually,
+
+1:17:22.480 --> 1:17:31.520
+ you can do it better the other way. When it comes to deciding on the future of the Python
+
+1:17:31.520 --> 1:17:38.800
+ core developer community, we now have, I think, five or six competing proposals for a constitution.
+
+1:17:40.960 --> 1:17:46.320
+ So that future, do you have a fear of that future? Do you have a hope for that future?
+
+1:17:46.320 --> 1:17:54.880
+ I'm very confident about that future. And by and large, I think that the debate has been very
+
+1:17:54.880 --> 1:18:06.000
+ healthy and productive. And I actually when when I wrote that resignation email, I knew that that
+
+1:18:06.000 --> 1:18:11.920
+ Python was in a very good spot and that the Python core development community that the group of
+
+1:18:11.920 --> 1:18:21.200
+ 50 or 100 people who sort of write or review most of the code that goes into Python, those people
+
+1:18:22.400 --> 1:18:31.200
+ get along very well most of the time. A large number of different areas of expertise are
+
+1:18:31.200 --> 1:18:42.240
+ represented at different levels of experience in the Python core dev community, different levels
+
+1:18:42.240 --> 1:18:49.040
+ of experience completely outside it in software development in general, large systems, small
+
+1:18:49.040 --> 1:19:00.000
+ systems, embedded systems. So I felt okay, resigning because I knew that that the community can
+
+1:19:00.000 --> 1:19:08.240
+ really take care of itself. And out of a grab bag of future feature developments, let me ask if
+
+1:19:08.240 --> 1:19:15.360
+ you can comment, maybe on all very quickly, concurrent programming parallel computing,
+
+1:19:15.920 --> 1:19:23.520
+ async IO, these are things that people have expressed hope, complained about, whatever
+
+1:19:23.520 --> 1:19:31.360
+ I have discussed on Reddit, async IO, so the parallelization in general, packaging, I was totally
+
+1:19:31.360 --> 1:19:36.400
+ close on this, I just use pip install stuff, but apparently, there's pip end of poetry, there's
+
+1:19:36.400 --> 1:19:41.760
+ these dependency packaging systems that manage dependencies and so on, they're emerging, and
+
+1:19:41.760 --> 1:19:47.920
+ there's a lot of confusion about what's what's the right thing to use. Then also, functional
+
+1:19:47.920 --> 1:19:57.760
+ programming, the ever, are we going to get more functional programming or not, this kind of idea,
+
+1:19:58.560 --> 1:20:07.440
+ and of course, the GIL connected to the parallelization, I suppose, the global interpreter
+
+1:20:07.440 --> 1:20:12.240
+ lock problem. Can you just comment on whichever you want to comment on?
+
+1:20:12.240 --> 1:20:22.640
+ Well, let's take the GIL and parallelization and async IO as one one topic.
+
+1:20:25.280 --> 1:20:35.840
+ I'm not that hopeful that Python will develop into a sort of high concurrency, high parallelism
+
+1:20:35.840 --> 1:20:44.480
+ language. That's sort of the way the language is designed, the way most users use the language,
+
+1:20:44.480 --> 1:20:50.080
+ the way the language is implemented, all make that a pretty unlikely future.
+
+1:20:50.080 --> 1:20:56.000
+ So you think it might not even need to really the way people use it, it might not be something
+
+1:20:56.000 --> 1:21:02.400
+ that should be of great concern. I think I think async IO is a special case, because it sort of
+
+1:21:02.400 --> 1:21:14.320
+ allows overlapping IO and only IO. And that is is a sort of best practice of supporting very
+
+1:21:14.320 --> 1:21:24.000
+ high throughput IO, many connections per second. I'm not worried about that. I think async IO
+
+1:21:24.000 --> 1:21:30.080
+ will evolve. There are a couple of competing packages, we have some very smart people who are
+
+1:21:30.080 --> 1:21:39.120
+ sort of pushing us in sort of to make async IO better. Parallel computing, I think that
+
+1:21:40.320 --> 1:21:46.160
+ Python is not the language for that. There are there are ways to work around it.
+
+1:21:47.120 --> 1:21:54.960
+ But you sort of you can't expect to write an algorithm in Python and have a compiler
+
+1:21:54.960 --> 1:22:01.200
+ automatically paralyze that what you can do is use a package like NumPy and there are a bunch of
+
+1:22:01.200 --> 1:22:10.080
+ other very powerful packages that sort of use all the CPUs available, because you tell the package,
+
+1:22:10.720 --> 1:22:17.200
+ here's the data, here's the abstract operation to apply over it, go at it, and then then we're
+
+1:22:17.200 --> 1:22:23.440
+ back in the C++ world. But the those packages are themselves implemented usually in C++.
+
+1:22:23.440 --> 1:22:26.960
+ That's right. That's where TensorFlow and all these packages come in where they parallelize
+
+1:22:26.960 --> 1:22:32.800
+ across GPUs, for example, they take care of that for you. So in terms of packaging, can you comment
+
+1:22:32.800 --> 1:22:43.760
+ on the packaging? Yeah, my packaging has always been my least favorite topic. It's a really tough
+
+1:22:43.760 --> 1:22:57.440
+ problem because the OS and the platform want to own packaging. But their packaging solution is not
+
+1:22:57.440 --> 1:23:04.240
+ specific to a language. Like, if you take Linux, there are two competing packaging solutions for
+
+1:23:04.240 --> 1:23:16.000
+ Linux, or for Unix in general. And but they all work across all languages. And several languages,
+
+1:23:16.000 --> 1:23:25.600
+ like Node, JavaScript, and Ruby, and Python all have their own packaging solutions that only work
+
+1:23:25.600 --> 1:23:34.400
+ within the ecosystem of that language. Well, what should you use? That is a tough problem.
+
+1:23:34.400 --> 1:23:43.520
+ My own own approach is I use the system packaging system to install Python, and I use the Python
+
+1:23:43.520 --> 1:23:50.240
+ packaging system then to install third party Python packages. That's what most people do.
+
+1:23:50.240 --> 1:23:58.160
+ 10 years ago, Python packaging was really a terrible situation. Nowadays, Pip is the future.
+
+1:23:58.160 --> 1:24:05.360
+ There is there is a separate ecosystem for numerical and scientific Python, Python based on
+
+1:24:05.360 --> 1:24:11.280
+ Anaconda. Those two can live together. I don't think there is a need for more than that.
+
+1:24:11.280 --> 1:24:16.800
+ Great. So that's that's packaging. That's, well, at least for me, that's that's where I've been
+
+1:24:16.800 --> 1:24:22.240
+ extremely happy. I didn't I didn't even know this was an issue until it was brought up. Well, in the
+
+1:24:22.240 --> 1:24:27.920
+ interest of time, let me sort of skip through a million other questions I have. So I watched the
+
+1:24:27.920 --> 1:24:33.840
+ five hour five five and a half hour oral history. They've done with the computer history museum.
+
+1:24:33.840 --> 1:24:38.480
+ And the nice thing about it, it gave this because of the linear progression of the interview, it
+
+1:24:38.480 --> 1:24:47.040
+ it gave this feeling of a life, you know, a life well lived with interesting things in it.
+
+1:24:47.040 --> 1:24:52.960
+ Sort of a pretty, I would say a good spend of of this little existence we have on earth.
+
+1:24:52.960 --> 1:24:59.920
+ So outside of your family, looking back, what about this journey are you really proud of?
+
+1:24:59.920 --> 1:25:10.240
+ Are there moments that stand out accomplishments ideas? Is it the creation of Python itself
+
+1:25:10.240 --> 1:25:15.040
+ that stands out as a thing that you look back and say, damn, I did pretty good there?
+
+1:25:17.600 --> 1:25:21.760
+ Well, I would say that Python is definitely the best thing I've ever done.
+
+1:25:21.760 --> 1:25:34.000
+ And I wouldn't sort of say just the creation of Python, but the way I sort of raised Python,
+
+1:25:34.000 --> 1:25:41.440
+ like a baby, I didn't just conceive a child, but I raised a child. And now I'm setting the child
+
+1:25:41.440 --> 1:25:48.800
+ free in the world. And I've set up the child to to sort of be able to take care of himself.
+
+1:25:48.800 --> 1:25:55.760
+ And I'm very proud of that. And as the announcer of Monty Python's Flying Circus used to say,
+
+1:25:55.760 --> 1:26:01.200
+ and now for something completely different, do you have a favorite Monty Python moment or a
+
+1:26:01.200 --> 1:26:05.680
+ moment in Hitchhiker's Guide or any other literature show or movie that cracks you up when you think
+
+1:26:05.680 --> 1:26:12.480
+ about it? Oh, you can always play me the Parrots, the dead Parrot sketch. Oh, that's brilliant.
+
+1:26:12.480 --> 1:26:19.360
+ Yeah, that's my favorite as well. Pushing up the daisies. Okay, Greta, thank you so much for
+
+1:26:19.360 --> 1:26:44.000
+ talking to me today. Lex, this has been a great conversation.
+

vtt/episode_008_small.vtt

@@ -0,0 +1,1418 @@
+WEBVTT
+
+00:00.000 --> 00:02.800
+ The following is a conversation with Eric Schmidt.
+
+00:02.800 --> 00:07.640
+ He was the CEO of Google for ten years and a chairman for six more, guiding the company
+
+00:07.640 --> 00:13.080
+ through an incredible period of growth and a series of world changing innovations.
+
+00:13.080 --> 00:19.280
+ He is one of the most impactful leaders in the era of the internet and the powerful voice
+
+00:19.280 --> 00:22.400
+ for the promise of technology in our society.
+
+00:22.400 --> 00:27.760
+ It was truly an honor to speak with him as part of the MIT course on artificial general
+
+00:27.760 --> 00:32.040
+ intelligence and the artificial intelligence podcast.
+
+00:32.040 --> 00:37.120
+ And now here's my conversation with Eric Schmidt.
+
+00:37.120 --> 00:40.840
+ What was the first moment when you fell in love with technology?
+
+00:40.840 --> 00:47.080
+ I grew up in the 1960s as a boy where every boy wanted to be an astronaut and part of
+
+00:47.080 --> 00:49.000
+ the space program.
+
+00:49.000 --> 00:54.400
+ So like everyone else of my age, we would go out to the cow pasture behind my house,
+
+00:54.400 --> 00:58.640
+ which was literally a cow pasture, and we would shoot model rockets off.
+
+00:58.640 --> 01:00.200
+ And that I think is the beginning.
+
+01:00.200 --> 01:05.680
+ And of course, generationally, today, it would be video games and all the amazing things
+
+01:05.680 --> 01:09.280
+ that you can do online with computers.
+
+01:09.280 --> 01:15.080
+ There's a transformative, inspiring aspect of science and math that maybe rockets would
+
+01:15.080 --> 01:17.560
+ bring, would instill in individuals.
+
+01:17.560 --> 01:21.720
+ You've mentioned yesterday that eighth grade math is where the journey through Mathematical
+
+01:21.720 --> 01:24.520
+ University diverges from many people.
+
+01:24.520 --> 01:27.080
+ It's this fork in the roadway.
+
+01:27.080 --> 01:31.160
+ There's a professor of math at Berkeley, Edward Franco.
+
+01:31.160 --> 01:32.720
+ I'm not sure if you're familiar with him.
+
+01:32.720 --> 01:33.720
+ I am.
+
+01:33.720 --> 01:35.400
+ He has written this amazing book.
+
+01:35.400 --> 01:41.960
+ I recommend to everybody called Love and Math, two of my favorite words.
+
+01:41.960 --> 01:49.680
+ He says that if painting was taught like math, then students would be asked to paint a fence,
+
+01:49.680 --> 01:54.520
+ which is his analogy of essentially how math is taught, and you never get a chance to discover
+
+01:54.520 --> 01:59.400
+ the beauty of the art of painting or the beauty of the art of math.
+
+01:59.400 --> 02:05.240
+ So how, when, and where did you discover that beauty?
+
+02:05.240 --> 02:12.040
+ I think what happens with people like myself is that your math enabled pretty early, and
+
+02:12.040 --> 02:16.640
+ all of a sudden you discover that you can use that to discover new insights.
+
+02:16.640 --> 02:22.120
+ The great scientists will all tell a story, the men and women who are fantastic today,
+
+02:22.120 --> 02:25.560
+ that somewhere when they were in high school or in college, they discovered that they could
+
+02:25.560 --> 02:30.760
+ discover something themselves, and that sense of building something, of having an impact
+
+02:30.760 --> 02:35.520
+ that you own drives knowledge, acquisition, and learning.
+
+02:35.520 --> 02:41.000
+ In my case, it was programming, and the notion that I could build things that had not existed
+
+02:41.000 --> 02:46.560
+ that I had built, that it had my name on it, and this was before open source, but you could
+
+02:46.560 --> 02:49.160
+ think of it as open source contributions.
+
+02:49.160 --> 02:53.760
+ So today, if I were a 16 or 17 year old boy, I'm sure that I would aspire as a computer
+
+02:53.760 --> 02:59.000
+ scientist to make a contribution like the open source heroes of the world today.
+
+02:59.000 --> 03:03.760
+ That would be what would be driving me, and I'd be trying and learning and making mistakes,
+
+03:03.760 --> 03:06.720
+ and so forth, in the ways that it works.
+
+03:06.720 --> 03:12.360
+ The repository that GitHub represents and that open source libraries represent is an
+
+03:12.360 --> 03:17.840
+ enormous bank of knowledge of all of the people who are doing that, and one of the lessons
+
+03:17.840 --> 03:22.240
+ that I learned at Google was that the world is a very big place, and there's an awful
+
+03:22.240 --> 03:26.360
+ lot of smart people, and an awful lot of them are underutilized.
+
+03:26.360 --> 03:32.240
+ So here's an opportunity, for example, building parts of programs, building new ideas to contribute
+
+03:32.240 --> 03:36.640
+ to the greater of society.
+
+03:36.640 --> 03:41.000
+ So in that moment in the 70s, the inspiring moment where there was nothing, and then you
+
+03:41.000 --> 03:44.800
+ created something through programming, that magical moment.
+
+03:44.800 --> 03:50.360
+ So in 1975, I think you've created a program called Lex, which I especially like because
+
+03:50.360 --> 03:51.560
+ my name is Lex.
+
+03:51.560 --> 03:52.560
+ So thank you.
+
+03:52.560 --> 03:58.240
+ Thank you for creating a brand that established a reputation that's long lasting reliable
+
+03:58.240 --> 04:01.240
+ and has a big impact on the world and still used today.
+
+04:01.240 --> 04:03.000
+ So thank you for that.
+
+04:03.000 --> 04:11.880
+ But more seriously, in that time, in the 70s, as an engineer, personal computers were being
+
+04:11.880 --> 04:12.880
+ born.
+
+04:12.880 --> 04:17.800
+ Do you think you'd be able to predict the 80s, 90s, and the aughts of where computers
+
+04:17.800 --> 04:18.800
+ would go?
+
+04:18.800 --> 04:23.160
+ I'm sure I could not and would not have gotten it right.
+
+04:23.160 --> 04:27.960
+ I was the beneficiary of the great work of many, many people who saw it clearer than I
+
+04:27.960 --> 04:29.160
+ did.
+
+04:29.160 --> 04:34.760
+ With Lex, I worked with a fellow named Michael Lesk, who was my supervisor, and he essentially
+
+04:34.760 --> 04:39.400
+ helped me architect and deliver a system that's still in use today.
+
+04:39.400 --> 04:43.800
+ After that, I worked at Xerox Palo Alto Research Center, where the Alto was invented, and the
+
+04:43.800 --> 04:50.600
+ Alto is the predecessor of the modern personal computer, or Macintosh, and so forth.
+
+04:50.600 --> 04:55.360
+ And the altos were very rare, and I had to drive an hour from Berkeley to go use them,
+
+04:55.360 --> 05:01.160
+ but I made a point of skipping classes and doing whatever it took to have access to this
+
+05:01.160 --> 05:02.480
+ extraordinary achievement.
+
+05:02.480 --> 05:05.000
+ I knew that they were consequential.
+
+05:05.000 --> 05:08.240
+ What I did not understand was scaling.
+
+05:08.240 --> 05:12.960
+ I did not understand what would happen when you had 100 million as opposed to 100.
+
+05:12.960 --> 05:17.360
+ And so since then, and I have learned the benefit of scale, I always look for things
+
+05:17.360 --> 05:23.160
+ which are going to scale to platforms, so mobile phones, Android, all those things.
+
+05:23.160 --> 05:27.560
+ There are many, many people in the world.
+
+05:27.560 --> 05:28.560
+ People really have needs.
+
+05:28.560 --> 05:32.600
+ They really will use these platforms, and you can build big businesses on top of them.
+
+05:32.600 --> 05:33.600
+ So it's interesting.
+
+05:33.600 --> 05:36.880
+ So when you see a piece of technology, now you think, what will this technology look
+
+05:36.880 --> 05:39.080
+ like when it's in the hands of a billion people?
+
+05:39.080 --> 05:40.160
+ That's right.
+
+05:40.160 --> 05:46.520
+ So an example would be that the market is so competitive now that if you can't figure
+
+05:46.520 --> 05:51.880
+ out a way for something to have a million users or a billion users, it probably is not
+
+05:51.880 --> 05:57.280
+ going to be successful because something else will become the general platform, and your
+
+05:57.280 --> 06:04.360
+ idea will become a lost idea or a specialized service with relatively few users.
+
+06:04.360 --> 06:06.000
+ So it's a path to generality.
+
+06:06.000 --> 06:07.720
+ It's a path to general platform use.
+
+06:07.720 --> 06:09.400
+ It's a path to broad applicability.
+
+06:09.400 --> 06:15.000
+ Now, there are plenty of good businesses that are tiny, so luxury goods, for example.
+
+06:15.000 --> 06:20.360
+ But if you want to have an impact at scale, you have to look for things which are of
+
+06:20.360 --> 06:25.200
+ common value, common pricing, common distribution, and solve common problems, the problems that
+
+06:25.200 --> 06:26.200
+ everyone has.
+
+06:26.200 --> 06:30.440
+ And by the way, people have lots of problems, information, medicine, health, education,
+
+06:30.440 --> 06:31.440
+ and so forth.
+
+06:31.440 --> 06:32.440
+ Work on those problems.
+
+06:32.440 --> 06:40.240
+ Like you said, you're a big fan of the middle class because there's so many of them by definition.
+
+06:40.240 --> 06:46.600
+ So any product, any thing that has a huge impact, it improves their lives is a great
+
+06:46.600 --> 06:48.960
+ business decision and it's just good for society.
+
+06:48.960 --> 06:53.520
+ And there's nothing wrong with starting off in the high end as long as you have a plan
+
+06:53.520 --> 06:55.520
+ to get to the middle class.
+
+06:55.520 --> 06:59.280
+ There's nothing wrong with starting with a specialized market in order to learn and to
+
+06:59.280 --> 07:01.080
+ build and to fund things.
+
+07:01.080 --> 07:04.520
+ So you start a luxury market to build a general purpose market.
+
+07:04.520 --> 07:09.640
+ But if you define yourself as only a narrow market, someone else can come along with a
+
+07:09.640 --> 07:14.320
+ general purpose market that can push you to the corner, can restrict the scale of operation,
+
+07:14.320 --> 07:17.320
+ can force you to be a lesser impact than you might be.
+
+07:17.320 --> 07:22.800
+ So it's very important to think in terms of broad businesses and broad impact, even if
+
+07:22.800 --> 07:26.360
+ you start in a little corner somewhere.
+
+07:26.360 --> 07:33.200
+ So as you look to the 70s, but also in the decades to come, and you saw computers, did
+
+07:33.200 --> 07:40.240
+ you see them as tools or was there a little element of another entity?
+
+07:40.240 --> 07:46.240
+ I remember a quote saying AI began with our dream to create the gods.
+
+07:46.240 --> 07:51.520
+ Is there a feeling when you wrote that program that you were creating another entity, giving
+
+07:51.520 --> 07:52.800
+ life to something?
+
+07:52.800 --> 07:58.880
+ I wish I could say otherwise, but I simply found the technology platforms so exciting.
+
+07:58.880 --> 08:00.400
+ That's what I was focused on.
+
+08:00.400 --> 08:04.640
+ I think the majority of the people that I've worked with, and there are a few exceptions,
+
+08:04.640 --> 08:09.960
+ Steve Jobs being an example, really saw this as a great technological play.
+
+08:09.960 --> 08:15.520
+ I think relatively few of the technical people understood the scale of its impact.
+
+08:15.520 --> 08:19.680
+ So I used NCP, which is a predecessor to TCPIP.
+
+08:19.680 --> 08:21.240
+ It just made sense to connect things.
+
+08:21.240 --> 08:26.240
+ We didn't think of it in terms of the internet, and then companies, and then Facebook, and
+
+08:26.240 --> 08:29.200
+ then Twitter, and then politics, and so forth.
+
+08:29.200 --> 08:30.800
+ We never did that build.
+
+08:30.800 --> 08:32.920
+ We didn't have that vision.
+
+08:32.920 --> 08:38.200
+ And I think most people, it's a rare person who can see compounding at scale.
+
+08:38.200 --> 08:41.520
+ Most people can see, if you ask people to predict the future, they'll say, they'll give
+
+08:41.520 --> 08:44.080
+ you an answer of six to nine months or 12 months.
+
+08:44.080 --> 08:47.560
+ Because that's about as far as people can imagine.
+
+08:47.560 --> 08:51.020
+ But there's an old saying, which actually was attributed to a professor at MIT a long
+
+08:51.020 --> 08:58.120
+ time ago, that we overestimate what can be done in one year, and we underestimate what
+
+08:58.120 --> 09:00.280
+ can be done in a decade.
+
+09:00.280 --> 09:05.560
+ And there's a great deal of evidence that these core platforms at hardware and software
+
+09:05.560 --> 09:07.800
+ take a decade.
+
+09:07.800 --> 09:09.600
+ So think about self driving cars.
+
+09:09.600 --> 09:12.160
+ Self driving cars were thought about in the 90s.
+
+09:12.160 --> 09:17.160
+ Over projects around them, the first DARPA Durand Challenge was roughly 2004.
+
+09:17.160 --> 09:19.760
+ So that's roughly 15 years ago.
+
+09:19.760 --> 09:25.400
+ And today we have self driving cars operating in a city in Arizona, right, so 15 years.
+
+09:25.400 --> 09:31.720
+ And we still have a ways to go before they're more generally available.
+
+09:31.720 --> 09:33.840
+ So you've spoken about the importance.
+
+09:33.840 --> 09:37.080
+ You just talked about predicting into the future.
+
+09:37.080 --> 09:41.640
+ You've spoken about the importance of thinking five years ahead and having a plan for those
+
+09:41.640 --> 09:42.640
+ five years.
+
+09:42.640 --> 09:47.840
+ And the way to say it is that almost everybody has a one year plan.
+
+09:47.840 --> 09:50.960
+ Almost no one has a proper five year plan.
+
+09:50.960 --> 09:55.160
+ And the key thing to having a five year plan is having a model for what's going to happen
+
+09:55.160 --> 09:57.000
+ under the underlying platforms.
+
+09:57.000 --> 09:59.840
+ So here's an example.
+
+09:59.840 --> 10:05.120
+ Computer Moore's Law, as we know it, the thing that powered improvements in CPUs has largely
+
+10:05.120 --> 10:10.400
+ halted in its traditional shrinking mechanism, because the costs have just gotten so high.
+
+10:10.400 --> 10:12.200
+ It's getting harder and harder.
+
+10:12.200 --> 10:16.600
+ But there's plenty of algorithmic improvements and specialized hardware improvements.
+
+10:16.600 --> 10:21.240
+ So you need to understand the nature of those improvements and where they'll go in order
+
+10:21.240 --> 10:24.360
+ to understand how it will change the platform.
+
+10:24.360 --> 10:28.000
+ In the area of network connectivity, what are the gains that are going to be possible
+
+10:28.000 --> 10:29.480
+ in wireless?
+
+10:29.480 --> 10:35.720
+ It looks like there's an enormous expansion of wireless connectivity at many different
+
+10:35.720 --> 10:36.960
+ bands, right?
+
+10:36.960 --> 10:40.520
+ And that we will primarily, historically, I've always thought that we were primarily
+
+10:40.520 --> 10:45.040
+ going to be using fiber, but now it looks like we're going to be using fiber plus very
+
+10:45.040 --> 10:51.560
+ powerful high bandwidth sort of short distance connectivity to bridge the last mile, right?
+
+10:51.560 --> 10:53.100
+ That's an amazing achievement.
+
+10:53.100 --> 10:56.880
+ If you know that, then you're going to build your systems differently.
+
+10:56.880 --> 10:59.800
+ By the way, those networks have different latency properties, right?
+
+10:59.800 --> 11:05.040
+ Because they're more symmetric, the algorithms feel faster for that reason.
+
+11:05.040 --> 11:09.920
+ And so when you think about whether it's a fiber or just technologies in general.
+
+11:09.920 --> 11:15.920
+ So there's this barber, wooden poem or quote that I really like.
+
+11:15.920 --> 11:20.400
+ It's from the champions of the impossible rather than the slaves of the possible that
+
+11:20.400 --> 11:23.280
+ evolution draws its creative force.
+
+11:23.280 --> 11:27.840
+ So in predicting the next five years, I'd like to talk about the impossible and the
+
+11:27.840 --> 11:28.840
+ possible.
+
+11:28.840 --> 11:34.720
+ Well, and again, one of the great things about humanity is that we produce dreamers, right?
+
+11:34.720 --> 11:37.760
+ We literally have people who have a vision and a dream.
+
+11:37.760 --> 11:43.400
+ They are, if you will, disagreeable in the sense that they disagree with the, they disagree
+
+11:43.400 --> 11:46.240
+ with what the sort of zeitgeist is.
+
+11:46.240 --> 11:48.040
+ They say there is another way.
+
+11:48.040 --> 11:49.040
+ They have a belief.
+
+11:49.040 --> 11:50.320
+ They have a vision.
+
+11:50.320 --> 11:56.560
+ If you look at science, science is always marked by such people who, who went against
+
+11:56.560 --> 12:01.360
+ some conventional wisdom, collected the knowledge at the time and assembled it in a way that
+
+12:01.360 --> 12:03.760
+ produced a powerful platform.
+
+12:03.760 --> 12:11.120
+ And you've been amazingly honest about in an inspiring way about things you've been wrong
+
+12:11.120 --> 12:14.800
+ about predicting and you've obviously been right about a lot of things.
+
+12:14.800 --> 12:23.880
+ But in this kind of tension, how do you balance as a company in predicting the next five years,
+
+12:23.880 --> 12:26.520
+ the impossible, planning for the impossible.
+
+12:26.520 --> 12:32.720
+ So listening to those crazy dreamers, letting them do, letting them run away and make the
+
+12:32.720 --> 12:38.760
+ impossible real, make it happen and slow, you know, that's how programmers often think
+
+12:38.760 --> 12:44.800
+ and slowing things down and saying, well, this is the rational, this is the possible,
+
+12:44.800 --> 12:49.160
+ the pragmatic, the, the dreamer versus the pragmatist.
+
+12:49.160 --> 12:56.680
+ So it's helpful to have a model which encourages a predictable revenue stream as well as the
+
+12:56.680 --> 12:58.720
+ ability to do new things.
+
+12:58.720 --> 13:03.120
+ So in Google's case, we're big enough and well enough managed and so forth that we have
+
+13:03.120 --> 13:06.600
+ a pretty good sense of what our revenue will be for the next year or two, at least for
+
+13:06.600 --> 13:07.960
+ a while.
+
+13:07.960 --> 13:13.720
+ And so we have enough cash generation that we can make bets.
+
+13:13.720 --> 13:18.760
+ And indeed, Google has become alphabet so the corporation is organized around these
+
+13:18.760 --> 13:19.760
+ bets.
+
+13:19.760 --> 13:25.560
+ And these bets are in areas of fundamental importance to, to the world, whether it's
+
+13:25.560 --> 13:31.920
+ digital intelligence, medical technology, self driving cars, connectivity through balloons,
+
+13:31.920 --> 13:33.440
+ on and on and on.
+
+13:33.440 --> 13:36.080
+ And there's more coming and more coming.
+
+13:36.080 --> 13:41.480
+ So one way you could express this is that the current business is successful enough
+
+13:41.480 --> 13:44.720
+ that we have the luxury of making bets.
+
+13:44.720 --> 13:48.960
+ And another one that you could say is that we have the, the wisdom of being able to see
+
+13:48.960 --> 13:53.920
+ that a corporate structure needs to be created to enhance the likelihood of the success of
+
+13:53.920 --> 13:55.320
+ those bets.
+
+13:55.320 --> 13:59.760
+ So we essentially turned ourselves into a conglomerate of bets.
+
+13:59.760 --> 14:04.360
+ And then this underlying corporation Google, which is itself innovative.
+
+14:04.360 --> 14:08.160
+ So in order to pull this off, you have to have a bunch of belief systems.
+
+14:08.160 --> 14:12.080
+ And one of them is that you have to have bottoms up and tops down the bottoms up.
+
+14:12.080 --> 14:13.600
+ We call 20% time.
+
+14:13.600 --> 14:17.040
+ And the idea is that people can spend 20% of the time on whatever they want.
+
+14:17.040 --> 14:21.960
+ And the top down is that our founders in particular have a keen eye on technology and they're
+
+14:21.960 --> 14:24.000
+ reviewing things constantly.
+
+14:24.000 --> 14:27.520
+ So an example would be they'll, they'll hear about an idea or I'll hear about something
+
+14:27.520 --> 14:28.880
+ and it sounds interesting.
+
+14:28.880 --> 14:34.920
+ Let's go visit them and then let's begin to assemble the pieces to see if that's possible.
+
+14:34.920 --> 14:39.920
+ And if you do this long enough, you get pretty good at predicting what's likely to work.
+
+14:39.920 --> 14:42.120
+ So that's, that's a beautiful balance that struck.
+
+14:42.120 --> 14:44.560
+ Is this something that applies at all scale?
+
+14:44.560 --> 14:53.960
+ So seems seems to be that the Sergei, again, 15 years ago, came up with a concept that
+
+14:53.960 --> 14:59.040
+ called 10% of the budget should be on things that are unrelated.
+
+14:59.040 --> 15:05.040
+ It was called 70, 20, 10, 70% of our time on core business, 20% on adjacent business
+
+15:05.040 --> 15:06.920
+ and 10% on other.
+
+15:06.920 --> 15:11.200
+ And he proved mathematically, of course, he's a brilliant mathematician, that you needed
+
+15:11.200 --> 15:18.800
+ that 10% right to make the sum of the growth work and it turns out he was right.
+
+15:18.800 --> 15:24.320
+ So getting into the world of artificial intelligence, you've, you've talked quite extensively and
+
+15:24.320 --> 15:32.160
+ effectively to the impact in the near term, the positive impact of artificial intelligence,
+
+15:32.160 --> 15:38.720
+ whether it's machine, especially machine learning in medical applications and education
+
+15:38.720 --> 15:44.160
+ and just making information more accessible, right in the AI community, there is a kind
+
+15:44.160 --> 15:49.520
+ of debate, so there's this shroud of uncertainty as we face this new world with artificial
+
+15:49.520 --> 15:50.800
+ intelligence in it.
+
+15:50.800 --> 15:57.560
+ And there is some people like Elon Musk, you've disagreed on at least on the degree of emphasis
+
+15:57.560 --> 16:00.800
+ he places on the existential threat of AI.
+
+16:00.800 --> 16:07.120
+ So I've spoken with Stuart Russell, Max Tagmark, who share Elon Musk's view, and Yoshio Benjio,
+
+16:07.120 --> 16:09.240
+ Steven Pinker, who do not.
+
+16:09.240 --> 16:13.320
+ And so there's a, there's a, there's a lot of very smart people who are thinking about
+
+16:13.320 --> 16:17.280
+ this stuff, disagreeing, which is really healthy, of course.
+
+16:17.280 --> 16:22.800
+ So what do you think is the healthiest way for the AI community to, and really for the
+
+16:22.800 --> 16:30.880
+ general public to think about AI and the concern of the technology being mismanaged
+
+16:30.880 --> 16:33.000
+ in some, in some kind of way.
+
+16:33.000 --> 16:37.560
+ So the source of education for the general public has been robot killer movies.
+
+16:37.560 --> 16:38.760
+ Right.
+
+16:38.760 --> 16:44.840
+ And Terminator, et cetera, and the one thing I can assure you we're not building are those
+
+16:44.840 --> 16:46.200
+ kinds of solutions.
+
+16:46.200 --> 16:51.240
+ Furthermore, if they were to show up, someone would notice and unplug them, right?
+
+16:51.240 --> 16:57.760
+ So as exciting as those movies are, and they're great movies, where the killer robots to start,
+
+16:57.760 --> 17:00.520
+ we would find a way to stop them, right?
+
+17:00.520 --> 17:04.320
+ So I'm not concerned about that.
+
+17:04.320 --> 17:08.680
+ And much of this has to do with the timeframe of conversation.
+
+17:08.680 --> 17:16.120
+ So you can imagine a situation a hundred years from now, when the human brain is fully understood,
+
+17:16.120 --> 17:19.880
+ and the next generation and next generation of brilliant MIT scientists have figured
+
+17:19.880 --> 17:25.960
+ all this out, we're going to have a large number of ethics questions, right, around science
+
+17:25.960 --> 17:29.760
+ and thinking and robots and computers and so forth and so on.
+
+17:29.760 --> 17:32.360
+ So it depends on the question of the timeframe.
+
+17:32.360 --> 17:37.280
+ In the next five to 10 years, we're not facing those questions.
+
+17:37.280 --> 17:42.000
+ What we're facing in the next five to 10 years is how do we spread this disruptive technology
+
+17:42.000 --> 17:46.520
+ as broadly as possible to gain the maximum benefit of it?
+
+17:46.520 --> 17:51.880
+ The primary benefit should be in healthcare and in education, healthcare because it's
+
+17:51.880 --> 17:52.880
+ obvious.
+
+17:52.880 --> 17:55.840
+ We're all the same, even though we somehow believe we're not.
+
+17:55.840 --> 18:00.400
+ As a medical matter, the fact that we have big data about our health will save lives,
+
+18:00.400 --> 18:05.520
+ allow us to deal with skin cancer and other cancers, ophthalmological problems.
+
+18:05.520 --> 18:10.080
+ There's people working on psychological diseases and so forth using these techniques.
+
+18:10.080 --> 18:11.680
+ I go on and on.
+
+18:11.680 --> 18:15.840
+ The promise of AI in medicine is extraordinary.
+
+18:15.840 --> 18:20.360
+ There are many, many companies and startups and funds and solutions and we will all live
+
+18:20.360 --> 18:22.120
+ much better for that.
+
+18:22.120 --> 18:25.680
+ The same argument in education.
+
+18:25.680 --> 18:31.760
+ Can you imagine that for each generation of child and even adult, you have a tutor educator
+
+18:31.760 --> 18:37.320
+ that's AI based, that's not a human but is properly trained, that helps you get smarter,
+
+18:37.320 --> 18:41.440
+ helps you address your language difficulties or your math difficulties or what have you.
+
+18:41.440 --> 18:43.400
+ Why don't we focus on those two?
+
+18:43.400 --> 18:49.240
+ The gains societally of making humans smarter and healthier are enormous.
+
+18:49.240 --> 18:54.000
+ Those translate for decades and decades and will all benefit from them.
+
+18:54.000 --> 18:58.560
+ There are people who are working on AI safety, which is the issue that you're describing.
+
+18:58.560 --> 19:02.920
+ There are conversations in the community that should there be such problems, what should
+
+19:02.920 --> 19:04.360
+ the rules be like?
+
+19:04.360 --> 19:09.360
+ Google, for example, has announced its policies with respect to AI safety, which I certainly
+
+19:09.360 --> 19:14.320
+ support and I think most everybody would support and they make sense.
+
+19:14.320 --> 19:19.760
+ It helps guide the research but the killer robots are not arriving this year and they're
+
+19:19.760 --> 19:23.840
+ not even being built.
+
+19:23.840 --> 19:31.040
+ On that line of thinking, you said the timescale, in this topic or other topics, have you found
+
+19:31.040 --> 19:37.920
+ a useful, on the business side or the intellectual side, to think beyond 5, 10 years, to think
+
+19:37.920 --> 19:39.480
+ 50 years out?
+
+19:39.480 --> 19:42.000
+ Has it ever been useful or productive?
+
+19:42.000 --> 19:49.040
+ In our industry, there are essentially no examples of 50 year predictions that have been correct.
+
+19:49.040 --> 19:54.320
+ Let's review AI, which was largely invented here at MIT and a couple of other universities
+
+19:54.320 --> 19:57.840
+ in 1956, 1957, 1958.
+
+19:57.840 --> 20:01.800
+ The original claims were a decade or two.
+
+20:01.800 --> 20:08.040
+ When I was a PhD student, I studied AI a bit and it entered during my looking at it a period
+
+20:08.040 --> 20:13.880
+ which is known as AI winter, which went on for about 30 years, which is a whole generation
+
+20:13.880 --> 20:18.800
+ of scientists and a whole group of people who didn't make a lot of progress because the
+
+20:18.800 --> 20:22.160
+ algorithms had not improved and the computers had not improved.
+
+20:22.160 --> 20:26.160
+ It took some brilliant mathematicians, starting with a fellow named Jeff Hinton at Toronto
+
+20:26.160 --> 20:33.120
+ in Montreal, who basically invented this deep learning model which empowers us today.
+
+20:33.120 --> 20:40.400
+ The seminal work there was 20 years ago and in the last 10 years, it's become popularized.
+
+20:40.400 --> 20:43.520
+ Think about the time frames for that level of discovery.
+
+20:43.520 --> 20:46.080
+ It's very hard to predict.
+
+20:46.080 --> 20:50.240
+ Many people think that we'll be flying around in the equivalent of flying cars.
+
+20:50.240 --> 20:51.240
+ Who knows?
+
+20:51.240 --> 20:56.680
+ My own view, if I want to go out on a limb, is to say that we know a couple of things
+
+20:56.680 --> 20:58.000
+ about 50 years from now.
+
+20:58.000 --> 21:00.480
+ We know that there'll be more people alive.
+
+21:00.480 --> 21:04.000
+ We know that we'll have to have platforms that are more sustainable because the earth
+
+21:04.000 --> 21:09.440
+ is limited in the ways we all know and that the kind of platforms that are going to get
+
+21:09.440 --> 21:13.000
+ billed will be consistent with the principles that I've described.
+
+21:13.000 --> 21:17.560
+ They will be much more empowering of individuals, they'll be much more sensitive to the ecology
+
+21:17.560 --> 21:20.440
+ because they have to be, they just have to be.
+
+21:20.440 --> 21:24.160
+ I also think that humans are going to be a great deal smarter and I think they're going
+
+21:24.160 --> 21:28.320
+ to be a lot smarter because of the tools that I've discussed with you and of course people
+
+21:28.320 --> 21:29.320
+ will live longer.
+
+21:29.320 --> 21:32.080
+ Life extension is continuing apace.
+
+21:32.080 --> 21:36.840
+ A baby born today has a reasonable chance of living to 100, which is pretty exciting.
+
+21:36.840 --> 21:40.760
+ It's well past the 21st century, so we better take care of them.
+
+21:40.760 --> 21:46.160
+ You mentioned interesting statistic on some very large percentage, 60, 70% of people may
+
+21:46.160 --> 21:48.360
+ live in cities.
+
+21:48.360 --> 21:53.880
+ Today more than half the world lives in cities and one of the great stories of humanity in
+
+21:53.880 --> 21:57.560
+ the last 20 years has been the rural to urban migration.
+
+21:57.560 --> 22:02.720
+ This has occurred in the United States, it's occurred in Europe, it's occurring in Asia
+
+22:02.720 --> 22:04.760
+ and it's occurring in Africa.
+
+22:04.760 --> 22:09.280
+ When people move to cities, the cities get more crowded, but believe it or not their health
+
+22:09.280 --> 22:15.560
+ gets better, their productivity gets better, their IQ and educational capabilities improve,
+
+22:15.560 --> 22:19.840
+ so it's good news that people are moving to cities, but we have to make them livable
+
+22:19.840 --> 22:22.800
+ and safe.
+
+22:22.800 --> 22:28.240
+ So you, first of all, you are, but you've also worked with some of the greatest leaders
+
+22:28.240 --> 22:30.180
+ in the history of tech.
+
+22:30.180 --> 22:37.080
+ What insights do you draw from the difference in leadership styles of yourself, Steve Jobs,
+
+22:37.080 --> 22:45.320
+ Elon Musk, Larry Page, now the new CEO, Sandra Pichai and others from the, I would say, calm
+
+22:45.320 --> 22:49.600
+ sages to the mad geniuses?
+
+22:49.600 --> 22:53.880
+ One of the things that I learned as a young executive is that there's no single formula
+
+22:53.880 --> 22:56.200
+ for leadership.
+
+22:56.200 --> 23:00.080
+ They try to teach one, but that's not how it really works.
+
+23:00.080 --> 23:04.360
+ There are people who just understand what they need to do and they need to do it quickly.
+
+23:04.360 --> 23:06.800
+ Those people are often entrepreneurs.
+
+23:06.800 --> 23:09.080
+ They just know and they move fast.
+
+23:09.080 --> 23:13.400
+ There are other people who are systems thinkers and planners, that's more who I am, somewhat
+
+23:13.400 --> 23:18.760
+ more conservative, more thorough in execution, a little bit more risk averse.
+
+23:18.760 --> 23:24.120
+ There's also people who are sort of slightly insane, right, in the sense that they are
+
+23:24.120 --> 23:29.040
+ emphatic and charismatic and they feel it and they drive it and so forth.
+
+23:29.040 --> 23:31.440
+ There's no single formula to success.
+
+23:31.440 --> 23:35.320
+ There is one thing that unifies all of the people that you named, which is very high
+
+23:35.320 --> 23:41.240
+ intelligence, at the end of the day, the thing that characterizes all of them is that they
+
+23:41.240 --> 23:46.360
+ saw the world quicker, faster, they processed information faster, they didn't necessarily
+
+23:46.360 --> 23:50.160
+ make the right decisions all the time, but they were on top of it.
+
+23:50.160 --> 23:54.600
+ The other thing that's interesting about all those people is they all started young.
+
+23:54.600 --> 23:58.560
+ Think about Steve Jobs starting Apple roughly at 18 or 19.
+
+23:58.560 --> 24:01.840
+ Think about Bill Gates starting at roughly 2021.
+
+24:01.840 --> 24:07.040
+ Think about by the time they were 30, Mark Zuckerberg, a more good example at 1920.
+
+24:07.040 --> 24:13.720
+ By the time they were 30, they had 10 years, at 30 years old, they had 10 years of experience
+
+24:13.720 --> 24:19.920
+ of dealing with people and products and shipments and the press and business and so forth.
+
+24:19.920 --> 24:24.480
+ It's incredible how much experience they had compared to the rest of us who were busy getting
+
+24:24.480 --> 24:25.480
+ our PhDs.
+
+24:25.480 --> 24:26.480
+ Yes, exactly.
+
+24:26.480 --> 24:32.760
+ We should celebrate these people because they've just had more life experience and that helps
+
+24:32.760 --> 24:34.520
+ inform the judgment.
+
+24:34.520 --> 24:41.360
+ At the end of the day, when you're at the top of these organizations, all the easy questions
+
+24:41.360 --> 24:43.680
+ have been dealt with.
+
+24:43.680 --> 24:45.840
+ How should we design the buildings?
+
+24:45.840 --> 24:48.400
+ Where should we put the colors on our product?
+
+24:48.400 --> 24:51.440
+ What should the box look like?
+
+24:51.440 --> 24:55.520
+ The problems, that's why it's so interesting to be in these rooms, the problems that they
+
+24:55.520 --> 25:00.200
+ face in terms of the way they operate, the way they deal with their employees, their
+
+25:00.200 --> 25:04.160
+ customers, their innovation are profoundly challenging.
+
+25:04.160 --> 25:09.360
+ Each of the companies is demonstrably different culturally.
+
+25:09.360 --> 25:11.800
+ They are not, in fact, cut of the same.
+
+25:11.800 --> 25:16.680
+ They behave differently based on input, their internal cultures are different, their compensation
+
+25:16.680 --> 25:24.920
+ schemes are different, their values are different, so there's proof that diversity works.
+
+25:24.920 --> 25:33.440
+ So when faced with a tough decision, in need of advice, it's been said that the best thing
+
+25:33.440 --> 25:39.840
+ one can do is to find the best person in the world who can give that advice and find a
+
+25:39.840 --> 25:44.880
+ way to be in a room with them, one on one and ask.
+
+25:44.880 --> 25:51.920
+ So here we are, and let me ask in a long winded way, I wrote this down, in 1998 there were
+
+25:51.920 --> 26:01.960
+ many good search engines, Lycos, Excite, Altavista, Infoseek, AskJeeves maybe, Yahoo even.
+
+26:01.960 --> 26:07.040
+ So Google stepped in and disrupted everything, they disrupted the nature of search, the nature
+
+26:07.040 --> 26:12.040
+ of our access to information, the way we discover new knowledge.
+
+26:12.040 --> 26:19.120
+ So now it's 2018, actually 20 years later, there are many good personal AI assistants,
+
+26:19.120 --> 26:22.360
+ including of course the best from Google.
+
+26:22.360 --> 26:28.720
+ So you've spoken in medical and education, the impact of such an AI assistant could bring.
+
+26:28.720 --> 26:34.920
+ So we arrive at this question, so it's a personal one for me, but I hope my situation represents
+
+26:34.920 --> 26:41.200
+ that of many other, as we said, dreamers and the crazy engineers.
+
+26:41.200 --> 26:46.680
+ So my whole life, I've dreamed of creating such an AI assistant.
+
+26:46.680 --> 26:50.800
+ Every step I've taken has been towards that goal, now I'm a research scientist in Human
+
+26:50.800 --> 26:58.920
+ Senate AI here at MIT, so the next step for me as I sit here, facing my passion, is to
+
+26:58.920 --> 27:04.880
+ do what Larry and Sergey did in 1998, this simple start up.
+
+27:04.880 --> 27:10.640
+ And so here's my simple question, given the low odds of success, the timing and luck required,
+
+27:10.640 --> 27:14.280
+ the countless other factors that can't be controlled or predicted, which is all the
+
+27:14.280 --> 27:16.560
+ things that Larry and Sergey faced.
+
+27:16.560 --> 27:23.080
+ Is there some calculation, some strategy to follow in this step, or do you simply follow
+
+27:23.080 --> 27:26.560
+ the passion just because there's no other choice?
+
+27:26.560 --> 27:32.880
+ I think the people who are in universities are always trying to study the extraordinarily
+
+27:32.880 --> 27:37.360
+ chaotic nature of innovation and entrepreneurship.
+
+27:37.360 --> 27:42.880
+ My answer is that they didn't have that conversation, they just did it.
+
+27:42.880 --> 27:48.840
+ They sensed a moment when, in the case of Google, there was all of this data that needed
+
+27:48.840 --> 27:53.940
+ to be organized and they had a better algorithm, they had invented a better way.
+
+27:53.940 --> 28:01.040
+ So today with Human Senate AI, which is your area of research, there must be new approaches.
+
+28:01.040 --> 28:07.320
+ It's such a big field, there must be new approaches, different from what we and others are doing.
+
+28:07.320 --> 28:12.320
+ There must be startups to fund, there must be research projects to try, there must be
+
+28:12.320 --> 28:15.200
+ graduate students to work on new approaches.
+
+28:15.200 --> 28:19.120
+ Here at MIT, there are people who are looking at learning from the standpoint of looking
+
+28:19.120 --> 28:23.840
+ at child learning, how do children learn starting at each one?
+
+28:23.840 --> 28:25.560
+ And the work is fantastic.
+
+28:25.560 --> 28:30.120
+ Those approaches are different from the approach that most people are taking.
+
+28:30.120 --> 28:33.980
+ Perhaps that's a bet that you should make, or perhaps there's another one.
+
+28:33.980 --> 28:40.200
+ But at the end of the day, the successful entrepreneurs are not as crazy as they sound.
+
+28:40.200 --> 28:43.200
+ They see an opportunity based on what's happened.
+
+28:43.200 --> 28:45.400
+ Let's use Uber as an example.
+
+28:45.400 --> 28:49.840
+ As Travis sells the story, he and his cofounder were sitting in Paris and they had this idea
+
+28:49.840 --> 28:52.160
+ because they couldn't get a cab.
+
+28:52.160 --> 28:56.800
+ And they said, we have smartphones and the rest is history.
+
+28:56.800 --> 29:04.040
+ So what's the equivalent of that Travis Eiffel Tower, where is a cab moment that you could,
+
+29:04.040 --> 29:08.800
+ as an entrepreneur, take advantage of, whether it's in Human Senate AI or something else?
+
+29:08.800 --> 29:11.480
+ That's the next great start up.
+
+29:11.480 --> 29:13.760
+ And the psychology of that moment.
+
+29:13.760 --> 29:20.120
+ So when Sergey and Larry talk about, and listen to a few interviews, it's very nonchalant.
+
+29:20.120 --> 29:25.280
+ Well, here's the very fascinating web data.
+
+29:25.280 --> 29:29.080
+ And here's an algorithm we have for, you know, we just kind of want to play around with that
+
+29:29.080 --> 29:30.080
+ data.
+
+29:30.080 --> 29:32.520
+ And it seems like that's a really nice way to organize this data.
+
+29:32.520 --> 29:38.000
+ Well, I should say what happened, remember, is that they were graduate students at Stanford
+
+29:38.000 --> 29:41.320
+ and they thought this is interesting, so they built a search engine and they kept it in
+
+29:41.320 --> 29:43.400
+ their room.
+
+29:43.400 --> 29:47.520
+ And they had to get power from the room next door because they were using too much power
+
+29:47.520 --> 29:48.520
+ in the room.
+
+29:48.520 --> 29:51.640
+ So they ran an extension cord over.
+
+29:51.640 --> 29:55.360
+ And then they went and they found a house and they had Google World headquarters of
+
+29:55.360 --> 29:57.600
+ five people to start the company.
+
+29:57.600 --> 30:02.560
+ And they raised $100,000 from Andy Bechtelstein, who was the sun founder to do this, and Dave
+
+30:02.560 --> 30:04.520
+ Chariton and a few others.
+
+30:04.520 --> 30:11.960
+ The point is their beginnings were very simple, but they were based on a powerful insight.
+
+30:11.960 --> 30:14.320
+ That is a replicable model for any startup.
+
+30:14.320 --> 30:16.520
+ It has to be a powerful insight.
+
+30:16.520 --> 30:17.680
+ The beginnings are simple.
+
+30:17.680 --> 30:19.960
+ And there has to be an innovation.
+
+30:19.960 --> 30:24.280
+ In Larry and Sergey's case, it was PageRank, which was a brilliant idea, one of the most
+
+30:24.280 --> 30:26.880
+ cited papers in the world today.
+
+30:26.880 --> 30:29.880
+ What's the next one?
+
+30:29.880 --> 30:37.280
+ So you're one of, if I may say, richest people in the world, and yet it seems that money
+
+30:37.280 --> 30:43.200
+ is simply a side effect of your passions and not an inherent goal.
+
+30:43.200 --> 30:48.360
+ But it's a, you're a fascinating person to ask.
+
+30:48.360 --> 30:55.080
+ So much of our society at the individual level and at the company level and its nations is
+
+30:55.080 --> 30:58.920
+ driven by the desire for wealth.
+
+30:58.920 --> 31:01.280
+ What do you think about this drive?
+
+31:01.280 --> 31:07.000
+ And what have you learned about, if I may romanticize the notion, the meaning of life,
+
+31:07.000 --> 31:10.520
+ having achieved success on so many dimensions?
+
+31:10.520 --> 31:16.960
+ There have been many studies of human happiness and above some threshold, which is typically
+
+31:16.960 --> 31:23.600
+ relatively low for this conversation, there's no difference in happiness about money.
+
+31:23.600 --> 31:30.120
+ The happiness is correlated with meaning and purpose, a sense of family, a sense of impact.
+
+31:30.120 --> 31:34.440
+ So if you organize your life, assuming you have enough to get around and have a nice
+
+31:34.440 --> 31:40.400
+ home and so forth, you'll be far happier if you figure out what you care about and work
+
+31:40.400 --> 31:41.800
+ on that.
+
+31:41.800 --> 31:44.120
+ It's often being in service to others.
+
+31:44.120 --> 31:47.840
+ It's a great deal of evidence that people are happiest when they're serving others
+
+31:47.840 --> 31:49.640
+ and not themselves.
+
+31:49.640 --> 31:57.480
+ This goes directly against the sort of press induced excitement about powerful and wealthy
+
+31:57.480 --> 32:01.840
+ leaders of one kind and indeed, these are consequential people.
+
+32:01.840 --> 32:06.720
+ But if you are in a situation where you've been very fortunate as I have, you also have
+
+32:06.720 --> 32:12.160
+ to take that as a responsibility and you have to basically work both to educate others and
+
+32:12.160 --> 32:16.760
+ give them that opportunity, but also use that wealth to advance human society.
+
+32:16.760 --> 32:20.440
+ In my case, I'm particularly interested in using the tools of artificial intelligence
+
+32:20.440 --> 32:22.800
+ and machine learning to make society better.
+
+32:22.800 --> 32:24.000
+ I've mentioned education.
+
+32:24.000 --> 32:29.040
+ I've mentioned inequality and middle class and things like this, all of which are a passion
+
+32:29.040 --> 32:30.160
+ of mine.
+
+32:30.160 --> 32:31.920
+ It doesn't matter what you do.
+
+32:31.920 --> 32:36.560
+ It matters that you believe in it, that it's important to you and that your life will be
+
+32:36.560 --> 32:40.480
+ far more satisfying if you spend your life doing that.
+
+32:40.480 --> 32:45.320
+ I think there's no better place to end than a discussion of the meaning of life.
+
+32:45.320 --> 32:46.320
+ Eric, thank you so much.
+
+32:46.320 --> 32:47.320
+ Thank you very much.
+
+32:47.320 --> 33:16.320
+ Thank you.
+

vtt/episode_009_small.vtt

@@ -0,0 +1,2240 @@
+WEBVTT
+
+00:00.000 --> 00:05.040
+ The following is a conversation from Stuart Russell. He's a professor of computer science at UC
+
+00:05.040 --> 00:11.360
+ Berkeley and a coauthor of a book that introduced me and millions of other people to the amazing world
+
+00:11.360 --> 00:18.320
+ of AI called Artificial Intelligence The Modern Approach. So it was an honor for me to have this
+
+00:18.320 --> 00:24.480
+ conversation as part of MIT course on artificial journal intelligence and the artificial intelligence
+
+00:24.480 --> 00:30.800
+ podcast. If you enjoy it, please subscribe on YouTube, iTunes or your podcast provider of choice
+
+00:31.360 --> 00:37.600
+ or simply connect with me on Twitter at Lex Freedman spelled F R I D. And now here's my
+
+00:37.600 --> 00:46.160
+ conversation with Stuart Russell. So you've mentioned in 1975 in high school you've created
+
+00:46.160 --> 00:54.160
+ one of your first AI programs that played chess. Were you ever able to build a program that
+
+00:54.160 --> 01:02.080
+ beat you at chess or another board game? So my program never beat me at chess.
+
+01:03.520 --> 01:10.480
+ I actually wrote the program at Imperial College. So I used to take the bus every Wednesday with a
+
+01:10.480 --> 01:17.200
+ box of cards this big and shove them into the card reader and they gave us eight seconds of CPU time.
+
+01:17.200 --> 01:24.720
+ It took about five seconds to read the cards in and compile the code. So we had three seconds of
+
+01:24.720 --> 01:30.960
+ CPU time, which was enough to make one move, you know, with a not very deep search. And then we
+
+01:30.960 --> 01:34.960
+ would print that move out and then we'd have to go to the back of the queue and wait to feed the
+
+01:34.960 --> 01:40.480
+ cards in again. How deep was the search? Well, are we talking about two moves? So no, I think we've
+
+01:40.480 --> 01:48.000
+ got we got an eight move, eight, you know, depth eight with alpha beta. And we had some tricks of
+
+01:48.000 --> 01:54.480
+ our own about move ordering and some pruning of the tree. And we were still able to beat that
+
+01:54.480 --> 02:00.960
+ program. Yeah, yeah, I was a reasonable chess player in my youth. I did an Othello program
+
+02:01.680 --> 02:05.920
+ and a backgammon program. So when I got to Berkeley, I worked a lot on
+
+02:05.920 --> 02:12.560
+ what we call meta reasoning, which really means reasoning about reasoning. And in the case of
+
+02:13.200 --> 02:18.320
+ a game playing program, you need to reason about what parts of the search tree you're actually
+
+02:18.320 --> 02:23.440
+ going to explore, because the search tree is enormous, you know, bigger than the number of
+
+02:23.440 --> 02:30.960
+ atoms in the universe. And the way programs succeed and the way humans succeed is by only
+
+02:30.960 --> 02:36.160
+ looking at a small fraction of the search tree. And if you look at the right fraction, you play
+
+02:36.160 --> 02:41.360
+ really well. If you look at the wrong fraction, if you waste your time thinking about things that
+
+02:41.360 --> 02:45.840
+ are never going to happen, the moves that no one's ever going to make, then you're going to lose,
+
+02:45.840 --> 02:53.760
+ because you won't be able to figure out the right decision. So that question of how machines can
+
+02:53.760 --> 02:59.760
+ manage their own computation, how they decide what to think about is the meta reasoning question.
+
+02:59.760 --> 03:05.920
+ We developed some methods for doing that. And very simply, a machine should think about
+
+03:06.640 --> 03:11.920
+ whatever thoughts are going to improve its decision quality. We were able to show that
+
+03:12.640 --> 03:18.240
+ both for a fellow, which is a standard two player game, and for backgammon, which includes
+
+03:19.040 --> 03:24.000
+ dice rolls, so it's a two player game with uncertainty. For both of those cases, we could
+
+03:24.000 --> 03:30.480
+ come up with algorithms that were actually much more efficient than the standard alpha beta search,
+
+03:31.120 --> 03:36.560
+ which chess programs at the time were using. And that those programs could beat me.
+
+03:38.080 --> 03:44.720
+ And I think you can see the same basic ideas in AlphaGo and AlphaZero today.
+
+03:44.720 --> 03:52.560
+ The way they explore the tree is using a form of meta reasoning to select what to think about
+
+03:52.560 --> 03:57.840
+ based on how useful it is to think about it. Is there any insights you can describe
+
+03:57.840 --> 04:03.040
+ without Greek symbols of how do we select which paths to go down?
+
+04:04.240 --> 04:10.560
+ There's really two kinds of learning going on. So as you say, AlphaGo learns to evaluate board
+
+04:10.560 --> 04:17.680
+ to evaluate board position. So it can look at a go board. And it actually has probably a super
+
+04:17.680 --> 04:25.760
+ human ability to instantly tell how promising that situation is. To me, the amazing thing about
+
+04:25.760 --> 04:34.560
+ AlphaGo is not that it can be the world champion with its hands tied behind his back. But the fact that
+
+04:34.560 --> 04:41.360
+ if you stop it from searching altogether, so you say, okay, you're not allowed to do
+
+04:41.360 --> 04:46.480
+ any thinking ahead. You can just consider each of your legal moves and then look at the
+
+04:47.120 --> 04:53.280
+ resulting situation and evaluate it. So what we call a depth one search. So just the immediate
+
+04:53.280 --> 04:57.920
+ outcome of your moves and decide if that's good or bad. That version of AlphaGo
+
+04:57.920 --> 05:05.200
+ can still play at a professional level. And human professionals are sitting there for
+
+05:05.200 --> 05:13.440
+ five, 10 minutes deciding what to do. And AlphaGo in less than a second can instantly intuit what
+
+05:13.440 --> 05:18.880
+ is the right move to make based on its ability to evaluate positions. And that is remarkable
+
+05:19.760 --> 05:26.080
+ because we don't have that level of intuition about go. We actually have to think about the
+
+05:26.080 --> 05:35.200
+ situation. So anyway, that capability that AlphaGo has is one big part of why it beats humans.
+
+05:35.840 --> 05:44.560
+ The other big part is that it's able to look ahead 40, 50, 60 moves into the future. And
+
+05:46.880 --> 05:51.200
+ if it was considering all possibilities, 40 or 50 or 60 moves into the future,
+
+05:51.200 --> 06:02.240
+ that would be 10 to the 200 possibilities. So way more than atoms in the universe and so on.
+
+06:02.240 --> 06:09.680
+ So it's very, very selective about what it looks at. So let me try to give you an intuition about
+
+06:10.880 --> 06:15.360
+ how you decide what to think about. It's a combination of two things. One is
+
+06:15.360 --> 06:22.000
+ how promising it is. So if you're already convinced that a move is terrible,
+
+06:22.560 --> 06:26.560
+ there's no point spending a lot more time convincing yourself that it's terrible.
+
+06:27.520 --> 06:33.600
+ Because it's probably not going to change your mind. So the real reason you think is because
+
+06:33.600 --> 06:39.920
+ there's some possibility of changing your mind about what to do. And is that changing your mind
+
+06:39.920 --> 06:46.000
+ that would result then in a better final action in the real world. So that's the purpose of thinking
+
+06:46.800 --> 06:53.520
+ is to improve the final action in the real world. And so if you think about a move that is guaranteed
+
+06:53.520 --> 06:58.000
+ to be terrible, you can convince yourself it's terrible, you're still not going to change your
+
+06:58.000 --> 07:04.320
+ mind. But on the other hand, suppose you had a choice between two moves, one of them you've
+
+07:04.320 --> 07:10.400
+ already figured out is guaranteed to be a draw, let's say. And then the other one looks a little
+
+07:10.400 --> 07:14.000
+ bit worse. Like it looks fairly likely that if you make that move, you're going to lose.
+
+07:14.640 --> 07:20.720
+ But there's still some uncertainty about the value of that move. There's still some possibility
+
+07:20.720 --> 07:25.920
+ that it will turn out to be a win. Then it's worth thinking about that. So even though it's
+
+07:25.920 --> 07:31.280
+ less promising on average than the other move, which is guaranteed to be a draw, there's still
+
+07:31.280 --> 07:36.160
+ some purpose in thinking about it because there's a chance that you'll change your mind and discover
+
+07:36.160 --> 07:42.080
+ that in fact it's a better move. So it's a combination of how good the move appears to be
+
+07:42.080 --> 07:48.000
+ and how much uncertainty there is about its value. The more uncertainty, the more it's worth thinking
+
+07:48.000 --> 07:52.800
+ about because there's a higher upside if you want to think of it that way. And of course in the
+
+07:52.800 --> 07:59.920
+ beginning, especially in the AlphaGo Zero formulation, it's everything is shrouded in
+
+07:59.920 --> 08:06.240
+ uncertainty. So you're really swimming in a sea of uncertainty. So it benefits you to
+
+08:07.520 --> 08:11.120
+ I mean, actually falling in the same process as you described, but because you're so uncertain
+
+08:11.120 --> 08:15.280
+ about everything, you basically have to try a lot of different directions.
+
+08:15.280 --> 08:22.400
+ Yeah. So the early parts of the search tree are fairly bushy that it will look at a lot
+
+08:22.400 --> 08:27.840
+ of different possibilities, but fairly quickly, the degree of certainty about some of the moves.
+
+08:27.840 --> 08:31.760
+ I mean, if a move is really terrible, you'll pretty quickly find out, right? You'll lose
+
+08:31.760 --> 08:37.200
+ half your pieces or half your territory. And then you'll say, okay, this is not worth thinking
+
+08:37.200 --> 08:45.280
+ about anymore. And then so further down, the tree becomes very long and narrow. And you're following
+
+08:45.280 --> 08:54.800
+ various lines of play, 10, 20, 30, 40, 50 moves into the future. And that again is something
+
+08:54.800 --> 09:01.920
+ that human beings have a very hard time doing mainly because they just lack the short term memory.
+
+09:02.480 --> 09:09.440
+ You just can't remember a sequence of moves. That's 50 moves long. And you can't imagine
+
+09:09.440 --> 09:15.520
+ the board correctly for that many moves into the future. Of course, the top players,
+
+09:16.400 --> 09:19.280
+ I'm much more familiar with chess, but the top players probably have,
+
+09:19.280 --> 09:26.480
+ they have echoes of the same kind of intuition instinct that in a moment's time, AlphaGo applies
+
+09:27.280 --> 09:31.760
+ when they see a board. I mean, they've seen those patterns, human beings have seen those
+
+09:31.760 --> 09:37.680
+ patterns before at the top, at the grandmaster level. It seems that there is some
+
+09:40.000 --> 09:45.920
+ similarities, or maybe it's our imagination creates a vision of those similarities, but it
+
+09:45.920 --> 09:53.120
+ feels like this kind of pattern recognition that the AlphaGo approaches are using is similar to
+
+09:53.120 --> 10:00.560
+ what human beings at the top level are using. I think there's some truth to that.
+
+10:01.520 --> 10:08.960
+ But not entirely. Yeah, I mean, I think the extent to which a human grandmaster can reliably
+
+10:10.080 --> 10:13.680
+ instantly recognize the right move and instantly recognize the values of position.
+
+10:13.680 --> 10:19.120
+ I think that's a little bit overrated. But if you sacrifice a queen, for example,
+
+10:19.120 --> 10:23.840
+ I mean, there's these, there's these beautiful games of chess with Bobby Fisher, somebody where
+
+10:24.640 --> 10:32.720
+ it's seeming to make a bad move. And I'm not sure there's a perfect degree of calculation
+
+10:32.720 --> 10:37.440
+ involved where they've calculated all the possible things that happen. But there's an
+
+10:37.440 --> 10:45.440
+ instinct there, right, that somehow adds up to. Yeah, so I think what happens is you get a sense
+
+10:45.440 --> 10:51.680
+ that there's some possibility in the position, even if you make a weird looking move, that it
+
+10:51.680 --> 11:05.120
+ opens up some lines of calculation that otherwise would be definitely bad. And it's that intuition
+
+11:05.120 --> 11:13.920
+ that there's something here in this position that might might yield a win down the set. And then
+
+11:13.920 --> 11:20.640
+ you follow that. Right. And in some sense, when a chess player is following a line in his or her
+
+11:20.640 --> 11:27.120
+ mind, they're they mentally simulating what the other person is going to do, what the opponent
+
+11:27.120 --> 11:33.680
+ is going to do. And they can do that as long as the moves are kind of forced, right, as long as
+
+11:33.680 --> 11:39.440
+ there's a we call a forcing variation where the opponent doesn't really have much choice how to
+
+11:39.440 --> 11:45.200
+ respond. And then you see if you can force them into a situation where you win. We see plenty
+
+11:45.200 --> 11:53.520
+ of mistakes, even in grandmaster games, where they just miss some simple three, four, five move
+
+11:54.560 --> 12:00.400
+ combination that wasn't particularly apparent in the position, but was still there.
+
+12:00.400 --> 12:07.360
+ That's the thing that makes us human. Yeah. So when you mentioned that in Othello, those games
+
+12:07.360 --> 12:13.760
+ were after some meta reasoning improvements and research was able to beat you. How did that make
+
+12:13.760 --> 12:21.280
+ you feel part of the meta reasoning capability that it had was based on learning. And,
+
+12:23.280 --> 12:28.160
+ and you could sit down the next day and you could just feel that it had got a lot smarter.
+
+12:28.160 --> 12:33.280
+ You know, and all of a sudden, you really felt like you're sort of pressed against
+
+12:34.480 --> 12:40.800
+ the wall because it was it was much more aggressive and was totally unforgiving of any
+
+12:40.800 --> 12:47.760
+ minor mistake that you might make. And actually, it seemed understood the game better than I did.
+
+12:47.760 --> 12:55.520
+ And Gary Kasparov has this quote where during his match against Deep Blue, he said he suddenly
+
+12:55.520 --> 13:01.680
+ felt that there was a new kind of intelligence across the board. Do you think that's a scary or
+
+13:01.680 --> 13:10.240
+ an exciting possibility for Kasparov and for yourself in the context of chess purely sort of
+
+13:10.240 --> 13:16.720
+ in this like that feeling, whatever that is, I think it's definitely an exciting feeling.
+
+13:17.600 --> 13:23.680
+ You know, this is what made me work on AI in the first place was as soon as I really understood
+
+13:23.680 --> 13:30.080
+ what a computer was, I wanted to make it smart. You know, I started out with the first program I
+
+13:30.080 --> 13:38.640
+ wrote was for the Sinclair Programmable Calculator. And I think you could write a 21 step algorithm.
+
+13:38.640 --> 13:44.160
+ That was the biggest program you could write something like that and do little arithmetic
+
+13:44.160 --> 13:49.440
+ calculations. So I think I implemented Newton's method for square roots and a few other things
+
+13:49.440 --> 13:56.640
+ like that. But then, you know, I thought, okay, if I just had more space, I could make this thing
+
+13:56.640 --> 14:10.560
+ intelligent. And so I started thinking about AI. And I think the thing that's scary is not the
+
+14:10.560 --> 14:19.520
+ chess program, because you know, chess programs, they're not in the taking over the world business.
+
+14:19.520 --> 14:29.440
+ But if you extrapolate, you know, there are things about chess that don't resemble the real
+
+14:29.440 --> 14:37.600
+ world, right? We know, we know the rules of chess. The chess board is completely visible
+
+14:37.600 --> 14:43.280
+ to the program where, of course, the real world is not. Most the real world is not visible from
+
+14:43.280 --> 14:52.400
+ wherever you're sitting, so to speak. And to overcome those kinds of problems, you need
+
+14:52.400 --> 14:58.240
+ qualitatively different algorithms. Another thing about the real world is that, you know, we
+
+14:58.240 --> 15:07.520
+ we regularly plan ahead on the timescales involving billions or trillions of steps. Now,
+
+15:07.520 --> 15:13.760
+ we don't plan those in detail. But, you know, when you choose to do a PhD at Berkeley,
+
+15:14.800 --> 15:20.480
+ that's a five year commitment that amounts to about a trillion motor control steps that you
+
+15:20.480 --> 15:26.160
+ will eventually be committed to. Including going up the stairs, opening doors,
+
+15:26.160 --> 15:32.880
+ a drinking water type. Yeah, I mean, every every finger movement while you're typing every character
+
+15:32.880 --> 15:37.280
+ of every paper and the thesis and everything. So you're not committing in advance to the specific
+
+15:37.280 --> 15:43.760
+ motor control steps, but you're still reasoning on a timescale that will eventually reduce to
+
+15:44.400 --> 15:50.000
+ trillions of motor control actions. And so for all these reasons,
+
+15:50.000 --> 15:58.160
+ you know, AlphaGo and Deep Blue and so on don't represent any kind of threat to humanity. But
+
+15:58.160 --> 16:08.320
+ they are a step towards it, right? And progress in AI occurs by essentially removing one by one
+
+16:08.320 --> 16:14.640
+ these assumptions that make problems easy, like the assumption of complete observability
+
+16:14.640 --> 16:21.120
+ of the situation, right? We remove that assumption, you need a much more complicated kind of computing
+
+16:22.160 --> 16:26.000
+ design and you need something that actually keeps track of all the things you can't see
+
+16:26.000 --> 16:31.920
+ and tries to estimate what's going on. And there's inevitable uncertainty in that. So it becomes a
+
+16:31.920 --> 16:38.160
+ much more complicated problem. But, you know, we are removing those assumptions, we are starting to
+
+16:38.160 --> 16:44.400
+ have algorithms that can cope with much longer timescales, cope with uncertainty that can cope
+
+16:44.400 --> 16:53.360
+ with partial observability. And so each of those steps sort of magnifies by a thousand the range
+
+16:53.360 --> 16:58.400
+ of things that we can do with AI systems. So the way I started in AI, I wanted to be a psychiatrist
+
+16:58.400 --> 17:03.840
+ for a long time and understand the mind in high school, and of course program and so on. And I
+
+17:03.840 --> 17:10.640
+ showed up University of Illinois to an AI lab and they said, okay, I don't have time for you, but here
+
+17:10.640 --> 17:18.480
+ is a book, AI Modern Approach, I think it was the first edition at the time. Here, go learn this.
+
+17:18.480 --> 17:23.120
+ And I remember the lay of the land was, well, it's incredible that we solved chess, but we'll
+
+17:23.120 --> 17:30.480
+ never solve go. I mean, it was pretty certain that go in the way we thought about systems that reason
+
+17:31.520 --> 17:36.080
+ wasn't possible to solve. And now we've solved it. So it's a very... Well, I think I would have said
+
+17:36.080 --> 17:44.080
+ that it's unlikely we could take the kind of algorithm that was used for chess and just get
+
+17:44.080 --> 17:55.680
+ it to scale up and work well for go. And at the time, what we thought was that in order to solve
+
+17:55.680 --> 18:01.600
+ go, we would have to do something similar to the way humans manage the complexity of go,
+
+18:01.600 --> 18:06.960
+ which is to break it down into kind of sub games. So when a human thinks about a go board,
+
+18:06.960 --> 18:12.480
+ they think about different parts of the board as sort of weakly connected to each other.
+
+18:12.480 --> 18:16.880
+ And they think about, okay, within this part of the board, here's how things could go.
+
+18:16.880 --> 18:20.560
+ In that part of board, here's how things could go. And then you try to sort of couple those
+
+18:20.560 --> 18:26.400
+ two analyses together and deal with the interactions and maybe revise your views of how things are
+
+18:26.400 --> 18:32.160
+ going to go in each part. And then you've got maybe five, six, seven, 10 parts of the board. And
+
+18:33.440 --> 18:40.640
+ that actually resembles the real world much more than chess does. Because in the real world,
+
+18:41.440 --> 18:49.200
+ we have work, we have home life, we have sport, whatever different kinds of activities, shopping,
+
+18:49.200 --> 18:57.040
+ these all are connected to each other, but they're weakly connected. So when I'm typing a paper,
+
+18:58.480 --> 19:03.600
+ I don't simultaneously have to decide which order I'm going to get the milk and the butter.
+
+19:04.400 --> 19:09.760
+ That doesn't affect the typing. But I do need to realize, okay, better finish this
+
+19:10.320 --> 19:14.080
+ before the shops close because I don't have anything, I don't have any food at home.
+
+19:14.080 --> 19:20.560
+ So there's some weak connection, but not in the way that chess works, where everything is tied
+
+19:20.560 --> 19:27.600
+ into a single stream of thought. So the thought was that go to solve go would have to make progress
+
+19:27.600 --> 19:31.600
+ on stuff that would be useful for the real world. And in a way, AlphaGo is a little bit disappointing
+
+19:32.480 --> 19:38.160
+ because the program designed for AlphaGo is actually not that different from
+
+19:38.160 --> 19:45.840
+ from Deep Blue or even from Arthur Samuel's Jacob playing program from the 1950s.
+
+19:48.160 --> 19:54.560
+ And in fact, the two things that make AlphaGo work is one is this amazing ability to evaluate
+
+19:54.560 --> 19:59.200
+ the positions. And the other is the meta reasoning capability, which allows it to
+
+19:59.200 --> 20:06.960
+ to explore some paths in the tree very deeply and to abandon other paths very quickly.
+
+20:06.960 --> 20:14.640
+ So this word meta reasoning, while technically correct, inspires perhaps the wrong
+
+20:16.000 --> 20:21.360
+ degree of power that AlphaGo has, for example, the word reasoning is a powerful word. So let me
+
+20:21.360 --> 20:29.840
+ ask you sort of, do you were part of the symbolic AI world for a while, like where AI was, there's
+
+20:29.840 --> 20:38.960
+ a lot of excellent interesting ideas there that unfortunately met a winter. And so do you think
+
+20:38.960 --> 20:46.800
+ it reemerges? Oh, so I would say, yeah, it's not quite as simple as that. So the AI winter,
+
+20:46.800 --> 20:54.400
+ the first winter that was actually named as such was the one in the late 80s.
+
+20:56.400 --> 21:00.880
+ And that came about because in the mid 80s, there was
+
+21:03.280 --> 21:10.480
+ really a concerted attempt to push AI out into the real world using what was called
+
+21:10.480 --> 21:17.280
+ expert system technology. And for the most part, that technology was just not ready for prime
+
+21:17.280 --> 21:27.200
+ time. They were trying in many cases to do a form of uncertain reasoning, judgment combinations of
+
+21:27.200 --> 21:34.640
+ evidence diagnosis, those kinds of things, which was simply invalid. And when you try to apply
+
+21:34.640 --> 21:40.960
+ invalid reasoning methods to real problems, you can fudge it for small versions of the problem.
+
+21:40.960 --> 21:47.200
+ But when it starts to get larger, the thing just falls apart. So many companies found that
+
+21:49.040 --> 21:53.440
+ the stuff just didn't work. And they were spending tons of money on consultants to
+
+21:53.440 --> 21:59.600
+ try to make it work. And there were other practical reasons, like they were asking
+
+21:59.600 --> 22:07.760
+ the companies to buy incredibly expensive Lisp machine workstations, which were literally
+
+22:07.760 --> 22:17.680
+ between $50,000 and $100,000 in 1980s money, which would be between $150,000 and $300,000 per
+
+22:17.680 --> 22:24.000
+ workstation in current prices. Then the bottom line, they weren't seeing a profit from it.
+
+22:24.000 --> 22:29.840
+ Yeah. In many cases, I think there were some successes. There's no doubt about that. But
+
+22:30.880 --> 22:37.760
+ people, I would say, over invested. Every major company was starting an AI department just like
+
+22:37.760 --> 22:45.840
+ now. And I worry a bit that we might see similar disappointments, not because the
+
+22:45.840 --> 22:57.600
+ current technology is invalid, but it's limited in its scope. And it's almost the dual of the
+
+22:57.600 --> 23:03.360
+ scope problems that expert systems had. What have you learned from that hype cycle? And
+
+23:03.360 --> 23:09.760
+ what can we do to prevent another winter, for example? Yeah. So when I'm giving talks these
+
+23:09.760 --> 23:17.520
+ days, that's one of the warnings that I give. So there's two part warning slide. One is that
+
+23:18.480 --> 23:24.000
+ rather than data being the new oil, data is the new snake oil. That's a good line. And then
+
+23:26.000 --> 23:35.440
+ the other is that we might see a very visible failure in some of the major application areas.
+
+23:35.440 --> 23:42.400
+ And I think self driving cars would be the flagship. And I think
+
+23:43.600 --> 23:48.560
+ when you look at the history, so the first self driving car was on the freeway,
+
+23:51.200 --> 24:00.400
+ driving itself, changing lanes, overtaking in 1987. And so it's more than 30 years.
+
+24:00.400 --> 24:06.720
+ And that kind of looks like where we are today, right? Prototypes on the freeway,
+
+24:06.720 --> 24:13.760
+ changing lanes and overtaking. Now, I think significant progress has been made, particularly
+
+24:13.760 --> 24:20.560
+ on the perception side. So we worked a lot on autonomous vehicles in the early, mid 90s at
+
+24:20.560 --> 24:29.040
+ Berkeley. And we had our own big demonstrations. We put congressmen into self driving cars and
+
+24:29.040 --> 24:36.000
+ had them zooming along the freeway. And the problem was clearly perception.
+
+24:37.520 --> 24:42.880
+ At the time, the problem was perception. Yeah. So in simulation, with perfect perception,
+
+24:42.880 --> 24:47.200
+ you could actually show that you can drive safely for a long time, even if the other cars
+
+24:47.200 --> 24:55.360
+ are misbehaving and so on. But simultaneously, we worked on machine vision for detecting cars and
+
+24:55.360 --> 25:03.040
+ tracking pedestrians and so on. And we couldn't get the reliability of detection and tracking
+
+25:03.040 --> 25:11.440
+ up to a high enough level, particularly in bad weather conditions, nighttime rainfall.
+
+25:11.440 --> 25:16.000
+ Good enough for demos, but perhaps not good enough to cover the general operation.
+
+25:16.000 --> 25:20.800
+ Yeah. So the thing about driving is, so suppose you're a taxi driver and you drive every day,
+
+25:20.800 --> 25:27.360
+ eight hours a day for 10 years, that's 100 million seconds of driving. And any one of those
+
+25:27.360 --> 25:33.280
+ seconds, you can make a fatal mistake. So you're talking about eight nines of reliability.
+
+25:34.960 --> 25:43.840
+ Now, if your vision system only detects 98.3% of the vehicles, that's sort of one
+
+25:43.840 --> 25:52.720
+ on a bit nine reliability. So you have another seven orders of magnitude to go. And this is
+
+25:52.720 --> 25:57.920
+ what people don't understand. They think, oh, because I had a successful demo, I'm pretty much
+
+25:57.920 --> 26:07.440
+ done. But you're not even within seven orders of magnitude of being done. And that's the difficulty.
+
+26:07.440 --> 26:14.320
+ And it's not, can I follow a white line? That's not the problem. We follow a white line all the
+
+26:14.320 --> 26:22.160
+ way across the country. But it's the weird stuff that happens. It's all the edge cases. Yeah.
+
+26:22.160 --> 26:30.640
+ The edge case, other drivers doing weird things. So if you talk to Google, so they had actually
+
+26:30.640 --> 26:36.560
+ a very classical architecture where you had machine vision, which would detect all the
+
+26:36.560 --> 26:41.920
+ other cars and pedestrians and the white lines and the road signs. And then basically,
+
+26:42.480 --> 26:49.680
+ that was fed into a logical database. And then you had a classical 1970s rule based expert system
+
+26:52.000 --> 26:55.680
+ telling you, okay, if you're in the middle lane, and there's a bicyclist in the right lane,
+
+26:55.680 --> 27:03.040
+ who is signaling this, then then do that, right? And what they found was that every day that go
+
+27:03.040 --> 27:07.760
+ out and there'd be another situation that the rules didn't cover. So they come to a traffic
+
+27:07.760 --> 27:11.680
+ circle and there's a little girl riding her bicycle the wrong way around the traffic circle.
+
+27:11.680 --> 27:17.520
+ Okay, what do you do? We don't have a rule. Oh my God. Okay, stop. And then they come back
+
+27:17.520 --> 27:24.400
+ and add more rules. And they just found that this was not really converging. And if you think about
+
+27:24.400 --> 27:31.280
+ it, right, how do you deal with an unexpected situation, meaning one that you've never previously
+
+27:31.280 --> 27:37.200
+ encountered and the sort of the reasoning required to figure out the solution for that
+
+27:37.200 --> 27:42.800
+ situation has never been done. It doesn't match any previous situation in terms of the kind of
+
+27:42.800 --> 27:49.520
+ reasoning you have to do. Well, in chess programs, this happens all the time. You're constantly
+
+27:49.520 --> 27:54.560
+ coming up with situations you haven't seen before. And you have to reason about them and you have
+
+27:54.560 --> 27:59.840
+ to think about, okay, here are the possible things I could do. Here are the outcomes. Here's how
+
+27:59.840 --> 28:04.560
+ desirable the outcomes are and then pick the right one. In the 90s, we were saying, okay,
+
+28:04.560 --> 28:08.160
+ this is how you're going to have to do automated vehicles. They're going to have to have a look
+
+28:08.160 --> 28:14.400
+ ahead capability. But the look ahead for driving is more difficult than it is for chess. Because
+
+28:14.400 --> 28:20.720
+ of humans. Right, there's humans and they're less predictable than chess pieces. Well,
+
+28:20.720 --> 28:28.240
+ then you have an opponent in chess who's also somewhat unpredictable. But for example, in chess,
+
+28:28.240 --> 28:33.600
+ you always know the opponent's intention. They're trying to beat you. Whereas in driving, you don't
+
+28:33.600 --> 28:39.040
+ know, is this guy trying to turn left or has he just forgotten to turn off his turn signal? Or is
+
+28:39.040 --> 28:45.680
+ he drunk? Or is he changing the channel on his radio or whatever it might be, you got to try and
+
+28:45.680 --> 28:52.560
+ figure out the mental state, the intent of the other drivers to forecast the possible evolutions
+
+28:52.560 --> 28:58.160
+ of their trajectories. And then you got to figure out, okay, which is the trajectory for me that's
+
+28:58.160 --> 29:04.000
+ going to be safest. And those all interact with each other because the other drivers are going
+
+29:04.000 --> 29:09.120
+ to react to your trajectory and so on. So, you know, they've got the classic merging onto the
+
+29:09.120 --> 29:14.640
+ freeway problem where you're kind of racing a vehicle that's already on the freeway and you're
+
+29:14.640 --> 29:17.680
+ are you going to pull ahead of them or are you going to let them go first and pull in behind
+
+29:17.680 --> 29:23.680
+ and you get this sort of uncertainty about who's going first. So all those kinds of things
+
+29:23.680 --> 29:34.720
+ mean that you need a decision making architecture that's very different from either a rule based
+
+29:34.720 --> 29:41.360
+ system or it seems to me a kind of an end to end neural network system. You know, so just as Alpha
+
+29:41.360 --> 29:47.360
+ Go is pretty good when it doesn't do any look ahead, but it's way, way, way, way better when it does.
+
+29:47.360 --> 29:52.720
+ I think the same is going to be true for driving. You can have a driving system that's pretty good
+
+29:54.080 --> 29:59.280
+ when it doesn't do any look ahead, but that's not good enough. You know, and we've already seen
+
+29:59.920 --> 30:07.440
+ multiple deaths caused by poorly designed machine learning algorithms that don't really
+
+30:07.440 --> 30:13.600
+ understand what they're doing. Yeah, and on several levels, I think it's on the perception side,
+
+30:13.600 --> 30:19.520
+ there's mistakes being made by those algorithms where the perception is very shallow on the
+
+30:19.520 --> 30:26.720
+ planning side, the look ahead, like you said, and the thing that we come up against that's
+
+30:28.560 --> 30:32.080
+ really interesting when you try to deploy systems in the real world is
+
+30:33.280 --> 30:37.680
+ you can't think of an artificial intelligence system as a thing that responds to the world always.
+
+30:38.320 --> 30:41.600
+ You have to realize that it's an agent that others will respond to as well.
+
+30:41.600 --> 30:47.200
+ Well, so in order to drive successfully, you can't just try to do obstacle avoidance.
+
+30:47.840 --> 30:51.520
+ You can't pretend that you're invisible, right? You're the invisible car.
+
+30:52.400 --> 30:57.280
+ It doesn't work that way. I mean, but you have to assert, yet others have to be scared of you,
+
+30:57.280 --> 31:04.160
+ just there's this tension, there's this game. So we study a lot of work with pedestrians.
+
+31:04.160 --> 31:09.360
+ If you approach pedestrians as purely an obstacle avoidance, so you're doing look
+
+31:09.360 --> 31:15.040
+ ahead as in modeling the intent, they're not going to take advantage of you.
+
+31:15.040 --> 31:20.080
+ They're not going to respect you at all. There has to be a tension, a fear, some amount of
+
+31:20.080 --> 31:26.720
+ uncertainty. That's how we have created. Or at least just a kind of a resoluteness.
+
+31:28.000 --> 31:32.000
+ You have to display a certain amount of resoluteness. You can't be too tentative.
+
+31:32.000 --> 31:42.480
+ Yeah. So the solutions then become pretty complicated. You get into game theoretic
+
+31:42.480 --> 31:50.960
+ analyses. So at Berkeley now, we're working a lot on this kind of interaction between machines
+
+31:50.960 --> 32:03.600
+ and humans. And that's exciting. So my colleague, Anka Dragan, actually, if you formulate the problem
+
+32:03.600 --> 32:08.800
+ game theoretically and you just let the system figure out the solution, it does interesting,
+
+32:08.800 --> 32:16.640
+ unexpected things. Like sometimes at a stop sign, if no one is going first, the car will
+
+32:16.640 --> 32:23.200
+ actually back up a little. It's just to indicate to the other cars that they should go. And that's
+
+32:23.200 --> 32:28.480
+ something it invented entirely by itself. That's interesting. We didn't say this is the language
+
+32:28.480 --> 32:36.240
+ of communication at stop signs. It figured it out. That's really interesting. So let me one just
+
+32:36.240 --> 32:42.960
+ step back for a second. Just this beautiful philosophical notion. So Pamela McCordick in
+
+32:42.960 --> 32:50.320
+ 1979 wrote AI began with the ancient wish to forge the gods. So when you think about the
+
+32:50.320 --> 32:57.520
+ history of our civilization, do you think that there is an inherent desire to create,
+
+32:58.960 --> 33:05.680
+ let's not say gods, but to create superintelligence? Is it inherent to us? Is it in our genes,
+
+33:05.680 --> 33:13.680
+ that the natural arc of human civilization is to create things that are of greater and greater
+
+33:13.680 --> 33:21.680
+ power and perhaps echoes of ourselves? So to create the gods, as Pamela said.
+
+33:21.680 --> 33:34.160
+ It may be. I mean, we're all individuals, but certainly we see over and over again in history
+
+33:35.760 --> 33:42.320
+ individuals who thought about this possibility. Hopefully, I'm not being too philosophical here.
+
+33:42.320 --> 33:48.560
+ But if you look at the arc of this, where this is going and we'll talk about AI safety,
+
+33:48.560 --> 33:55.840
+ we'll talk about greater and greater intelligence, do you see that when you created the Othello
+
+33:55.840 --> 34:01.680
+ program and you felt this excitement, what was that excitement? Was it the excitement of a tinkerer
+
+34:01.680 --> 34:10.240
+ who created something cool, like a clock? Or was there a magic, or was it more like a child being
+
+34:10.240 --> 34:17.520
+ born? Yeah. So I mean, I certainly understand that viewpoint. And if you look at the light
+
+34:17.520 --> 34:26.640
+ hill report, so in the 70s, there was a lot of controversy in the UK about AI and whether it
+
+34:26.640 --> 34:34.720
+ was for real and how much the money the government should invest. So it's a long story, but the
+
+34:34.720 --> 34:43.280
+ government commissioned a report by Lighthill, who was a physicist, and he wrote a very damning
+
+34:43.280 --> 34:53.920
+ report about AI, which I think was the point. And he said that these are frustrated men who
+
+34:54.480 --> 35:05.760
+ unable to have children would like to create life as a kind of replacement, which I think is
+
+35:05.760 --> 35:21.600
+ really pretty unfair. But there is a kind of magic, I would say, when you build something
+
+35:25.680 --> 35:29.760
+ and what you're building in is really just you're building in some understanding of the
+
+35:29.760 --> 35:37.120
+ principles of learning and decision making. And to see those principles actually then
+
+35:37.840 --> 35:47.920
+ turn into intelligent behavior in specific situations, it's an incredible thing. And
+
+35:47.920 --> 35:58.480
+ that is naturally going to make you think, okay, where does this end?
+
+36:00.080 --> 36:08.240
+ And so there's a there's magical, optimistic views of word and whatever your view of optimism is,
+
+36:08.240 --> 36:13.360
+ whatever your view of utopia is, it's probably different for everybody. But you've often talked
+
+36:13.360 --> 36:26.080
+ about concerns you have of how things might go wrong. So I've talked to Max Tegmark. There's a
+
+36:26.080 --> 36:33.360
+ lot of interesting ways to think about AI safety. You're one of the seminal people thinking about
+
+36:33.360 --> 36:39.360
+ this problem amongst sort of being in the weeds of actually solving specific AI problems,
+
+36:39.360 --> 36:44.080
+ you're also thinking about the big picture of where we're going. So can you talk about
+
+36:44.080 --> 36:49.200
+ several elements of it? Let's just talk about maybe the control problem. So this idea of
+
+36:50.800 --> 36:58.720
+ losing ability to control the behavior of our AI system. So how do you see that? How do you see
+
+36:58.720 --> 37:04.480
+ that coming about? What do you think we can do to manage it?
+
+37:04.480 --> 37:11.520
+ Well, so it doesn't take a genius to realize that if you make something that's smarter than you,
+
+37:11.520 --> 37:20.320
+ you might have a problem. Alan Turing wrote about this and gave lectures about this,
+
+37:21.600 --> 37:32.480
+ 1951. He did a lecture on the radio. And he basically says, once the machine thinking method
+
+37:32.480 --> 37:45.600
+ starts, very quickly, they'll outstrip humanity. And if we're lucky, we might be able to turn off
+
+37:45.600 --> 37:52.160
+ the power at strategic moments, but even so, our species would be humbled. And actually,
+
+37:52.160 --> 37:56.240
+ I think it was wrong about that. If it's a sufficiently intelligent machine, it's not
+
+37:56.240 --> 38:00.160
+ going to let you switch it off. It's actually in competition with you.
+
+38:00.160 --> 38:05.840
+ So what do you think is meant just for a quick tangent if we shut off this
+
+38:05.840 --> 38:08.800
+ super intelligent machine that our species would be humbled?
+
+38:11.840 --> 38:20.560
+ I think he means that we would realize that we are inferior, that we only survive by the skin
+
+38:20.560 --> 38:27.440
+ of our teeth because we happen to get to the off switch just in time. And if we hadn't,
+
+38:27.440 --> 38:34.400
+ then we would have lost control over the earth. So are you more worried when you think about
+
+38:34.400 --> 38:41.600
+ this stuff about super intelligent AI or are you more worried about super powerful AI that's not
+
+38:41.600 --> 38:49.760
+ aligned with our values? So the paperclip scenarios kind of... I think so the main problem I'm
+
+38:49.760 --> 38:58.960
+ working on is the control problem, the problem of machines pursuing objectives that are, as you
+
+38:58.960 --> 39:06.720
+ say, not aligned with human objectives. And this has been the way we've thought about AI
+
+39:06.720 --> 39:15.120
+ since the beginning. You build a machine for optimizing and then you put in some objective
+
+39:15.120 --> 39:25.520
+ and it optimizes. And we can think of this as the king Midas problem. Because if the king Midas
+
+39:26.480 --> 39:32.640
+ put in this objective, everything I touch should turn to gold and the gods, that's like the machine,
+
+39:32.640 --> 39:39.360
+ they said, okay, done. You now have this power and of course his food and his drink and his family
+
+39:39.360 --> 39:50.080
+ all turned to gold and then he dies of misery and starvation. It's a warning, it's a failure mode that
+
+39:50.080 --> 39:56.160
+ pretty much every culture in history has had some story along the same lines. There's the
+
+39:56.160 --> 40:01.920
+ genie that gives you three wishes and third wish is always, please undo the first two wishes because
+
+40:01.920 --> 40:11.920
+ I messed up. And when Arthur Samuel wrote his checker playing program, which learned to play
+
+40:11.920 --> 40:16.800
+ checkers considerably better than Arthur Samuel could play and actually reached a pretty decent
+
+40:16.800 --> 40:25.040
+ standard, Norbert Wiener, who was one of the major mathematicians of the 20th century, he's sort of
+
+40:25.040 --> 40:32.560
+ the father of modern automation control systems. He saw this and he basically extrapolated
+
+40:33.360 --> 40:43.680
+ as Turing did and said, okay, this is how we could lose control. And specifically that
+
+40:45.520 --> 40:50.960
+ we have to be certain that the purpose we put into the machine is the purpose which we really
+
+40:50.960 --> 40:59.760
+ desire. And the problem is, we can't do that. Right. You mean we're not, it's a very difficult
+
+40:59.760 --> 41:05.440
+ to encode, to put our values on paper is really difficult, or you're just saying it's impossible?
+
+41:09.120 --> 41:15.360
+ The line is great between the two. So theoretically, it's possible, but in practice,
+
+41:15.360 --> 41:23.520
+ it's extremely unlikely that we could specify correctly in advance the full range of concerns
+
+41:23.520 --> 41:29.360
+ of humanity. You talked about cultural transmission of values, I think is how humans to human
+
+41:29.360 --> 41:36.320
+ transmission of values happens, right? Well, we learn, yeah, I mean, as we grow up, we learn about
+
+41:36.320 --> 41:42.640
+ the values that matter, how things should go, what is reasonable to pursue and what isn't
+
+41:42.640 --> 41:47.920
+ reasonable to pursue. I think machines can learn in the same kind of way. Yeah. So I think that
+
+41:49.120 --> 41:54.480
+ what we need to do is to get away from this idea that you build an optimizing machine and then you
+
+41:54.480 --> 42:03.200
+ put the objective into it. Because if it's possible that you might put in a wrong objective, and we
+
+42:03.200 --> 42:08.880
+ already know this is possible because it's happened lots of times, right? That means that the machine
+
+42:08.880 --> 42:17.760
+ should never take an objective that's given as gospel truth. Because once it takes the objective
+
+42:17.760 --> 42:26.800
+ as gospel truth, then it believes that whatever actions it's taking in pursuit of that objective
+
+42:26.800 --> 42:31.200
+ are the correct things to do. So you could be jumping up and down and saying, no, no, no, no,
+
+42:31.200 --> 42:36.480
+ you're going to destroy the world, but the machine knows what the true objective is and is pursuing
+
+42:36.480 --> 42:42.640
+ it and tough luck to you. And this is not restricted to AI, right? This is, I think,
+
+42:43.360 --> 42:48.880
+ many of the 20th century technologies, right? So in statistics, you minimize a loss function,
+
+42:48.880 --> 42:54.320
+ the loss function is exogenously specified in control theory, you minimize a cost function,
+
+42:54.320 --> 42:59.840
+ in operations research, you maximize a reward function, and so on. So in all these disciplines,
+
+42:59.840 --> 43:08.560
+ this is how we conceive of the problem. And it's the wrong problem. Because we cannot specify
+
+43:08.560 --> 43:15.360
+ with certainty the correct objective, right? We need uncertainty, we need the machine to be
+
+43:15.360 --> 43:19.440
+ uncertain about what it is that it's supposed to be maximizing.
+
+43:19.440 --> 43:25.200
+ It's my favorite idea of yours. I've heard you say somewhere, well, I shouldn't pick favorites,
+
+43:25.200 --> 43:32.640
+ but it just sounds beautiful. We need to teach machines humility. It's a beautiful way to put
+
+43:32.640 --> 43:40.320
+ it. I love it. That they're humble. They know that they don't know what it is they're supposed
+
+43:40.320 --> 43:48.240
+ to be doing. And that those objectives, I mean, they exist, they're within us, but we may not
+
+43:48.240 --> 43:56.160
+ be able to explicate them. We may not even know how we want our future to go.
+
+43:57.040 --> 44:06.800
+ So exactly. And a machine that's uncertain is going to be differential to us. So if we say,
+
+44:06.800 --> 44:11.840
+ don't do that, well, now the machines learn something a bit more about our true objectives,
+
+44:11.840 --> 44:16.480
+ because something that it thought was reasonable in pursuit of our objective,
+
+44:16.480 --> 44:20.800
+ it turns out not to be so now it's learned something. So it's going to defer because it
+
+44:20.800 --> 44:30.240
+ wants to be doing what we really want. And that point, I think, is absolutely central
+
+44:30.240 --> 44:37.920
+ to solving the control problem. And it's a different kind of AI when you take away this
+
+44:37.920 --> 44:44.560
+ idea that the objective is known, then in fact, a lot of the theoretical frameworks that we're so
+
+44:44.560 --> 44:53.520
+ familiar with, you know, mark off decision processes, goal based planning, you know,
+
+44:53.520 --> 44:59.280
+ standard game research, all of these techniques actually become inapplicable.
+
+45:01.040 --> 45:11.120
+ And you get a more complicated problem because because now the interaction with the human becomes
+
+45:11.120 --> 45:20.400
+ part of the problem. Because the human by making choices is giving you more information about
+
+45:21.280 --> 45:25.360
+ the true objective and that information helps you achieve the objective better.
+
+45:26.640 --> 45:32.000
+ And so that really means that you're mostly dealing with game theoretic problems where you've
+
+45:32.000 --> 45:38.000
+ got the machine and the human and they're coupled together, rather than a machine going off by itself
+
+45:38.000 --> 45:43.600
+ with a fixed objective. Which is fascinating on the machine and the human level that we,
+
+45:44.400 --> 45:51.920
+ when you don't have an objective means you're together coming up with an objective. I mean,
+
+45:51.920 --> 45:56.160
+ there's a lot of philosophy that, you know, you could argue that life doesn't really have meaning.
+
+45:56.160 --> 46:01.680
+ We we together agree on what gives it meaning and we kind of culturally create
+
+46:01.680 --> 46:08.560
+ things that give why the heck we are in this earth anyway. We together as a society create
+
+46:08.560 --> 46:13.680
+ that meaning and you have to learn that objective. And one of the biggest, I thought that's where
+
+46:13.680 --> 46:19.200
+ you were going to go for a second. One of the biggest troubles we run into outside of statistics
+
+46:19.200 --> 46:26.240
+ and machine learning and AI in just human civilization is when you look at, I came from,
+
+46:26.240 --> 46:32.160
+ I was born in the Soviet Union. And the history of the 20th century, we ran into the most trouble,
+
+46:32.160 --> 46:40.160
+ us humans, when there was a certainty about the objective. And you do whatever it takes to achieve
+
+46:40.160 --> 46:46.480
+ that objective, whether you're talking about Germany or communist Russia, you get into trouble
+
+46:46.480 --> 46:52.960
+ with humans. And I would say with corporations, in fact, some people argue that we don't have
+
+46:52.960 --> 46:57.840
+ to look forward to a time when AI systems take over the world, they already have. And they call
+
+46:57.840 --> 47:04.880
+ corporations, right? That corporations happen to be using people as components right now.
+
+47:05.920 --> 47:11.680
+ But they are effectively algorithmic machines, and they're optimizing an objective, which is
+
+47:11.680 --> 47:18.080
+ quarterly profit that isn't aligned with overall well being of the human race. And they are
+
+47:18.080 --> 47:24.160
+ destroying the world. They are primarily responsible for our inability to tackle climate change.
+
+47:24.960 --> 47:30.400
+ So I think that's one way of thinking about what's going on with corporations. But
+
+47:31.840 --> 47:39.680
+ I think the point you're making is valid, that there are many systems in the real world where
+
+47:39.680 --> 47:48.480
+ we've sort of prematurely fixed on the objective and then decoupled the machine from those that
+
+47:48.480 --> 47:54.720
+ are supposed to be serving. And I think you see this with government, right? Government is supposed
+
+47:54.720 --> 48:02.720
+ to be a machine that serves people. But instead, it tends to be taken over by people who have their
+
+48:02.720 --> 48:08.160
+ own objective and use government to optimize that objective, regardless of what people want.
+
+48:08.160 --> 48:16.080
+ Do you find appealing the idea of almost arguing machines where you have multiple AI systems with
+
+48:16.080 --> 48:22.400
+ a clear fixed objective? We have in government the red team and the blue team that are very fixed
+
+48:22.400 --> 48:28.240
+ on their objectives. And they argue, and it kind of maybe would disagree, but it kind of seems to
+
+48:28.240 --> 48:39.440
+ make it work somewhat that the duality of it, okay, let's go 100 years back when there was still
+
+48:39.440 --> 48:44.480
+ was going on or at the founding of this country, there was disagreements and that disagreement is
+
+48:44.480 --> 48:52.160
+ where so it was a balance between certainty and forced humility because the power was distributed.
+
+48:52.160 --> 49:05.280
+ Yeah, I think that the nature of debate and disagreement argument takes as a premise the idea
+
+49:05.280 --> 49:12.800
+ that you could be wrong, which means that you're not necessarily absolutely convinced that your
+
+49:12.800 --> 49:19.520
+ objective is the correct one. If you were absolutely convinced, there'd be no point
+
+49:19.520 --> 49:24.160
+ in having any discussion or argument because you would never change your mind. And there wouldn't
+
+49:24.160 --> 49:32.080
+ be any sort of synthesis or anything like that. So I think you can think of argumentation as an
+
+49:32.080 --> 49:44.640
+ implementation of a form of uncertain reasoning. I've been reading recently about utilitarianism
+
+49:44.640 --> 49:54.960
+ and the history of efforts to define in a sort of clear mathematical way a if you like a formula for
+
+49:54.960 --> 50:02.320
+ moral or political decision making. And it's really interesting that the parallels between
+
+50:02.320 --> 50:08.720
+ the philosophical discussions going back 200 years and what you see now in discussions about
+
+50:08.720 --> 50:15.040
+ existential risk because it's almost exactly the same. So someone would say, okay, well,
+
+50:15.040 --> 50:21.600
+ here's a formula for how we should make decisions. So utilitarianism is roughly each person has a
+
+50:21.600 --> 50:27.680
+ utility function and then we make decisions to maximize the sum of everybody's utility.
+
+50:28.720 --> 50:36.480
+ And then people point out, well, in that case, the best policy is one that leads to
+
+50:36.480 --> 50:42.480
+ the enormously vast population, all of whom are living a life that's barely worth living.
+
+50:43.520 --> 50:50.640
+ And this is called the repugnant conclusion. And another version is that we should maximize
+
+50:51.200 --> 50:57.680
+ pleasure and that's what we mean by utility. And then you'll get people effectively saying,
+
+50:57.680 --> 51:02.480
+ well, in that case, we might as well just have everyone hooked up to a heroin drip. And they
+
+51:02.480 --> 51:08.720
+ didn't use those words. But that debate was happening in the 19th century, as it is now
+
+51:09.920 --> 51:17.600
+ about AI, that if we get the formula wrong, we're going to have AI systems working towards
+
+51:17.600 --> 51:22.080
+ an outcome that in retrospect, would be exactly wrong.
+
+51:22.080 --> 51:26.960
+ Do you think there's has beautifully put so the echoes are there. But do you think,
+
+51:26.960 --> 51:34.640
+ I mean, if you look at Sam Harris, our imagination worries about the AI version of that, because
+
+51:34.640 --> 51:44.640
+ of the speed at which the things going wrong in the utilitarian context could happen.
+
+51:45.840 --> 51:47.280
+ Is that a worry for you?
+
+51:47.280 --> 51:55.360
+ Yeah, I think that in most cases, not in all, but if we have a wrong political idea,
+
+51:55.360 --> 52:01.200
+ we see it starting to go wrong. And we're not completely stupid. And so we said, okay,
+
+52:02.000 --> 52:10.160
+ maybe that was a mistake. Let's try something different. And also, we're very slow and inefficient
+
+52:10.160 --> 52:15.520
+ about implementing these things and so on. So you have to worry when you have corporations
+
+52:15.520 --> 52:20.800
+ or political systems that are extremely efficient. But when we look at AI systems,
+
+52:20.800 --> 52:27.840
+ or even just computers in general, right, they have this different characteristic
+
+52:28.400 --> 52:35.040
+ from ordinary human activity in the past. So let's say you were a surgeon. You had some idea
+
+52:35.040 --> 52:40.480
+ about how to do some operation, right? Well, and let's say you were wrong, right, that that way
+
+52:40.480 --> 52:45.840
+ of doing the operation would mostly kill the patient. Well, you'd find out pretty quickly,
+
+52:45.840 --> 52:56.000
+ like after three, maybe three or four tries, right? But that isn't true for pharmaceutical
+
+52:56.000 --> 53:03.040
+ companies, because they don't do three or four operations. They manufacture three or four billion
+
+53:03.040 --> 53:08.800
+ pills and they sell them. And then they find out maybe six months or a year later that, oh,
+
+53:08.800 --> 53:14.880
+ people are dying of heart attacks or getting cancer from this drug. And so that's why we have the FDA,
+
+53:14.880 --> 53:22.960
+ right? Because of the scalability of pharmaceutical production. And there have been some unbelievably
+
+53:22.960 --> 53:34.320
+ bad episodes in the history of pharmaceuticals and adulteration of products and so on that have
+
+53:34.320 --> 53:37.520
+ killed tens of thousands or paralyzed hundreds of thousands of people.
+
+53:39.360 --> 53:43.280
+ Now, with computers, we have that same scalability problem that you can
+
+53:43.280 --> 53:49.520
+ sit there and type for i equals one to five billion, two, right? And all of a sudden,
+
+53:49.520 --> 53:55.360
+ you're having an impact on a global scale. And yet we have no FDA, right? There's absolutely no
+
+53:55.360 --> 54:02.480
+ controls at all over what a bunch of undergraduates with too much caffeine can do to the world.
+
+54:03.440 --> 54:09.600
+ And, you know, we look at what happened with Facebook, well, social media in general, and
+
+54:09.600 --> 54:18.480
+ click through optimization. So you have a simple feedback algorithm that's trying to just optimize
+
+54:18.480 --> 54:24.080
+ click through, right? That sounds reasonable, right? Because you don't want to be feeding people
+
+54:24.080 --> 54:33.200
+ ads that they don't care about or not interested in. And you might even think of that process as
+
+54:33.200 --> 54:42.160
+ simply adjusting the the feeding of ads or news articles or whatever it might be to match people's
+
+54:42.160 --> 54:50.000
+ preferences, right? Which sounds like a good idea. But in fact, that isn't how the algorithm works,
+
+54:50.880 --> 54:59.760
+ right? You make more money. The algorithm makes more money. If it can better predict what people
+
+54:59.760 --> 55:06.400
+ are going to click on, because then it can feed them exactly that, right? So the way to maximize
+
+55:06.400 --> 55:13.360
+ click through is actually to modify the people, to make them more predictable. And one way to do
+
+55:13.360 --> 55:21.280
+ that is to feed them information which will change their behavior and preferences towards
+
+55:21.920 --> 55:27.600
+ extremes that make them predictable. Whatever is the nearest extreme or the nearest predictable
+
+55:27.600 --> 55:33.840
+ point, that's where you're going to end up. And the machines will force you there.
+
+55:34.480 --> 55:40.400
+ Now, and I think there's a reasonable argument to say that this, among other things, is
+
+55:40.400 --> 55:48.880
+ contributing to the destruction of democracy in the world. And where was the oversight
+
+55:50.160 --> 55:55.600
+ of this process? Where were the people saying, okay, you would like to apply this algorithm to
+
+55:55.600 --> 56:01.120
+ five billion people on the face of the earth? Can you show me that it's safe? Can you show
+
+56:01.120 --> 56:07.040
+ me that it won't have various kinds of negative effects? No, there was no one asking that question.
+
+56:07.040 --> 56:14.800
+ There was no one placed between the undergrads with too much caffeine and the human race.
+
+56:15.520 --> 56:20.480
+ It's just they just did it. And some way outside the scope of my knowledge,
+
+56:20.480 --> 56:27.120
+ so economists would argue that the invisible hand, so the capitalist system, it was the
+
+56:27.120 --> 56:32.480
+ oversight. So if you're going to corrupt society with whatever decision you make as a company,
+
+56:32.480 --> 56:38.640
+ then that's going to be reflected in people not using your product. That's one model of oversight.
+
+56:39.280 --> 56:48.000
+ We shall see. But in the meantime, but you might even have broken the political system
+
+56:48.000 --> 56:54.960
+ that enables capitalism to function. Well, you've changed it. So we should see. Yeah.
+
+56:54.960 --> 57:01.360
+ Change is often painful. So my question is absolutely, it's fascinating. You're absolutely
+
+57:01.360 --> 57:07.840
+ right that there was zero oversight on algorithms that can have a profound civilization changing
+
+57:09.120 --> 57:15.440
+ effect. So do you think it's possible? I mean, I haven't, have you seen government?
+
+57:15.440 --> 57:22.800
+ So do you think it's possible to create regulatory bodies oversight over AI algorithms,
+
+57:22.800 --> 57:28.400
+ which are inherently such cutting edge set of ideas and technologies?
+
+57:30.960 --> 57:37.520
+ Yeah, but I think it takes time to figure out what kind of oversight, what kinds of controls.
+
+57:37.520 --> 57:42.960
+ I mean, it took time to design the FDA regime. Some people still don't like it and they want
+
+57:42.960 --> 57:50.400
+ to fix it. And I think there are clear ways that it could be improved. But the whole notion that
+
+57:50.400 --> 57:56.400
+ you have stage one, stage two, stage three, and here are the criteria for what you have to do
+
+57:56.400 --> 58:02.000
+ to pass a stage one trial, right? We haven't even thought about what those would be for algorithms.
+
+58:02.000 --> 58:10.320
+ So I mean, I think there are, there are things we could do right now with regard to bias, for
+
+58:10.320 --> 58:19.040
+ example, we have a pretty good technical handle on how to detect algorithms that are propagating
+
+58:19.040 --> 58:26.720
+ bias that exists in data sets, how to debias those algorithms, and even what it's going to cost you
+
+58:26.720 --> 58:34.480
+ to do that. So I think we could start having some standards on that. I think there are things to do
+
+58:34.480 --> 58:41.840
+ with impersonation and falsification that we could, we could work on. So I think, yeah.
+
+58:43.360 --> 58:50.240
+ Or the very simple point. So impersonation is a machine acting as if it was a person.
+
+58:51.440 --> 58:58.800
+ I can't see a real justification for why we shouldn't insist that machine self identify as
+
+58:58.800 --> 59:08.480
+ machines. Where is the social benefit in fooling people into thinking that this is really a person
+
+59:08.480 --> 59:15.120
+ when it isn't? I don't mind if it uses a human like voice that's easy to understand. That's fine.
+
+59:15.120 --> 59:22.000
+ But it should just say, I'm a machine in some form. And now many people are speaking to that.
+
+59:22.800 --> 59:26.400
+ I would think relatively obvious facts. So I think most people... Yeah. I mean,
+
+59:26.400 --> 59:32.400
+ there is actually a law in California that bans impersonation, but only in certain
+
+59:33.280 --> 59:41.280
+ restricted circumstances. So for the purpose of engaging in a fraudulent transaction and for
+
+59:41.280 --> 59:48.160
+ the purpose of modifying someone's voting behavior. So those are the circumstances where
+
+59:48.160 --> 59:55.520
+ machines have to self identify. But I think, arguably, it should be in all circumstances.
+
+59:56.400 --> 1:00:03.280
+ And then when you talk about deep fakes, we're just at the beginning. But already,
+
+1:00:03.280 --> 1:00:10.480
+ it's possible to make a movie of anybody saying anything in ways that are pretty hard to detect.
+
+1:00:11.520 --> 1:00:15.040
+ Including yourself because you're on camera now and your voice is coming through with high
+
+1:00:15.040 --> 1:00:19.360
+ resolution. Yeah. So you could take what I'm saying and replace it with pretty much anything
+
+1:00:19.360 --> 1:00:24.400
+ else you wanted me to be saying. And even it would change my lips and facial expressions to fit.
+
+1:00:26.960 --> 1:00:35.280
+ And there's actually not much in the way of real legal protection against that.
+
+1:00:35.920 --> 1:00:38.640
+ I think in the commercial area, you could say, yeah, that's...
+
+1:00:38.640 --> 1:00:46.000
+ You're using my brand and so on. There are rules about that. But in the political sphere, I think,
+
+1:00:47.600 --> 1:00:52.560
+ at the moment, it's anything goes. So that could be really, really damaging.
+
+1:00:53.920 --> 1:01:02.400
+ And let me just try to make not an argument, but try to look back at history and say something
+
+1:01:02.400 --> 1:01:09.680
+ dark, in essence, is while regulation seems to be... Oversight seems to be exactly the
+
+1:01:09.680 --> 1:01:14.480
+ right thing to do here. It seems that human beings, what they naturally do is they wait
+
+1:01:14.480 --> 1:01:20.080
+ for something to go wrong. If you're talking about nuclear weapons, you can't talk about
+
+1:01:20.080 --> 1:01:26.000
+ nuclear weapons being dangerous until somebody actually, like the United States drops the bomb,
+
+1:01:26.000 --> 1:01:34.960
+ or Chernobyl melting. Do you think we will have to wait for things going wrong in a way that's
+
+1:01:34.960 --> 1:01:39.840
+ obviously damaging to society, not an existential risk, but obviously damaging?
+
+1:01:42.320 --> 1:01:48.000
+ Or do you have faith that... I hope not. But I mean, I think we do have to look at history.
+
+1:01:48.000 --> 1:01:57.600
+ So the two examples you gave, nuclear weapons and nuclear power, are very, very interesting because
+
+1:01:59.520 --> 1:02:07.840
+ nuclear weapons, we knew in the early years of the 20th century that atoms contained a huge
+
+1:02:07.840 --> 1:02:13.280
+ amount of energy. We had E equals MC squared. We knew the mass differences between the different
+
+1:02:13.280 --> 1:02:20.640
+ atoms and their components, and we knew that you might be able to make an incredibly powerful
+
+1:02:20.640 --> 1:02:28.000
+ explosive. So H.G. Wells wrote science fiction book, I think, in 1912. Frederick Soddy, who was the
+
+1:02:28.000 --> 1:02:35.760
+ guy who discovered isotopes as a Nobel Prize winner, he gave a speech in 1915 saying that
+
+1:02:37.840 --> 1:02:42.160
+ one pound of this new explosive would be the equivalent of 150 tons of dynamite,
+
+1:02:42.160 --> 1:02:50.720
+ which turns out to be about right. And this was in World War I, so he was imagining how much worse
+
+1:02:51.360 --> 1:02:57.600
+ the world war would be if we were using that kind of explosive. But the physics establishment
+
+1:02:57.600 --> 1:03:05.760
+ simply refused to believe that these things could be made. Including the people who were making it.
+
+1:03:06.400 --> 1:03:11.280
+ Well, so they were doing the nuclear physics. I mean, eventually were the ones who made it.
+
+1:03:11.280 --> 1:03:21.280
+ You talk about Fermi or whoever. Well, so up to the development was mostly theoretical. So it was
+
+1:03:21.280 --> 1:03:28.320
+ people using sort of primitive kinds of particle acceleration and doing experiments at the level
+
+1:03:28.320 --> 1:03:36.480
+ of single particles or collections of particles. They weren't yet thinking about how to actually
+
+1:03:36.480 --> 1:03:40.160
+ make a bomb or anything like that. But they knew the energy was there and they figured if they
+
+1:03:40.160 --> 1:03:46.720
+ understood it better, it might be possible. But the physics establishment, their view, and I think
+
+1:03:46.720 --> 1:03:51.840
+ because they did not want it to be true, their view was that it could not be true.
+
+1:03:53.360 --> 1:04:00.240
+ That this could not provide a way to make a super weapon. And there was this famous
+
+1:04:01.120 --> 1:04:08.240
+ speech given by Rutherford, who was the sort of leader of nuclear physics. And it was on
+
+1:04:08.240 --> 1:04:14.800
+ September 11, 1933. And he said, you know, anyone who talks about the possibility of
+
+1:04:14.800 --> 1:04:21.600
+ obtaining energy from transformation of atoms is talking complete moonshine. And the next
+
+1:04:22.800 --> 1:04:28.560
+ morning, Leo Zillard read about that speech and then invented the nuclear chain reaction.
+
+1:04:28.560 --> 1:04:35.920
+ And so as soon as he invented, as soon as he had that idea, that you could make a chain reaction
+
+1:04:35.920 --> 1:04:40.640
+ with neutrons because neutrons were not repelled by the nucleus so they could enter the nucleus
+
+1:04:40.640 --> 1:04:48.480
+ and then continue the reaction. As soon as he has that idea, he instantly realized that the world
+
+1:04:48.480 --> 1:04:58.160
+ was in deep doo doo. Because this is 1933, right? Hitler had recently come to power in Germany.
+
+1:04:58.160 --> 1:05:09.280
+ Zillard was in London. He eventually became a refugee and he came to the US. And in the
+
+1:05:09.280 --> 1:05:14.880
+ process of having the idea about the chain reaction, he figured out basically how to make
+
+1:05:14.880 --> 1:05:22.800
+ a bomb and also how to make a reactor. And he patented the reactor in 1934. But because
+
+1:05:22.800 --> 1:05:28.480
+ of the situation, the great power conflict situation that he could see happening,
+
+1:05:29.200 --> 1:05:38.160
+ he kept that a secret. And so between then and the beginning of World War II,
+
+1:05:39.600 --> 1:05:48.000
+ people were working, including the Germans, on how to actually create neutron sources,
+
+1:05:48.000 --> 1:05:54.720
+ what specific fission reactions would produce neutrons of the right energy to continue the
+
+1:05:54.720 --> 1:06:02.480
+ reaction. And that was demonstrated in Germany, I think in 1938, if I remember correctly. The first
+
+1:06:03.760 --> 1:06:17.280
+ nuclear weapon patent was 1939 by the French. So this was actually going on well before World War
+
+1:06:17.280 --> 1:06:22.720
+ II really got going. And then the British probably had the most advanced capability
+
+1:06:22.720 --> 1:06:27.920
+ in this area. But for safety reasons, among others, and bless just sort of just resources,
+
+1:06:28.720 --> 1:06:33.520
+ they moved the program from Britain to the US. And then that became Manhattan Project.
+
+1:06:34.480 --> 1:06:37.920
+ So the reason why we couldn't
+
+1:06:37.920 --> 1:06:48.320
+ have any kind of oversight of nuclear weapons and nuclear technology was because we were basically
+
+1:06:48.320 --> 1:06:57.520
+ already in an arms race in a war. But you mentioned in the 20s and 30s, so what are the echoes?
+
+1:07:00.000 --> 1:07:04.320
+ The way you've described the story, I mean, there's clearly echoes. What do you think most AI
+
+1:07:04.320 --> 1:07:11.440
+ researchers, folks who are really close to the metal, they really are not concerned about AI,
+
+1:07:11.440 --> 1:07:16.960
+ they don't think about it, whether it's they don't want to think about it. But what are the,
+
+1:07:16.960 --> 1:07:23.760
+ yeah, why do you think that is? What are the echoes of the nuclear situation to the current AI
+
+1:07:23.760 --> 1:07:33.120
+ situation? And what can we do about it? I think there is a kind of motivated cognition, which is
+
+1:07:33.120 --> 1:07:40.640
+ a term in psychology means that you believe what you would like to be true, rather than what is
+
+1:07:40.640 --> 1:07:50.240
+ true. And it's unsettling to think that what you're working on might be the end of the human race,
+
+1:07:50.800 --> 1:07:58.400
+ obviously. So you would rather instantly deny it and come up with some reason why it couldn't be
+
+1:07:58.400 --> 1:08:07.440
+ true. And I collected a long list of regions that extremely intelligent, competent AI scientists
+
+1:08:08.080 --> 1:08:16.560
+ have come up with for why we shouldn't worry about this. For example, calculators are superhuman at
+
+1:08:16.560 --> 1:08:21.680
+ arithmetic and they haven't taken over the world, so there's nothing to worry about. Well, okay,
+
+1:08:21.680 --> 1:08:29.440
+ my five year old could have figured out why that was an unreasonable and really quite weak argument.
+
+1:08:31.520 --> 1:08:40.800
+ Another one was, while it's theoretically possible that you could have superhuman AI
+
+1:08:41.760 --> 1:08:46.480
+ destroy the world, it's also theoretically possible that a black hole could materialize
+
+1:08:46.480 --> 1:08:51.760
+ right next to the earth and destroy humanity. I mean, yes, it's theoretically possible,
+
+1:08:51.760 --> 1:08:56.720
+ quantum theoretically, extremely unlikely that it would just materialize right there.
+
+1:08:58.400 --> 1:09:04.720
+ But that's a completely bogus analogy because if the whole physics community on earth was working
+
+1:09:04.720 --> 1:09:11.680
+ to materialize a black hole in near earth orbit, wouldn't you ask them, is that a good idea? Is
+
+1:09:11.680 --> 1:09:19.040
+ that going to be safe? What if you succeed? And that's the thing. The AI community is sort of
+
+1:09:19.040 --> 1:09:26.240
+ refused to ask itself, what if you succeed? And initially, I think that was because it was too
+
+1:09:26.240 --> 1:09:35.520
+ hard, but Alan Turing asked himself that and he said, we'd be toast. If we were lucky, we might
+
+1:09:35.520 --> 1:09:40.000
+ be able to switch off the power but probably we'd be toast. But there's also an aspect
+
+1:09:40.000 --> 1:09:49.680
+ that because we're not exactly sure what the future holds, it's not clear exactly so technically
+
+1:09:49.680 --> 1:09:58.480
+ what to worry about, sort of how things go wrong. And so there is something it feels like, maybe
+
+1:09:58.480 --> 1:10:04.400
+ you can correct me if I'm wrong, but there's something paralyzing about worrying about something
+
+1:10:04.400 --> 1:10:10.000
+ that logically is inevitable. But you don't really know what that will look like.
+
+1:10:10.720 --> 1:10:19.440
+ Yeah, I think that's a reasonable point. And it's certainly in terms of existential risks,
+
+1:10:19.440 --> 1:10:26.800
+ it's different from asteroid collides with the earth, which again is quite possible. It's
+
+1:10:26.800 --> 1:10:33.040
+ happened in the past. It'll probably happen again. We don't know right now. But if we did detect an
+
+1:10:33.040 --> 1:10:39.200
+ asteroid that was going to hit the earth in 75 years time, we'd certainly be doing something
+
+1:10:39.200 --> 1:10:43.600
+ about it. Well, it's clear there's got big rock and we'll probably have a meeting and see what
+
+1:10:43.600 --> 1:10:49.040
+ do we do about the big rock with AI. Right, with AI. I mean, there are very few people who think it's
+
+1:10:49.040 --> 1:10:54.400
+ not going to happen within the next 75 years. I know Rod Brooks doesn't think it's going to happen.
+
+1:10:55.200 --> 1:11:00.880
+ Maybe Andrew Ng doesn't think it's happened. But a lot of the people who work day to day,
+
+1:11:00.880 --> 1:11:07.360
+ you know, as you say, at the rock face, they think it's going to happen. I think the median
+
+1:11:08.640 --> 1:11:14.320
+ estimate from AI researchers is somewhere in 40 to 50 years from now. Or maybe, you know,
+
+1:11:14.320 --> 1:11:20.000
+ I think in Asia, they think it's going to be even faster than that. I'm a little bit
+
+1:11:21.280 --> 1:11:25.840
+ more conservative. I think it probably take longer than that. But I think, you know, as
+
+1:11:25.840 --> 1:11:32.400
+ happened with nuclear weapons, it can happen overnight that you have these breakthroughs.
+
+1:11:32.400 --> 1:11:38.240
+ And we need more than one breakthrough. But, you know, it's on the order of half a dozen.
+
+1:11:38.800 --> 1:11:43.920
+ This is a very rough scale. But so half a dozen breakthroughs of that nature
+
+1:11:45.840 --> 1:11:53.600
+ would have to happen for us to reach superhuman AI. But the AI research community is
+
+1:11:53.600 --> 1:12:00.640
+ vast now, the massive investments from governments, from corporations, tons of really,
+
+1:12:00.640 --> 1:12:05.760
+ really smart people. You just have to look at the rate of progress in different areas of AI
+
+1:12:05.760 --> 1:12:10.800
+ to see that things are moving pretty fast. So to say, oh, it's just going to be thousands of years.
+
+1:12:11.920 --> 1:12:18.160
+ I don't see any basis for that. You know, I see, you know, for example, the
+
+1:12:18.160 --> 1:12:28.640
+ Stanford 100 year AI project, which is supposed to be sort of, you know, the serious establishment view,
+
+1:12:29.520 --> 1:12:33.440
+ their most recent report actually said it's probably not even possible.
+
+1:12:34.160 --> 1:12:34.720
+ Oh, wow.
+
+1:12:35.280 --> 1:12:42.960
+ Right. Which if you want a perfect example of people in denial, that's it. Because, you know,
+
+1:12:42.960 --> 1:12:49.760
+ for the whole history of AI, we've been saying to philosophers who said it wasn't possible. Well,
+
+1:12:49.760 --> 1:12:53.520
+ you have no idea what you're talking about. Of course, it's possible. Right. Give me an
+
+1:12:53.520 --> 1:12:59.760
+ argument for why it couldn't happen. And there isn't one. Right. And now, because people are
+
+1:12:59.760 --> 1:13:04.000
+ worried that maybe AI might get a bad name, or I just don't want to think about this,
+
+1:13:05.200 --> 1:13:09.520
+ they're saying, okay, well, of course, it's not really possible. You know, imagine, right? Imagine
+
+1:13:09.520 --> 1:13:16.000
+ if, you know, the leaders of the cancer biology community got up and said, well, you know, of
+
+1:13:16.000 --> 1:13:23.680
+ course, curing cancer, it's not really possible. There'd be a complete outrage and dismay. And,
+
+1:13:25.040 --> 1:13:30.800
+ you know, I find this really a strange phenomenon. So,
+
+1:13:30.800 --> 1:13:38.640
+ okay, so if you accept it as possible, and if you accept that it's probably going to happen,
+
+1:13:40.560 --> 1:13:44.400
+ the point that you're making that, you know, how does it go wrong?
+
+1:13:46.320 --> 1:13:51.600
+ A valid question without that, without an answer to that question, then you're stuck with what I
+
+1:13:51.600 --> 1:13:56.160
+ call the gorilla problem, which is, you know, the problem that the gorillas face, right? They
+
+1:13:56.160 --> 1:14:02.800
+ made something more intelligent than them, namely us a few million years ago, and now they're in
+
+1:14:02.800 --> 1:14:09.360
+ deep doo doo. So there's really nothing they can do. They've lost the control. They failed to solve
+
+1:14:09.360 --> 1:14:16.240
+ the control problem of controlling humans. And so they've lost. So we don't want to be in that
+
+1:14:16.240 --> 1:14:22.320
+ situation. And if the gorilla problem is the only formulation you have, there's not a lot you can do.
+
+1:14:22.320 --> 1:14:28.240
+ Right. Other than to say, okay, we should try to stop. You know, we should just not make the humans
+
+1:14:28.240 --> 1:14:33.120
+ or, or in this case, not make the AI. And I think that's really hard to do.
+
+1:14:35.120 --> 1:14:42.480
+ To, I'm not actually proposing that that's a feasible course of action. And I also think that,
+
+1:14:43.040 --> 1:14:46.080
+ you know, if properly controlled AI could be incredibly beneficial.
+
+1:14:46.080 --> 1:14:54.960
+ So the, but it seems to me that there's a, there's a consensus that one of the major
+
+1:14:54.960 --> 1:15:02.320
+ failure modes is this loss of control that we create AI systems that are pursuing incorrect
+
+1:15:02.320 --> 1:15:11.040
+ objectives. And because the AI system believes it knows what the objective is, it has no incentive
+
+1:15:11.040 --> 1:15:16.800
+ to listen to us anymore, so to speak, right? It's just carrying out the,
+
+1:15:17.920 --> 1:15:22.160
+ the strategy that it, it has computed as being the optimal solution.
+
+1:15:24.320 --> 1:15:31.040
+ And, you know, it may be that in the process, it needs to acquire more resources to increase the
+
+1:15:31.600 --> 1:15:37.360
+ possibility of success or prevent various failure modes by defending itself against interference.
+
+1:15:37.360 --> 1:15:42.640
+ And so that collection of problems, I think, is something we can address.
+
+1:15:45.280 --> 1:15:55.360
+ The other problems are roughly speaking, you know, misuse, right? So even if we solve the control
+
+1:15:55.360 --> 1:16:00.960
+ problem, we make perfectly safe controllable AI systems. Well, why, you know, why does Dr.
+
+1:16:00.960 --> 1:16:05.600
+ Evil going to use those, right? He wants to just take over the world and he'll make unsafe AI systems
+
+1:16:05.600 --> 1:16:12.000
+ that then get out of control. So that's one problem, which is sort of a, you know, partly a
+
+1:16:12.000 --> 1:16:20.880
+ policing problem, partly a sort of a cultural problem for the profession of how we teach people
+
+1:16:21.760 --> 1:16:26.560
+ what kinds of AI systems are safe. You talk about autonomous weapon system and how pretty much
+
+1:16:26.560 --> 1:16:31.920
+ everybody agrees that there's too many ways that that can go horribly wrong. You have this great
+
+1:16:31.920 --> 1:16:36.560
+ Slotabots movie that kind of illustrates that beautifully. Well, I want to talk about that.
+
+1:16:36.560 --> 1:16:41.440
+ That's another, there's another topic I'm happy to talk about the, I just want to mention that
+
+1:16:41.440 --> 1:16:48.080
+ what I see is the third major failure mode, which is overuse, not so much misuse, but overuse of AI,
+
+1:16:49.680 --> 1:16:55.280
+ that we become overly dependent. So I call this the warly problems. If you've seen the warly,
+
+1:16:55.280 --> 1:17:00.800
+ the movie, all right, all the humans are on the spaceship and the machines look after everything
+
+1:17:00.800 --> 1:17:07.680
+ for them. And they just watch TV and drink big gulps. And they're all sort of obese and stupid.
+
+1:17:07.680 --> 1:17:17.040
+ And they sort of totally lost any notion of human autonomy. And, you know, so in effect, right,
+
+1:17:18.240 --> 1:17:23.520
+ this would happen like the slow boiling frog, right, we would gradually turn over
+
+1:17:24.320 --> 1:17:28.480
+ more and more of the management of our civilization to machines as we are already doing.
+
+1:17:28.480 --> 1:17:34.560
+ And this, you know, this, if this process continues, you know, we sort of gradually
+
+1:17:34.560 --> 1:17:41.440
+ switch from sort of being the masters of technology to just being the guests, right?
+
+1:17:41.440 --> 1:17:45.840
+ So, so we become guests on a cruise ship, you know, which is fine for a week, but not,
+
+1:17:46.480 --> 1:17:53.520
+ not for the rest of eternity, right? You know, and it's almost irreversible, right? Once you,
+
+1:17:53.520 --> 1:18:00.000
+ once you lose the incentive to, for example, you know, learn to be an engineer or a doctor
+
+1:18:00.800 --> 1:18:08.000
+ or a sanitation operative or any other of the, the infinitely many ways that we
+
+1:18:08.000 --> 1:18:13.200
+ maintain and propagate our civilization. You know, if you, if you don't have the
+
+1:18:13.200 --> 1:18:18.400
+ incentive to do any of that, you won't. And then it's really hard to recover.
+
+1:18:18.400 --> 1:18:23.440
+ And of course they add just one of the technologies that could, that third failure mode result in that.
+
+1:18:23.440 --> 1:18:27.360
+ There's probably other technology in general detaches us from.
+
+1:18:28.400 --> 1:18:33.440
+ It does a bit, but the, the, the difference is that in terms of the knowledge to,
+
+1:18:34.080 --> 1:18:39.360
+ to run our civilization, you know, up to now we've had no alternative, but to put it into
+
+1:18:39.360 --> 1:18:44.400
+ people's heads, right? And if you, if you, software with Google, I mean, so software in
+
+1:18:44.400 --> 1:18:51.600
+ general, so computers in general, but, but the, you know, the knowledge of how, you know, how
+
+1:18:51.600 --> 1:18:56.560
+ a sanitation system works, you know, that's an AI has to understand that it's no good putting it
+
+1:18:56.560 --> 1:19:02.960
+ into Google. So, I mean, we, we've always put knowledge in on paper, but paper doesn't run
+
+1:19:02.960 --> 1:19:07.520
+ our civilization. It only runs when it goes from the paper into people's heads again, right? So
+
+1:19:07.520 --> 1:19:13.920
+ we've always propagated civilization through human minds and we've spent about a trillion
+
+1:19:13.920 --> 1:19:19.440
+ person years doing that literally, right? You, you can work it out. It's about, right? There's
+
+1:19:19.440 --> 1:19:25.120
+ about just over a hundred billion people who've ever lived and each of them has spent about 10
+
+1:19:25.120 --> 1:19:30.640
+ years learning stuff to keep their civilization going. And so that's a trillion person years we
+
+1:19:30.640 --> 1:19:35.760
+ put into this effort. Beautiful way to describe all of civilization. And now we're, you know,
+
+1:19:35.760 --> 1:19:39.840
+ we're in danger of throwing that away. So this is a problem that AI can't solve. It's not a
+
+1:19:39.840 --> 1:19:47.120
+ technical problem. It's a, you know, and if we do our job right, the AI systems will say, you know,
+
+1:19:47.120 --> 1:19:52.800
+ the human race doesn't in the long run want to be passengers in a cruise ship. The human race
+
+1:19:52.800 --> 1:19:59.840
+ wants autonomy. This is part of human preferences. So we, the AI systems are not going to do this
+
+1:19:59.840 --> 1:20:05.440
+ stuff for you. You've got to do it for yourself, right? I'm not going to carry you to the top of
+
+1:20:05.440 --> 1:20:11.840
+ Everest in an autonomous helicopter. You have to climb it if you want to get the benefit and so on. So
+
+1:20:14.160 --> 1:20:20.160
+ but I'm afraid that because we are short sighted and lazy, we're going to override the AI systems.
+
+1:20:20.880 --> 1:20:27.520
+ And, and there's an amazing short story that I recommend to everyone that I talk to about this
+
+1:20:27.520 --> 1:20:36.080
+ called the machine stops written in 1909 by Ian Forster, who, you know, wrote novels about the
+
+1:20:36.080 --> 1:20:41.200
+ British Empire and sort of things that became costume dramas on the BBC. But he wrote this one
+
+1:20:41.200 --> 1:20:49.280
+ science fiction story, which is an amazing vision of the future. It has, it has basically iPads.
+
+1:20:49.280 --> 1:20:57.680
+ It has video conferencing. It has MOOCs. It has computer and computer induced obesity. I mean,
+
+1:20:57.680 --> 1:21:02.960
+ literally, the whole thing is what people spend their time doing is giving online courses or
+
+1:21:02.960 --> 1:21:07.920
+ listening to online courses and talking about ideas. But they never get out there in the real
+
+1:21:07.920 --> 1:21:13.680
+ world. They don't really have a lot of face to face contact. Everything is done online.
+
+1:21:13.680 --> 1:21:19.680
+ You know, so all the things we're worrying about now were described in the story and and then the
+
+1:21:19.680 --> 1:21:26.640
+ human race becomes more and more dependent on the machine loses knowledge of how things really run
+
+1:21:27.600 --> 1:21:35.200
+ and then becomes vulnerable to collapse. And so it's a it's a pretty unbelievably amazing
+
+1:21:35.200 --> 1:21:41.520
+ story for someone writing in 1909 to imagine all this. Plus, yeah. So there's very few people
+
+1:21:41.520 --> 1:21:46.080
+ that represent artificial intelligence more than you, Stuart Russell.
+
+1:21:46.960 --> 1:21:50.880
+ If you say it's okay, that's very kind. So it's all my fault.
+
+1:21:50.880 --> 1:21:59.680
+ It's all your fault. No, right. You're often brought up as the person. Well, Stuart Russell,
+
+1:22:00.560 --> 1:22:04.960
+ like the AI person is worried about this. That's why you should be worried about it.
+
+1:22:06.080 --> 1:22:11.280
+ Do you feel the burden of that? I don't know if you feel that at all. But when I talk to people,
+
+1:22:11.280 --> 1:22:16.800
+ like from you talk about people outside of computer science, when they think about this,
+
+1:22:16.800 --> 1:22:22.560
+ Stuart Russell is worried about AI safety, you should be worried too. Do you feel the burden
+
+1:22:22.560 --> 1:22:31.520
+ of that? I mean, in a practical sense, yeah, because I get, you know, a dozen, sometimes
+
+1:22:31.520 --> 1:22:39.600
+ 25 invitations a day to talk about it, to give interviews, to write press articles and so on.
+
+1:22:39.600 --> 1:22:47.120
+ So in that very practical sense, I'm seeing that people are concerned and really interested about
+
+1:22:47.120 --> 1:22:53.680
+ this. Are you worried that you could be wrong, as all good scientists are? Of course. I worry about
+
+1:22:53.680 --> 1:23:00.400
+ that all the time. I mean, that's always been the way that I've worked, you know, is like I have an
+
+1:23:00.400 --> 1:23:06.320
+ argument in my head with myself, right? So I have some idea. And then I think, okay,
+
+1:23:06.320 --> 1:23:11.680
+ okay, how could that be wrong? Or did someone else already have that idea? So I'll go and
+
+1:23:12.800 --> 1:23:18.240
+ search in as much literature as I can to see whether someone else already thought of that
+
+1:23:18.240 --> 1:23:25.600
+ or even refuted it. So, you know, right now, I'm reading a lot of philosophy because,
+
+1:23:25.600 --> 1:23:37.920
+ you know, in the form of the debates over utilitarianism and other kinds of moral formulas,
+
+1:23:37.920 --> 1:23:44.320
+ shall we say, people have already thought through some of these issues. But, you know,
+
+1:23:44.320 --> 1:23:51.280
+ what one of the things I'm not seeing in a lot of these debates is this specific idea about
+
+1:23:51.280 --> 1:23:58.560
+ the importance of uncertainty in the objective, that this is the way we should think about machines
+
+1:23:58.560 --> 1:24:06.800
+ that are beneficial to humans. So this idea of provably beneficial machines based on explicit
+
+1:24:06.800 --> 1:24:15.200
+ uncertainty in the objective, you know, it seems to be, you know, my gut feeling is this is the core
+
+1:24:15.200 --> 1:24:21.200
+ of it. It's going to have to be elaborated in a lot of different directions. And they're a lot
+
+1:24:21.200 --> 1:24:27.440
+ of beneficial. Yeah, but they're, I mean, it has to be, right? We can't afford, you know,
+
+1:24:27.440 --> 1:24:33.040
+ hand wavy beneficial. Because there are, you know, whenever we do hand wavy stuff, there are
+
+1:24:33.040 --> 1:24:38.080
+ loopholes. And the thing about super intelligent machines is they find the loopholes. You know,
+
+1:24:38.080 --> 1:24:44.320
+ just like, you know, tax evaders, if you don't write your tax law properly, people will find
+
+1:24:44.320 --> 1:24:53.440
+ the loopholes and end up paying no tax. And so you should think of it this way. And getting those
+
+1:24:53.440 --> 1:25:03.440
+ definitions right, you know, it is really a long process, you know, so you can you can define
+
+1:25:03.440 --> 1:25:07.760
+ mathematical frameworks. And within that framework, you can prove mathematical theorems that, yes,
+
+1:25:07.760 --> 1:25:12.800
+ this will, you know, this this theoretical entity will be provably beneficial to that theoretical
+
+1:25:12.800 --> 1:25:20.160
+ entity. But that framework may not match the real world in some crucial way. So the long process
+
+1:25:20.160 --> 1:25:27.120
+ thinking through it to iterating and so on. Last question. Yep. You have 10 seconds to answer it.
+
+1:25:27.120 --> 1:25:34.480
+ What is your favorite sci fi movie about AI? I would say interstellar has my favorite robots.
+
+1:25:34.480 --> 1:25:42.160
+ Oh, beats space. Yeah, yeah, yeah. So so Tars, the robots, one of the robots in interstellar is
+
+1:25:42.160 --> 1:25:52.080
+ the way robots should behave. And I would say X Machina is in some ways the one, the one that
+
+1:25:52.080 --> 1:25:58.000
+ makes you think in a nervous kind of way about about where we're going.
+
+1:25:58.000 --> 1:26:13.920
+ Well, Stuart, thank you so much for talking today. Pleasure.
+

vtt/episode_010_small.vtt

@@ -0,0 +1,1331 @@
+WEBVTT
+
+00:00.000 --> 00:03.040
+ The following is a conversation with Peter Abiel.
+
+00:03.040 --> 00:07.760
+ He's a professor at UC Berkeley and the director of the Berkeley Robotics Learning Lab.
+
+00:07.760 --> 00:13.200
+ He's one of the top researchers in the world working on how to make robots understand and
+
+00:13.200 --> 00:18.480
+ interact with the world around them, especially using imitation and deeper enforcement learning.
+
+00:19.680 --> 00:24.160
+ This conversation is part of the MIT course on artificial general intelligence
+
+00:24.160 --> 00:29.040
+ and the artificial intelligence podcast. If you enjoy it, please subscribe on YouTube,
+
+00:29.040 --> 00:34.160
+ iTunes, or your podcast provider of choice, or simply connect with me on Twitter at Lex
+
+00:34.160 --> 00:40.560
+ Freedman, spelled F R I D. And now here's my conversation with Peter Abiel.
+
+00:41.440 --> 00:46.480
+ You've mentioned that if there was one person you could meet, it would be Roger Federer. So let
+
+00:46.480 --> 00:52.720
+ me ask, when do you think we will have a robot that fully autonomously can beat Roger Federer
+
+00:52.720 --> 00:59.840
+ at tennis? Roger Federer level player at tennis? Well, first, if you can make it happen for me
+
+00:59.840 --> 01:07.840
+ to meet Roger, let me know. In terms of getting a robot to beat him at tennis, it's kind of an
+
+01:07.840 --> 01:15.280
+ interesting question because for a lot of the challenges we think about in AI, the software
+
+01:15.280 --> 01:22.800
+ is really the missing piece. But for something like this, the hardware is nowhere near either. To
+
+01:22.800 --> 01:28.240
+ really have a robot that can physically run around, the Boston Dynamics robots are starting to get
+
+01:28.240 --> 01:34.560
+ there, but still not really human level ability to run around and then swing a racket.
+
+01:36.720 --> 01:40.160
+ So you think that's a hardware problem? I don't think it's a hardware problem only. I think it's
+
+01:40.160 --> 01:45.600
+ a hardware and a software problem. I think it's both. And I think they'll have independent progress.
+
+01:45.600 --> 01:53.360
+ So I'd say the hardware maybe in 10, 15 years. On clay, not grass. I mean, grass is probably hard.
+
+01:53.360 --> 02:00.080
+ With the sliding? Yeah. Well, clay, I'm not sure what's harder, grass or clay. The clay involves
+
+02:00.080 --> 02:09.360
+ sliding, which might be harder to master actually. Yeah. But you're not limited to bipedal. I mean,
+
+02:09.360 --> 02:12.560
+ I'm sure there's no... Well, if we can build a machine, it's a whole different question, of
+
+02:12.560 --> 02:18.000
+ course. If you can say, okay, this robot can be on wheels, it can move around on wheels and
+
+02:18.000 --> 02:24.880
+ can be designed differently, then I think that can be done sooner probably than a full humanoid
+
+02:24.880 --> 02:30.400
+ type of setup. What do you think of swing a racket? So you've worked at basic manipulation.
+
+02:31.120 --> 02:36.480
+ How hard do you think is the task of swinging a racket with a be able to hit a nice backhand
+
+02:36.480 --> 02:44.240
+ or a forehand? Let's say we just set up stationery, a nice robot arm, let's say. You know,
+
+02:44.240 --> 02:49.440
+ a standard industrial arm, and it can watch the ball come and then swing the racket.
+
+02:50.560 --> 02:57.600
+ It's a good question. I'm not sure it would be super hard to do. I mean, I'm sure it would require
+
+02:57.600 --> 03:01.520
+ a lot... If we do it with reinforcement learning, it would require a lot of trial and error. It's
+
+03:01.520 --> 03:06.960
+ not going to swing it right the first time around, but yeah, I don't see why I couldn't
+
+03:08.240 --> 03:12.320
+ swing it the right way. I think it's learnable. I think if you set up a ball machine, let's say
+
+03:12.320 --> 03:18.960
+ on one side and then a robot with a tennis racket on the other side, I think it's learnable
+
+03:20.160 --> 03:25.360
+ and maybe a little bit of pre training and simulation. Yeah, I think that's feasible.
+
+03:25.360 --> 03:28.880
+ I think the swinging the racket is feasible. It'd be very interesting to see how much precision it
+
+03:28.880 --> 03:37.760
+ can get. I mean, that's where... I mean, some of the human players can hit it on the lines,
+
+03:37.760 --> 03:44.320
+ which is very high precision. With spin. The spin is an interesting whether RL can learn to
+
+03:44.320 --> 03:48.160
+ put a spin on the ball. Well, you got me interested. Maybe someday we'll set this up.
+
+03:51.040 --> 03:55.440
+ Your answer is basically, okay, for this problem, it sounds fascinating, but for the general problem
+
+03:55.440 --> 03:59.840
+ of a tennis player, we might be a little bit farther away. What's the most impressive thing
+
+03:59.840 --> 04:06.720
+ you've seen a robot do in the physical world? So physically, for me, it's
+
+04:08.720 --> 04:16.560
+ the Boston Dynamics videos always just ring home and just super impressed. Recently, the robot
+
+04:16.560 --> 04:22.160
+ running up the stairs during the parkour type thing. I mean, yes, we don't know what's underneath.
+
+04:22.160 --> 04:26.400
+ They don't really write a lot of detail, but even if it's hard coded underneath,
+
+04:27.040 --> 04:30.800
+ which it might or might not be just the physical abilities of doing that parkour,
+
+04:30.800 --> 04:36.000
+ that's a very impressive robot right there. So have you met Spotmini or any of those robots in
+
+04:36.000 --> 04:43.040
+ person? I met Spotmini last year in April at the Mars event that Jeff Bezos organizes. They
+
+04:43.040 --> 04:49.840
+ brought it out there and it was nicely falling around Jeff. When Jeff left the room, they had it
+
+04:49.840 --> 04:55.680
+ following him along, which is pretty impressive. So I think there's some confidence to know that
+
+04:55.680 --> 05:00.080
+ there's no learning going on in those robots. The psychology of it, so while knowing that,
+
+05:00.080 --> 05:03.360
+ while knowing there's not, if there's any learning going on, it's very limited,
+
+05:03.920 --> 05:08.720
+ I met Spotmini earlier this year and knowing everything that's going on,
+
+05:09.520 --> 05:12.400
+ having one on one interaction, so I get to spend some time alone.
+
+05:14.400 --> 05:18.720
+ And there's immediately a deep connection on the psychological level,
+
+05:18.720 --> 05:22.320
+ even though you know the fundamentals, how it works, there's something magical.
+
+05:23.280 --> 05:29.040
+ So do you think about the psychology of interacting with robots in the physical world,
+
+05:29.040 --> 05:36.000
+ even you just showed me the PR2, the robot, and there was a little bit something like a face,
+
+05:37.040 --> 05:40.480
+ had a little bit something like a face, there's something that immediately draws you to it.
+
+05:40.480 --> 05:45.040
+ Do you think about that aspect of the robotics problem?
+
+05:45.040 --> 05:50.560
+ Well, it's very hard with Brett here. We'll give him a name, Berkeley Robot,
+
+05:50.560 --> 05:56.480
+ for the elimination of tedious tasks. It's very hard to not think of the robot as a person,
+
+05:56.480 --> 06:00.560
+ and it seems like everybody calls him a he for whatever reason, but that also makes it more
+
+06:00.560 --> 06:07.200
+ a person than if it was a it. And it seems pretty natural to think of it that way.
+
+06:07.200 --> 06:12.400
+ This past weekend really struck me, I've seen Pepper many times on videos,
+
+06:12.400 --> 06:18.640
+ but then I was at an event organized by, this was by Fidelity, and they had scripted Pepper to help
+
+06:19.280 --> 06:24.880
+ moderate some sessions, and they had scripted Pepper to have the personality of a child a
+
+06:24.880 --> 06:31.360
+ little bit. And it was very hard to not think of it as its own person in some sense, because it
+
+06:31.360 --> 06:35.120
+ was just kind of jumping, it would just jump into conversation making it very interactive.
+
+06:35.120 --> 06:38.720
+ Moderate would be saying Pepper would just jump in, hold on, how about me,
+
+06:38.720 --> 06:43.600
+ how about me, can I participate in this doing it, just like, okay, this is like like a person,
+
+06:43.600 --> 06:48.800
+ and that was 100% scripted. And even then it was hard not to have that sense of somehow
+
+06:48.800 --> 06:55.120
+ there is something there. So as we have robots interact in this physical world, is that a signal
+
+06:55.120 --> 07:00.160
+ that could be used in reinforcement learning? You've worked a little bit in this direction,
+
+07:00.160 --> 07:05.920
+ but do you think that that psychology can be somehow pulled in? Yes, that's a question I would
+
+07:05.920 --> 07:12.800
+ say a lot, a lot of people ask. And I think part of why they ask it is they're thinking about
+
+07:14.160 --> 07:18.560
+ how unique are we really still as people, like after they see some results, they see
+
+07:18.560 --> 07:23.200
+ a computer play go to say a computer do this that they're like, okay, but can it really have
+
+07:23.200 --> 07:28.960
+ emotion? Can it really interact with us in that way? And then once you're around robots,
+
+07:28.960 --> 07:33.760
+ you already start feeling it. And I think that kind of maybe methodologically, the way that I
+
+07:33.760 --> 07:38.560
+ think of it is, if you run something like reinforcement learnings about optimizing some
+
+07:38.560 --> 07:48.240
+ objective, and there's no reason that the objective couldn't be tied into how much
+
+07:48.240 --> 07:53.120
+ does a person like interacting with this system? And why could not the reinforcement learning system
+
+07:53.120 --> 07:59.040
+ optimize for the robot being fun to be around? And why wouldn't it then naturally become more
+
+07:59.040 --> 08:03.920
+ more interactive and more and more maybe like a person or like a pet? I don't know what it would
+
+08:03.920 --> 08:08.720
+ exactly be, but more and more have those features and acquire them automatically. As long as you
+
+08:08.720 --> 08:16.320
+ can formalize an objective of what it means to like something, how you exhibit what's the ground
+
+08:16.320 --> 08:21.360
+ truth? How do you get the reward from human? Because you have to somehow collect that information
+
+08:21.360 --> 08:27.120
+ from human. But you're saying if you can formulate as an objective, it can be learned.
+
+08:27.120 --> 08:30.800
+ There's no reason it couldn't emerge through learning. And maybe one way to formulate as an
+
+08:30.800 --> 08:35.840
+ objective, you wouldn't have to necessarily score it explicitly. So standard rewards are
+
+08:35.840 --> 08:41.920
+ numbers. And numbers are hard to come by. This is a 1.5 or 1.7 on some scale. It's very hard to do
+
+08:41.920 --> 08:47.680
+ for a person. But much easier is for a person to say, okay, what you did the last five minutes
+
+08:47.680 --> 08:53.600
+ was much nicer than we did the previous five minutes. And that now gives a comparison. And in fact,
+
+08:53.600 --> 08:58.080
+ there have been some results on that. For example, Paul Cristiano and collaborators at OpenEye had
+
+08:58.080 --> 09:05.840
+ the hopper, Mojoka hopper, one legged robot, the backflip, backflips purely from feedback. I like
+
+09:05.840 --> 09:11.280
+ this better than that. That's kind of equally good. And after a bunch of interactions, it figured
+
+09:11.280 --> 09:15.200
+ out what it was the person was asking for, namely a backflip. And so I think the same thing.
+
+09:16.080 --> 09:20.880
+ It wasn't trying to do a backflip. It was just getting a score from the comparison score from
+
+09:20.880 --> 09:27.760
+ the person based on person having a mind in their own mind. I wanted to do a backflip. But
+
+09:27.760 --> 09:32.480
+ the robot didn't know what it was supposed to be doing. It just knew that sometimes the person
+
+09:32.480 --> 09:37.120
+ said, this is better, this is worse. And then the robot figured out what the person was actually
+
+09:37.120 --> 09:42.560
+ after was a backflip. And I imagine the same would be true for things like more interactive
+
+09:42.560 --> 09:47.520
+ robots that the robot would figure out over time. Oh, this kind of thing apparently is appreciated
+
+09:47.520 --> 09:54.720
+ more than this other kind of thing. So when I first picked up Sutton's Richard Sutton's
+
+09:54.720 --> 10:02.480
+ reinforcement learning book, before sort of this deep learning, before the reemergence
+
+10:02.480 --> 10:07.600
+ of neural networks as a powerful mechanism for machine learning, IRL seemed to me like magic.
+
+10:07.600 --> 10:18.000
+ It was beautiful. So that seemed like what intelligence is, RRL reinforcement learning. So how
+
+10:18.000 --> 10:24.320
+ do you think we can possibly learn anything about the world when the reward for the actions is delayed
+
+10:24.320 --> 10:32.160
+ is so sparse? Like where is, why do you think RRL works? Why do you think you can learn anything
+
+10:32.160 --> 10:37.600
+ under such sparse rewards, whether it's regular reinforcement learning or deeper reinforcement
+
+10:37.600 --> 10:45.600
+ learning? What's your intuition? The kind of part of that is, why is RRL, why does it need
+
+10:45.600 --> 10:51.040
+ so many samples, so many experiences to learn from? Because really what's happening is when you
+
+10:51.040 --> 10:56.240
+ have a sparse reward, you do something maybe for like, I don't know, you take 100 actions and then
+
+10:56.240 --> 11:01.920
+ you get a reward, or maybe you get like a score of three. And I'm like, okay, three. Not sure what
+
+11:01.920 --> 11:06.960
+ that means. You go again and now you get two. And now you know that that sequence of 100 actions
+
+11:06.960 --> 11:10.640
+ that you did the second time around somehow was worse than the sequence of 100 actions you did
+
+11:10.640 --> 11:15.040
+ the first time around. But that's tough to now know which one of those were better or worse.
+
+11:15.040 --> 11:19.680
+ Some might have been good and bad in either one. And so that's why you need so many experiences.
+
+11:19.680 --> 11:24.080
+ But once you have enough experiences, effectively RRL is teasing that apart. It's starting to say,
+
+11:24.080 --> 11:28.640
+ okay, what is consistently there when you get a higher reward and what's consistently there when
+
+11:28.640 --> 11:34.080
+ you get a lower reward? And then kind of the magic of sometimes the policy grant update is to say,
+
+11:34.720 --> 11:39.520
+ now let's update the neural network to make the actions that were kind of present when things are
+
+11:39.520 --> 11:44.960
+ good, more likely, and make the actions that are present when things are not as good, less likely.
+
+11:44.960 --> 11:50.480
+ So that's that is the counterpoint. But it seems like you would need to run it a lot more than
+
+11:50.480 --> 11:55.120
+ you do. Even though right now, people could say that RRL is very inefficient. But it seems to be
+
+11:55.120 --> 12:01.200
+ way more efficient than one would imagine on paper, that the simple updates to the policy,
+
+12:01.760 --> 12:07.520
+ the policy gradient that somehow you can learn is exactly as I said, what are the common actions
+
+12:07.520 --> 12:11.680
+ that seem to produce some good results, that that somehow can learn anything.
+
+12:12.640 --> 12:16.800
+ It seems counterintuitive, at least. Is there some intuition behind it?
+
+12:16.800 --> 12:24.720
+ Yeah, so I think there's a few ways to think about this. The way I tend to think about it
+
+12:24.720 --> 12:29.920
+ mostly originally. And so when we started working on deep reinforcement learning here at Berkeley,
+
+12:29.920 --> 12:36.880
+ which was maybe 2011, 12, 13, around that time, John Shulman was a PhD student initially kind of
+
+12:36.880 --> 12:44.480
+ driving it forward here. And kind of the way we thought about it at the time was if you think
+
+12:44.480 --> 12:51.360
+ about rectified linear units or kind of rectifier type neural networks, what do you get? You get
+
+12:51.360 --> 12:56.320
+ something that's piecewise linear feedback control. And if you look at the literature,
+
+12:56.960 --> 13:02.080
+ linear feedback control is extremely successful, can solve many, many problems surprisingly well.
+
+13:03.520 --> 13:07.200
+ I remember, for example, when we did helicopter flight, if you're in a stationary flight regime,
+
+13:07.200 --> 13:12.080
+ not a non stationary, but a stationary flight regime like hover, you can use linear feedback
+
+13:12.080 --> 13:16.960
+ control to stabilize the helicopter, a very complex dynamical system. But the controller
+
+13:16.960 --> 13:22.240
+ is relatively simple. And so I think that's a big part of is that if you do feedback control,
+
+13:22.240 --> 13:25.280
+ even though the system you control can be very, very complex, often,
+
+13:26.000 --> 13:31.520
+ relatively simple control architectures can already do a lot. But then also just linear
+
+13:31.520 --> 13:35.840
+ is not good enough. And so one way you can think of these neural networks is that in some of the
+
+13:35.840 --> 13:40.880
+ tile the space, which people were already trying to do more by hand or with finite state machines,
+
+13:40.880 --> 13:44.560
+ say this linear controller here, this linear controller here, neural network,
+
+13:44.560 --> 13:48.160
+ learns to tell the spin say linear controller here, another linear controller here,
+
+13:48.160 --> 13:52.000
+ but it's more subtle than that. And so it's benefiting from this linear control aspect is
+
+13:52.000 --> 13:57.760
+ benefiting from the tiling, but it's somehow tiling it one dimension at a time. Because if
+
+13:57.760 --> 14:04.160
+ let's say you have a two layer network, even that hidden layer, you make a transition from active
+
+14:04.160 --> 14:09.600
+ to inactive or the other way around, that is essentially one axis, but not axis aligned, but
+
+14:09.600 --> 14:15.200
+ one direction that you change. And so you have this kind of very gradual tiling of the space,
+
+14:15.200 --> 14:19.840
+ we have a lot of sharing between the linear controllers that tile the space. And that was
+
+14:19.840 --> 14:25.280
+ always my intuition as to why to expect that this might work pretty well. It's essentially
+
+14:25.280 --> 14:30.000
+ leveraging the fact that linear feedback control is so good. But of course, not enough. And this
+
+14:30.000 --> 14:35.520
+ is a gradual tiling of the space with linear feedback controls that share a lot of expertise
+
+14:35.520 --> 14:41.120
+ across them. So that that's, that's really nice intuition. But do you think that scales to the
+
+14:41.120 --> 14:47.040
+ more and more general problems of when you start going up the number of control dimensions,
+
+14:48.160 --> 14:55.280
+ when you start going down in terms of how often you get a clean reward signal,
+
+14:55.280 --> 15:00.960
+ does that intuition carry forward to those crazy or weirder worlds that we think of as the real
+
+15:00.960 --> 15:10.000
+ world? So I think where things get really tricky in the real world compared to the things we've
+
+15:10.000 --> 15:13.920
+ looked at so far with great success and reinforcement learning is
+
+15:16.160 --> 15:21.920
+ the time scales, which takes us to an extreme. So when you think about the real world, I mean,
+
+15:22.800 --> 15:28.560
+ I don't know, maybe some student decided to do a PhD here, right? Okay, that's that's a decision,
+
+15:28.560 --> 15:34.000
+ that's a very high level decision. But if you think about their lives, I mean, any person's life,
+
+15:34.000 --> 15:39.360
+ it's a sequence of muscle fiber contractions and relaxations. And that's how you interact with
+
+15:39.360 --> 15:44.480
+ the world. And that's a very high frequency control thing. But it's ultimately what you do
+
+15:44.480 --> 15:49.280
+ and how you affect the world. Until I guess we have brain readings, you can maybe do it slightly
+
+15:49.280 --> 15:55.120
+ differently. But typically, that's how you affect the world. And the decision of doing a PhD is
+
+15:55.120 --> 16:00.240
+ like so abstract relative to what you're actually doing in the world. And I think that's where
+
+16:00.240 --> 16:07.360
+ credit assignment becomes just completely beyond what any current RL algorithm can do. And we need
+
+16:07.360 --> 16:13.360
+ hierarchical reasoning at a level that is just not available at all yet. Where do you think we can
+
+16:13.360 --> 16:19.360
+ pick up hierarchical reasoning by which mechanisms? Yeah, so maybe let me highlight what I think the
+
+16:19.360 --> 16:27.600
+ limitations are of what already was done 20, 30 years ago. In fact, you'll find reasoning systems
+
+16:27.600 --> 16:33.200
+ that reason over relatively long horizons. But the problem is that they were not grounded in the real
+
+16:33.200 --> 16:43.040
+ world. So people would have to hand design some kind of logical, dynamical descriptions of the
+
+16:43.040 --> 16:49.120
+ world. And that didn't tie into perception. And so that didn't tie into real objects and so forth.
+
+16:49.120 --> 16:57.920
+ And so that was a big gap. Now with deep learning, we start having the ability to really see with
+
+16:57.920 --> 17:02.800
+ sensors process that and understand what's in the world. And so it's a good time to try to
+
+17:02.800 --> 17:08.080
+ bring these things together. I see a few ways of getting there. One way to get there would be to say
+
+17:08.080 --> 17:12.160
+ deep learning can get bolted on somehow to some of these more traditional approaches.
+
+17:12.160 --> 17:16.160
+ Now bolted on would probably mean you need to do some kind of end to end training,
+
+17:16.160 --> 17:21.840
+ where you say, my deep learning processing somehow leads to a representation that in term
+
+17:22.720 --> 17:29.680
+ uses some kind of traditional underlying dynamical systems that can be used for planning.
+
+17:29.680 --> 17:33.920
+ And that's, for example, the direction of Eve Tamar and Thanard Kuritach here have been pushing
+
+17:33.920 --> 17:38.800
+ with causal info again. And of course, other people too, that that's that's one way. Can we
+
+17:38.800 --> 17:43.520
+ somehow force it into the form factor that is amenable to reasoning?
+
+17:43.520 --> 17:50.160
+ Another direction we've been thinking about for a long time and didn't make any progress on
+
+17:50.160 --> 17:56.880
+ was more information theoretic approaches. So the idea there was that what it means to take
+
+17:56.880 --> 18:03.840
+ high level action is to take and choose a latent variable now that tells you a lot about what's
+
+18:03.840 --> 18:08.640
+ going to be the case in the future, because that's what it means to to take a high level action.
+
+18:08.640 --> 18:14.480
+ I say, okay, what I decide I'm going to navigate to the gas station because I need to get
+
+18:14.480 --> 18:18.800
+ gas from my car. Well, that'll now take five minutes to get there. But the fact that I get
+
+18:18.800 --> 18:23.200
+ there, I could already tell that from the high level action I took much earlier.
+
+18:24.480 --> 18:30.080
+ That we had a very hard time getting success with, not saying it's a dead end,
+
+18:30.080 --> 18:34.160
+ necessarily, but we had a lot of trouble getting that to work. And then we started revisiting
+
+18:34.160 --> 18:39.600
+ the notion of what are we really trying to achieve? What we're trying to achieve is
+
+18:39.600 --> 18:42.880
+ not necessarily a hierarchy per se, but you could think about what does hierarchy give us?
+
+18:44.160 --> 18:50.560
+ What we hope it would give us is better credit assignment. What is better credit assignment
+
+18:50.560 --> 18:58.640
+ is giving us, it gives us faster learning. And so faster learning is ultimately maybe
+
+18:58.640 --> 19:03.840
+ what we're after. And so that's where we ended up with the RL squared paper on learning to
+
+19:03.840 --> 19:10.640
+ reinforcement learn, which at a time Rocky Dwan led. And that's exactly the meta learning
+
+19:10.640 --> 19:15.040
+ approach where we say, okay, we don't know how to design hierarchy. We know what we want to get
+
+19:15.040 --> 19:20.000
+ from it. Let's just enter and optimize for what we want to get from it and see if it might emerge.
+
+19:20.000 --> 19:24.720
+ And we saw things emerge. The maze navigation had consistent motion down hallways,
+
+19:25.920 --> 19:29.520
+ which is what you want. A hierarchical control should say, I want to go down this hallway.
+
+19:29.520 --> 19:33.040
+ And then when there is an option to take a turn, I can decide whether to take a turn or not and
+
+19:33.040 --> 19:38.480
+ repeat, even had the notion of, where have you been before or not to not revisit places you've
+
+19:38.480 --> 19:45.840
+ been before? It still didn't scale yet to the real world kind of scenarios I think you had in mind,
+
+19:45.840 --> 19:50.000
+ but it was some sign of life that maybe you can meta learn these hierarchical concepts.
+
+19:51.040 --> 19:58.000
+ I mean, it seems like through these meta learning concepts, we get at the, what I think is one of
+
+19:58.000 --> 20:05.120
+ the hardest and most important problems of AI, which is transfer learning. So it's generalization.
+
+20:06.240 --> 20:12.160
+ How far along this journey towards building general systems are we being able to do transfer
+
+20:12.160 --> 20:18.320
+ learning? Well, so there's some signs that you can generalize a little bit. But do you think
+
+20:18.320 --> 20:25.360
+ we're on the right path or totally different breakthroughs are needed to be able to transfer
+
+20:25.360 --> 20:34.000
+ knowledge between different learned models? Yeah, I'm pretty torn on this in that I think
+
+20:34.000 --> 20:44.400
+ there are some very impressive results already, right? I mean, I would say when even with the
+
+20:44.400 --> 20:50.160
+ initial kind of big breakthrough in 2012 with Alex net, right, the initial, the initial thing is,
+
+20:50.160 --> 20:57.600
+ okay, great. This does better on image net hands image recognition. But then immediately thereafter,
+
+20:57.600 --> 21:04.080
+ there was of course the notion that wow, what was learned on image net, and you now want to solve
+
+21:04.080 --> 21:11.280
+ a new task, you can fine tune Alex net for new tasks. And that was often found to be the even
+
+21:11.280 --> 21:15.920
+ bigger deal that you learn something that was reusable, which was not often the case before
+
+21:15.920 --> 21:19.520
+ usually machine learning, you learn something for one scenario. And that was it. And that's
+
+21:19.520 --> 21:23.200
+ really exciting. I mean, that's just a huge application. That's probably the biggest
+
+21:23.200 --> 21:28.960
+ success of transfer learning today, if in terms of scope and impact. That was a huge breakthrough.
+
+21:28.960 --> 21:37.040
+ And then recently, I feel like similar kind of by scaling things up, it seems like this has been
+
+21:37.040 --> 21:41.440
+ expanded upon like people training even bigger networks, they might transfer even better. If
+
+21:41.440 --> 21:46.480
+ you look that, for example, some of the opening results on language models. And so in the recent
+
+21:46.480 --> 21:53.600
+ Google results on language models, they are learned for just prediction. And then they get
+
+21:54.320 --> 21:59.600
+ reused for other tasks. And so I think there is something there where somehow if you train a
+
+21:59.600 --> 22:05.200
+ big enough model on enough things, it seems to transfer some deep mind results that I thought
+
+22:05.200 --> 22:12.160
+ were very impressive, the unreal results, where it was learning to navigate mazes in ways where
+
+22:12.160 --> 22:16.880
+ it wasn't just doing reinforcement learning, but it had other objectives was optimizing for. So I
+
+22:16.880 --> 22:23.680
+ think there's a lot of interesting results already. I think maybe where it's hard to wrap my head
+
+22:23.680 --> 22:30.160
+ around this, to which extent or when do we call something generalization, right? Or the levels
+
+22:30.160 --> 22:37.360
+ of generalization involved in these different tasks, right? So you draw this, by the way, just
+
+22:37.360 --> 22:43.280
+ to frame things. I've heard you say somewhere, it's the difference in learning to master versus
+
+22:43.280 --> 22:49.680
+ learning to generalize. That it's a nice line to think about. And I guess you're saying it's a gray
+
+22:49.680 --> 22:54.640
+ area of what learning to master and learning to generalize where one starts.
+
+22:54.640 --> 22:58.800
+ I think I might have heard this. I might have heard it somewhere else. And I think it might have
+
+22:58.800 --> 23:05.120
+ been one of your interviews, maybe the one with Yoshua Benjamin, 900% sure. But I like the example
+
+23:05.120 --> 23:12.000
+ and I'm going to not sure who it was, but the example was essentially if you use current deep
+
+23:12.000 --> 23:20.480
+ learning techniques, what we're doing to predict, let's say the relative motion of our planets,
+
+23:20.480 --> 23:27.680
+ it would do pretty well. But then now if a massive new mass enters our solar system,
+
+23:28.320 --> 23:32.880
+ it would probably not predict what will happen, right? And that's a different kind of
+
+23:32.880 --> 23:38.400
+ generalization. That's a generalization that relies on the ultimate simplest explanation
+
+23:38.400 --> 23:42.640
+ that we have available today to explain the motion of planets, whereas just pattern recognition
+
+23:42.640 --> 23:48.160
+ could predict our current solar system motion pretty well. No problem. And so I think that's
+
+23:48.160 --> 23:53.920
+ an example of a kind of generalization that is a little different from what we've achieved so far.
+
+23:54.480 --> 24:01.360
+ And it's not clear if just, you know, regularizing more and forcing it to come up with a simpler,
+
+24:01.360 --> 24:05.280
+ simpler, simpler explanation. Look, this is not simple, but that's what physics researchers do,
+
+24:05.280 --> 24:10.000
+ right, to say, can I make this even simpler? How simple can I get this? What's the simplest
+
+24:10.000 --> 24:14.560
+ equation that can explain everything, right? The master equation for the entire dynamics of the
+
+24:14.560 --> 24:20.960
+ universe. We haven't really pushed that direction as hard in deep learning, I would say. Not sure
+
+24:20.960 --> 24:24.960
+ if it should be pushed, but it seems a kind of generalization you get from that that you don't
+
+24:24.960 --> 24:30.400
+ get in our current methods so far. So I just talked to Vladimir Vapnik, for example, who was
+
+24:30.400 --> 24:39.200
+ a statistician in statistical learning, and he kind of dreams of creating the E equals Mc
+
+24:39.200 --> 24:44.400
+ squared for learning, right, the general theory of learning. Do you think that's a fruitless pursuit
+
+24:46.480 --> 24:50.560
+ in the near term, within the next several decades?
+
+24:51.680 --> 24:56.800
+ I think that's a really interesting pursuit. And in the following sense, in that there is a
+
+24:56.800 --> 25:05.440
+ lot of evidence that the brain is pretty modular. And so I wouldn't maybe think of it as the theory,
+
+25:05.440 --> 25:12.480
+ maybe, the underlying theory, but more kind of the principle where there have been findings where
+
+25:14.160 --> 25:20.240
+ people who are blind will use the part of the brain usually used for vision for other functions.
+
+25:20.240 --> 25:26.800
+ And even after some kind of, if people get rewired in some way, they might be able to reuse parts of
+
+25:26.800 --> 25:35.040
+ their brain for other functions. And so what that suggests is some kind of modularity. And I think
+
+25:35.040 --> 25:41.120
+ it is a pretty natural thing to strive for to see, can we find that modularity? Can we find this
+
+25:41.120 --> 25:45.440
+ thing? Of course, it's not every part of the brain is not exactly the same. Not everything can be
+
+25:45.440 --> 25:50.080
+ rewired arbitrarily. But if you think of things like the neocortex, which is a pretty big part of
+
+25:50.080 --> 25:56.880
+ the brain, that seems fairly modular from what the findings so far. Can you design something
+
+25:56.880 --> 26:01.840
+ equally modular? And if you can just grow it, it becomes more capable, probably. I think that would
+
+26:01.840 --> 26:07.200
+ be the kind of interesting underlying principle to shoot for that is not unrealistic.
+
+26:07.200 --> 26:14.400
+ Do you think you prefer math or empirical trial and error for the discovery of the essence of what
+
+26:14.400 --> 26:19.680
+ it means to do something intelligent? So reinforcement learning embodies both groups, right?
+
+26:19.680 --> 26:25.760
+ To prove that something converges, prove the bounds. And then at the same time, a lot of those
+
+26:25.760 --> 26:31.280
+ successes are, well, let's try this and see if it works. So which do you gravitate towards? How do
+
+26:31.280 --> 26:40.960
+ you think of those two parts of your brain? So maybe I would prefer we could make the progress
+
+26:41.600 --> 26:46.560
+ with mathematics. And the reason maybe I would prefer that is because often if you have something you
+
+26:46.560 --> 26:54.080
+ can mathematically formalize, you can leapfrog a lot of experimentation. And experimentation takes
+
+26:54.080 --> 27:01.440
+ a long time to get through. And a lot of trial and error, reinforcement learning, your research
+
+27:01.440 --> 27:05.040
+ process. But you need to do a lot of trial and error before you get to a success. So if you can
+
+27:05.040 --> 27:10.400
+ leapfrog that, to my mind, that's what the math is about. And hopefully once you do a bunch of
+
+27:10.400 --> 27:15.600
+ experiments, you start seeing a pattern, you can do some derivations that leapfrog some experiments.
+
+27:16.240 --> 27:20.160
+ But I agree with you. I mean, in practice, a lot of the progress has been such that we have not
+
+27:20.160 --> 27:25.840
+ been able to find the math that allows it to leapfrog ahead. And we are kind of making gradual
+
+27:25.840 --> 27:30.480
+ progress one step at a time. A new experiment here, a new experiment there that gives us new
+
+27:30.480 --> 27:35.280
+ insights and gradually building up, but not getting to something yet where we're just, okay,
+
+27:35.280 --> 27:39.920
+ here's an equation that now explains how, you know, that would be have been two years of
+
+27:39.920 --> 27:44.880
+ experimentation to get there. But this tells us what the results going to be. Unfortunately,
+
+27:44.880 --> 27:52.800
+ unfortunately, not so much yet. Not so much yet. But your hope is there. In trying to teach robots
+
+27:52.800 --> 28:01.200
+ or systems to do everyday tasks, or even in simulation, what do you think you're more excited
+
+28:01.200 --> 28:10.560
+ about? imitation learning or self play. So letting robots learn from humans, or letting robots plan
+
+28:10.560 --> 28:18.240
+ their own, try to figure out in their own way, and eventually play, eventually interact with humans,
+
+28:18.240 --> 28:23.200
+ or solve whatever problem is. What's the more exciting to you? What's more promising you think
+
+28:23.200 --> 28:34.240
+ is a research direction? So when we look at self play, what's so beautiful about it is,
+
+28:34.240 --> 28:37.680
+ goes back to kind of the challenges in reinforcement learning. So the challenge
+
+28:37.680 --> 28:43.200
+ of reinforcement learning is getting signal. And if you don't never succeed, you don't get any signal.
+
+28:43.200 --> 28:49.040
+ In self play, you're on both sides. So one of you succeeds. And the beauty is also one of you
+
+28:49.040 --> 28:53.520
+ fails. And so you see the contrast, you see the one version of me that did better than the other
+
+28:53.520 --> 28:58.400
+ version. And so every time you play yourself, you get signal. And so whenever you can turn
+
+28:58.400 --> 29:04.160
+ something into self play, you're in a beautiful situation where you can naturally learn much
+
+29:04.160 --> 29:10.080
+ more quickly than in most other reinforcement learning environments. So I think, I think if
+
+29:10.080 --> 29:15.760
+ somehow we can turn more reinforcement learning problems into self play formulations, that would
+
+29:15.760 --> 29:21.760
+ go really, really far. So far, self play has been largely around games where there is natural
+
+29:21.760 --> 29:25.440
+ opponents. But if we could do self play for other things, and let's say, I don't know,
+
+29:25.440 --> 29:29.360
+ a robot learns to build a house, I mean, that's a pretty advanced thing to try to do for a robot,
+
+29:29.360 --> 29:34.240
+ but maybe it tries to build a hut or something. If that can be done through self play, it would
+
+29:34.240 --> 29:38.560
+ learn a lot more quickly if somebody can figure it out. And I think that would be something where
+
+29:38.560 --> 29:42.560
+ it goes closer to kind of the mathematical leapfrogging where somebody figures out a
+
+29:42.560 --> 29:47.680
+ formalism to say, okay, any RL problem by playing this and this idea, you can turn it
+
+29:47.680 --> 29:50.480
+ into a self play problem where you get signal a lot more easily.
+
+29:52.400 --> 29:57.680
+ Reality is many problems, we don't know how to turn to self play. And so either we need to provide
+
+29:57.680 --> 30:02.640
+ detailed reward. That doesn't just reward for achieving a goal, but rewards for making progress,
+
+30:02.640 --> 30:06.480
+ and that becomes time consuming. And once you're starting to do that, let's say you want a robot
+
+30:06.480 --> 30:09.920
+ to do something, you need to give all this detailed reward. Well, why not just give a
+
+30:09.920 --> 30:15.920
+ demonstration? Because why not just show the robot. And now the question is, how do you show
+
+30:15.920 --> 30:20.240
+ the robot? One way to show is to tally operator robot and then robot really experiences things.
+
+30:20.800 --> 30:24.480
+ And that's nice, because that's really high signal to noise ratio data. And we've done a lot
+
+30:24.480 --> 30:29.360
+ of that. And you teach your robot skills. In just 10 minutes, you can teach your robot a new basic
+
+30:29.360 --> 30:33.360
+ skill, like, okay, pick up the bottle, place it somewhere else. That's a skill, no matter where
+
+30:33.360 --> 30:38.000
+ the bottle starts, maybe it always goes on to a target or something. That's fairly easy to teach
+
+30:38.000 --> 30:43.120
+ your robot with teleop. Now, what's even more interesting, if you can now teach your robot
+
+30:43.120 --> 30:48.480
+ through third person learning, where the robot watches you do something, and doesn't experience
+
+30:48.480 --> 30:52.880
+ it, but just watches it and says, okay, well, if you're showing me that, that means I should
+
+30:52.880 --> 30:56.880
+ be doing this. And I'm not going to be using your hand, because I don't get to control your hand,
+
+30:56.880 --> 31:02.000
+ but I'm going to use my hand, I do that mapping. And so that's where I think one of the big breakthroughs
+
+31:02.000 --> 31:07.520
+ has happened this year. This was led by Chelsea Finn here. It's almost like learning a machine
+
+31:07.520 --> 31:12.000
+ translation for demonstrations where you have a human demonstration and the robot learns to
+
+31:12.000 --> 31:17.440
+ translate it into what it means for the robot to do it. And that was a meta learning formulation,
+
+31:17.440 --> 31:23.440
+ learn from one to get the other. And that I think opens up a lot of opportunities to learn a lot
+
+31:23.440 --> 31:28.080
+ more quickly. So my focus is on autonomous vehicles. Do you think this approach of third
+
+31:28.080 --> 31:33.040
+ person watching is the autonomous driving is amenable to this kind of approach?
+
+31:33.840 --> 31:42.080
+ So for autonomous driving, I would say it's third person is slightly easier. And the reason I'm
+
+31:42.080 --> 31:48.320
+ going to say it's slightly easier to do with third person is because the car dynamics are very well
+
+31:48.320 --> 31:56.560
+ understood. So the easier than first person, you mean, or easier than. So I think the distinction
+
+31:56.560 --> 32:01.680
+ between third person and first person is not a very important distinction for autonomous driving.
+
+32:01.680 --> 32:07.760
+ They're very similar. Because the distinction is really about who turns the steering wheel.
+
+32:07.760 --> 32:15.280
+ And or maybe let me put it differently. How to get from a point where you are now to a point,
+
+32:15.280 --> 32:19.120
+ let's say a couple of meters in front of you. And that's a problem that's very well understood.
+
+32:19.120 --> 32:22.480
+ And that's the only distinction between third and first person there. Whereas with the robot
+
+32:22.480 --> 32:26.720
+ manipulation, interaction forces are very complex. And it's still a very different thing.
+
+32:27.840 --> 32:33.840
+ For autonomous driving, I think there's still the question imitation versus RL.
+
+32:33.840 --> 32:39.520
+ Well, so imitation gives you a lot more signal. I think where imitation is lacking and needs
+
+32:39.520 --> 32:47.600
+ some extra machinery is it doesn't in its normal format, doesn't think about goals or objectives.
+
+32:48.480 --> 32:52.240
+ And of course, there are versions of imitation learning, inverse reinforcement learning type
+
+32:52.240 --> 32:57.440
+ imitation, which also thinks about goals. I think then we're getting much closer. But I think it's
+
+32:57.440 --> 33:05.120
+ very hard to think of a fully reactive car generalizing well, if it really doesn't have a notion
+
+33:05.120 --> 33:10.720
+ of objectives to generalize well to the kind of general that you would want, you want more than
+
+33:10.720 --> 33:15.200
+ just that reactivity that you get from just behavioral cloning slash supervised learning.
+
+33:17.040 --> 33:22.560
+ So a lot of the work, whether it's self play or even imitation learning would benefit
+
+33:22.560 --> 33:27.440
+ significantly from simulation, from effective simulation, and you're doing a lot of stuff
+
+33:27.440 --> 33:32.400
+ in the physical world and in simulation, do you have hope for greater and greater
+
+33:33.520 --> 33:40.160
+ power of simulation loop being boundless, eventually, to where most of what we need
+
+33:40.160 --> 33:45.600
+ to operate in the physical world, what could be simulated to a degree that's directly
+
+33:45.600 --> 33:54.720
+ transferable to the physical world? Are we still very far away from that? So I think
+
+33:55.840 --> 34:03.200
+ we could even rephrase that question in some sense, please. And so the power of simulation,
+
+34:04.720 --> 34:09.760
+ as simulators get better and better, of course, becomes stronger, and we can learn more in
+
+34:09.760 --> 34:13.760
+ simulation. But there's also another version, which is where you say the simulator doesn't
+
+34:13.760 --> 34:19.120
+ even have to be that precise. As long as it's somewhat representative. And instead of trying
+
+34:19.120 --> 34:24.480
+ to get one simulator that is sufficiently precise to learn and transfer really well to the real
+
+34:24.480 --> 34:29.200
+ world, I'm going to build many simulators, ensemble of simulators, ensemble of simulators,
+
+34:30.080 --> 34:35.120
+ not any single one of them is sufficiently representative of the real world such that
+
+34:35.120 --> 34:41.760
+ it would work if you train in there. But if you train in all of them, then there is something
+
+34:41.760 --> 34:47.840
+ that's good in all of them. The real world will just be, you know, another one of them. That's,
+
+34:47.840 --> 34:50.720
+ you know, not identical to any one of them, but just another one of them.
+
+34:50.720 --> 34:53.120
+ Now, this sample from the distribution of simulators.
+
+34:53.120 --> 34:53.360
+ Exactly.
+
+34:53.360 --> 34:57.600
+ We do live in a simulation. So this is just one, one other one.
+
+34:57.600 --> 35:03.440
+ I'm not sure about that. But yeah, it's definitely a very advanced simulator if it is.
+
+35:03.440 --> 35:08.960
+ Yeah, it's a pretty good one. I've talked to Russell. It's something you think about a little bit
+
+35:08.960 --> 35:13.120
+ too. Of course, you're like really trying to build these systems. But do you think about the future
+
+35:13.120 --> 35:18.880
+ of AI? A lot of people have concern about safety. How do you think about AI safety as you build
+
+35:18.880 --> 35:24.960
+ robots that are operating the physical world? What is, yeah, how do you approach this problem
+
+35:24.960 --> 35:27.440
+ in an engineering kind of way in a systematic way?
+
+35:29.200 --> 35:36.720
+ So when a robot is doing things, you kind of have a few notions of safety to worry about. One is that
+
+35:36.720 --> 35:43.760
+ the robot is physically strong and of course could do a lot of damage. Same for cars, which we can
+
+35:43.760 --> 35:49.360
+ think of as robots do in some way. And this could be completely unintentional. So it could be not
+
+35:49.360 --> 35:54.240
+ the kind of long term AI safety concerns that, okay, AI is smarter than us. And now what do we do?
+
+35:54.240 --> 35:57.760
+ But it could be just very practical. Okay, this robot, if it makes a mistake,
+
+35:58.800 --> 36:04.080
+ what are the results going to be? Of course, simulation comes in a lot there to test in simulation.
+
+36:04.080 --> 36:10.960
+ It's a difficult question. And I'm always wondering, like I always wonder, let's say you look at,
+
+36:10.960 --> 36:14.000
+ let's go back to driving, because a lot of people know driving well, of course.
+
+36:15.120 --> 36:20.800
+ What do we do to test somebody for driving, right, to get a driver's license? What do they
+
+36:20.800 --> 36:27.680
+ really do? I mean, you fill out some tests, and then you drive and I mean, for a few minutes,
+
+36:27.680 --> 36:34.800
+ it's suburban California, that driving test is just you drive around the block, pull over, you
+
+36:34.800 --> 36:39.280
+ do a stop sign successfully, and then, you know, you pull over again, and you're pretty much done.
+
+36:40.000 --> 36:46.720
+ And you're like, okay, if a self driving car did that, would you trust it that it can drive?
+
+36:46.720 --> 36:49.840
+ And I'd be like, no, that's not enough for me to trust it. But somehow for humans,
+
+36:50.560 --> 36:54.480
+ we've figured out that somebody being able to do that is representative
+
+36:54.480 --> 36:59.840
+ of them being able to do a lot of other things. And so I think somehow for humans,
+
+36:59.840 --> 37:05.200
+ we figured out representative tests of what it means if you can do this, what you can really do.
+
+37:05.760 --> 37:09.840
+ Of course, testing humans, humans don't want to be tested at all times. Self driving cars or
+
+37:09.840 --> 37:13.760
+ robots could be tested more often probably, you can have replicas that get tested and are known
+
+37:13.760 --> 37:19.600
+ to be identical because they use the same neural net and so forth. But still, I feel like we don't
+
+37:19.600 --> 37:25.040
+ have this kind of unit tests or proper tests for robots. And I think there's something very
+
+37:25.040 --> 37:29.440
+ interesting to be thought about there, especially as you update things, your software improves,
+
+37:29.440 --> 37:34.640
+ you have a better self driving car suite, you update it. How do you know it's indeed more
+
+37:34.640 --> 37:41.440
+ capable on everything than what you had before that you didn't have any bad things creep into it?
+
+37:41.440 --> 37:45.680
+ So I think that's a very interesting direction of research that there is no real solution yet,
+
+37:45.680 --> 37:50.640
+ except that somehow for humans, we do because we say, okay, you have a driving test, you passed,
+
+37:50.640 --> 37:55.760
+ you can go on the road now and you must have accents every like a million or 10 million miles,
+
+37:55.760 --> 38:01.520
+ something pretty phenomenal compared to that short test that is being done.
+
+38:01.520 --> 38:06.000
+ So let me ask, you've mentioned, you've mentioned that Andrew Ang, by example,
+
+38:06.000 --> 38:11.440
+ showed you the value of kindness. And do you think the space of
+
+38:11.440 --> 38:20.240
+ of policies, good policies for humans and for AI is populated by policies that
+
+38:21.440 --> 38:28.880
+ with kindness or ones that are the opposite, exploitation, even evil. So if you just look
+
+38:28.880 --> 38:34.400
+ at the sea of policies we operate under as human beings, or if AI system had to operate in this
+
+38:34.400 --> 38:39.440
+ real world, do you think it's really easy to find policies that are full of kindness,
+
+38:39.440 --> 38:44.480
+ like we naturally fall into them? Or is it like a very hard optimization problem?
+
+38:47.920 --> 38:52.720
+ I mean, there is kind of two optimizations happening for humans, right? So for humans,
+
+38:52.720 --> 38:57.440
+ there's kind of the very long term optimization, which evolution has done for us. And we're kind of
+
+38:57.440 --> 39:02.640
+ predisposed to like certain things. And that's in some sense, what makes our learning easier,
+
+39:02.640 --> 39:10.000
+ because I mean, we know things like pain and hunger and thirst. And the fact that we know about those
+
+39:10.000 --> 39:13.840
+ is not something that we were taught. That's kind of innate. When we're hungry, we're unhappy.
+
+39:13.840 --> 39:20.720
+ When we're thirsty, we're unhappy. When we have pain, we're unhappy. And ultimately evolution
+
+39:20.720 --> 39:25.040
+ built that into us to think about those things. And so I think there is a notion that it seems
+
+39:25.040 --> 39:33.840
+ somehow humans evolved in general to prefer to get along in some ways. But at the same time,
+
+39:33.840 --> 39:43.040
+ also to be very territorial and kind of centric to their own tribe. It seems like that's the kind
+
+39:43.040 --> 39:47.360
+ of space we converged on to. I mean, I'm not an expert in anthropology, but it seems like we're
+
+39:47.360 --> 39:54.480
+ very kind of good within our own tribe, but need to be taught to be nice to other tribes.
+
+39:54.480 --> 39:58.000
+ Well, if you look at Steven Pinker, he highlights this pretty nicely in
+
+40:00.720 --> 40:05.520
+ Better Angels of Our Nature, where he talks about violence decreasing over time consistently.
+
+40:05.520 --> 40:11.360
+ So whatever tension, whatever teams we pick, it seems that the long arc of history goes
+
+40:11.360 --> 40:17.840
+ towards us getting along more and more. So do you think that
+
+40:17.840 --> 40:27.280
+ do you think it's possible to teach RRL based robots this kind of kindness, this kind of ability
+
+40:27.280 --> 40:33.040
+ to interact with humans, this kind of policy, even to let me ask, let me ask upon one, do you think
+
+40:33.040 --> 40:38.800
+ it's possible to teach RRL based robot to love a human being and to inspire that human to love
+
+40:38.800 --> 40:48.080
+ the robot back? So to like a RRL based algorithm that leads to a happy marriage? That's an interesting
+
+40:48.080 --> 40:56.080
+ question. Maybe I'll answer it with another question, right? Because I mean, but I'll come
+
+40:56.080 --> 41:02.000
+ back to it. So another question you can have is okay. I mean, how close does some people's
+
+41:02.000 --> 41:09.760
+ happiness get from interacting with just a really nice dog? Like, I mean, dogs, you come home,
+
+41:09.760 --> 41:14.000
+ that's what dogs do. They greet you. They're excited. It makes you happy when you come home
+
+41:14.000 --> 41:17.600
+ to your dog. You're just like, okay, this is exciting. They're always happy when I'm here.
+
+41:18.160 --> 41:22.560
+ I mean, if they don't greet you, because maybe whatever, your partner took them on a trip or
+
+41:22.560 --> 41:27.600
+ something, you might not be nearly as happy when you get home, right? And so the kind of,
+
+41:27.600 --> 41:33.600
+ it seems like the level of reasoning a dog has is pretty sophisticated, but then it's still not yet
+
+41:33.600 --> 41:38.240
+ at the level of human reasoning. And so it seems like we don't even need to achieve human level
+
+41:38.240 --> 41:44.320
+ reasoning to get like very strong affection with humans. And so my thinking is, why not, right?
+
+41:44.320 --> 41:51.360
+ Why couldn't, with an AI, couldn't we achieve the kind of level of affection that humans feel
+
+41:51.360 --> 41:59.280
+ among each other or with friendly animals and so forth? So question, is it a good thing for us
+
+41:59.280 --> 42:07.040
+ or not? That's another thing, right? Because I mean, but I don't see why not. Why not? Yeah.
+
+42:07.040 --> 42:12.640
+ So Elon Musk says love is the answer. Maybe he should say love is the objective function and
+
+42:12.640 --> 42:19.280
+ then RL is the answer, right? Well, maybe. Oh, Peter, thank you so much. I don't want to take
+
+42:19.280 --> 42:23.360
+ up more of your time. Thank you so much for talking today. Well, thanks for coming by.
+
+42:23.360 --> 42:53.200
+ Great to have you visit.
+

vtt/episode_011_small.vtt

@@ -0,0 +1,3773 @@
+WEBVTT
+
+00:00.000 --> 00:03.520
+ The following is a conversation with Jürgen Schmidhuber.
+
+00:03.520 --> 00:06.360
+ He's the codirector of its CS Swiss AI lab
+
+00:06.360 --> 00:10.360
+ and a cocreator of long short term memory networks.
+
+00:10.360 --> 00:13.720
+ LSDMs are used in billions of devices today
+
+00:13.720 --> 00:17.400
+ for speech recognition, translation, and much more.
+
+00:17.400 --> 00:20.800
+ Over 30 years, he has proposed a lot of interesting
+
+00:20.800 --> 00:24.800
+ out of the box ideas, a meta learning, adversarial networks,
+
+00:24.800 --> 00:28.720
+ computer vision, and even a formal theory of quote,
+
+00:28.720 --> 00:32.360
+ creativity, curiosity, and fun.
+
+00:32.360 --> 00:34.920
+ This conversation is part of the MIT course
+
+00:34.920 --> 00:36.520
+ on artificial general intelligence
+
+00:36.520 --> 00:38.840
+ and the artificial intelligence podcast.
+
+00:38.840 --> 00:41.960
+ If you enjoy it, subscribe on YouTube, iTunes,
+
+00:41.960 --> 00:43.960
+ or simply connect with me on Twitter
+
+00:43.960 --> 00:47.280
+ at Lex Friedman spelled F R I D.
+
+00:47.280 --> 00:51.480
+ And now here's my conversation with Jürgen Schmidhuber.
+
+00:53.080 --> 00:55.640
+ Early on you dreamed of AI systems
+
+00:55.640 --> 00:58.680
+ that self improve recursively.
+
+00:58.680 --> 01:01.440
+ When was that dream born?
+
+01:01.440 --> 01:02.840
+ When I was a baby.
+
+01:02.840 --> 01:04.000
+ No, that's not true.
+
+01:04.000 --> 01:06.200
+ When I was a teenager.
+
+01:06.200 --> 01:09.400
+ And what was the catalyst for that birth?
+
+01:09.400 --> 01:12.800
+ What was the thing that first inspired you?
+
+01:12.800 --> 01:15.000
+ When I was a boy, I...
+
+01:17.400 --> 01:19.880
+ I was thinking about what to do in my life
+
+01:19.880 --> 01:23.560
+ and then I thought the most exciting thing
+
+01:23.560 --> 01:27.160
+ is to solve the riddles of the universe.
+
+01:27.160 --> 01:30.720
+ And that means you have to become a physicist.
+
+01:30.720 --> 01:35.640
+ However, then I realized that there's something even grander.
+
+01:35.640 --> 01:39.680
+ You can try to build a machine.
+
+01:39.680 --> 01:41.920
+ That isn't really a machine any longer.
+
+01:41.920 --> 01:44.280
+ That learns to become a much better physicist
+
+01:44.280 --> 01:46.840
+ than I could ever hope to be.
+
+01:46.840 --> 01:50.120
+ And that's how I thought maybe I can multiply
+
+01:50.120 --> 01:54.320
+ my tiny little bit of creativity into infinity.
+
+01:54.320 --> 01:57.160
+ But ultimately, that creativity will be multiplied
+
+01:57.160 --> 01:59.160
+ to understand the universe around us.
+
+01:59.160 --> 02:05.640
+ That's the curiosity for that mystery that drove you.
+
+02:05.640 --> 02:08.320
+ Yes, so if you can build a machine
+
+02:08.320 --> 02:13.760
+ that learns to solve more and more complex problems
+
+02:13.760 --> 02:16.720
+ and more and more general problems over,
+
+02:16.720 --> 02:22.520
+ then you basically have solved all the problems.
+
+02:22.520 --> 02:25.960
+ At least all the solvable problems.
+
+02:25.960 --> 02:27.080
+ So how do you think...
+
+02:27.080 --> 02:31.440
+ What is the mechanism for that kind of general solver look like?
+
+02:31.440 --> 02:35.480
+ Obviously, we don't quite yet have one or know
+
+02:35.480 --> 02:37.040
+ how to build one boy of ideas
+
+02:37.040 --> 02:40.800
+ and you have had throughout your career several ideas about it.
+
+02:40.800 --> 02:43.600
+ So how do you think about that mechanism?
+
+02:43.600 --> 02:48.640
+ So in the 80s, I thought about how to build this machine
+
+02:48.640 --> 02:51.000
+ that learns to solve all these problems
+
+02:51.000 --> 02:54.120
+ that I cannot solve myself.
+
+02:54.120 --> 02:57.120
+ And I thought it is clear, it has to be a machine
+
+02:57.120 --> 03:00.880
+ that not only learns to solve this problem here
+
+03:00.880 --> 03:02.640
+ and this problem here,
+
+03:02.640 --> 03:06.240
+ but it also has to learn to improve
+
+03:06.240 --> 03:09.360
+ the learning algorithm itself.
+
+03:09.360 --> 03:12.480
+ So it has to have the learning algorithm
+
+03:12.480 --> 03:15.720
+ in a representation that allows it to inspect it
+
+03:15.720 --> 03:19.240
+ and modify it so that it can come up
+
+03:19.240 --> 03:22.080
+ with a better learning algorithm.
+
+03:22.080 --> 03:25.680
+ So I called that meta learning, learning to learn
+
+03:25.680 --> 03:28.040
+ and recursive self improvement.
+
+03:28.040 --> 03:29.840
+ That is really the pinnacle of that,
+
+03:29.840 --> 03:35.960
+ where you then not only learn how to improve
+
+03:35.960 --> 03:37.480
+ on that problem and on that,
+
+03:37.480 --> 03:41.080
+ but you also improve the way the machine improves
+
+03:41.080 --> 03:43.480
+ and you also improve the way it improves the way
+
+03:43.480 --> 03:45.720
+ it improves itself.
+
+03:45.720 --> 03:48.560
+ And that was my 1987 diploma thesis,
+
+03:48.560 --> 03:53.200
+ which was all about that hierarchy of meta learners
+
+03:53.200 --> 03:57.240
+ that have no computational limits
+
+03:57.240 --> 03:59.920
+ except for the well known limits
+
+03:59.920 --> 04:03.160
+ that Gödel identified in 1931
+
+04:03.160 --> 04:05.640
+ and for the limits of physics.
+
+04:06.480 --> 04:10.040
+ In the recent years, meta learning has gained popularity
+
+04:10.040 --> 04:12.760
+ in a specific kind of form.
+
+04:12.760 --> 04:16.000
+ You've talked about how that's not really meta learning
+
+04:16.000 --> 04:21.000
+ with neural networks, that's more basic transfer learning.
+
+04:21.480 --> 04:22.720
+ Can you talk about the difference
+
+04:22.720 --> 04:25.440
+ between the big general meta learning
+
+04:25.440 --> 04:27.960
+ and a more narrow sense of meta learning
+
+04:27.960 --> 04:30.880
+ the way it's used today, the way it's talked about today?
+
+04:30.880 --> 04:33.440
+ Let's take the example of a deep neural network
+
+04:33.440 --> 04:37.240
+ that has learned to classify images.
+
+04:37.240 --> 04:40.080
+ And maybe you have trained that network
+
+04:40.080 --> 04:43.800
+ on 100 different databases of images.
+
+04:43.800 --> 04:48.120
+ And now a new database comes along
+
+04:48.120 --> 04:52.000
+ and you want to quickly learn the new thing as well.
+
+04:53.400 --> 04:57.720
+ So one simple way of doing that is you take the network
+
+04:57.720 --> 05:02.440
+ which already knows 100 types of databases
+
+05:02.440 --> 05:06.320
+ and then you just take the top layer of that
+
+05:06.320 --> 05:11.320
+ and you retrain that using the new labeled data
+
+05:11.320 --> 05:14.720
+ that you have in the new image database.
+
+05:14.720 --> 05:17.320
+ And then it turns out that it really, really quickly
+
+05:17.320 --> 05:20.560
+ can learn that too, one shot basically,
+
+05:20.560 --> 05:24.280
+ because from the first 100 data sets,
+
+05:24.280 --> 05:27.520
+ it already has learned so much about computer vision
+
+05:27.520 --> 05:31.840
+ that it can reuse that and that is then almost good enough
+
+05:31.840 --> 05:34.240
+ to solve the new tasks except you need a little bit
+
+05:34.240 --> 05:37.040
+ of adjustment on the top.
+
+05:37.040 --> 05:40.200
+ So that is transfer learning
+
+05:40.200 --> 05:43.480
+ and it has been done in principle for many decades.
+
+05:43.480 --> 05:45.680
+ People have done similar things for decades.
+
+05:47.360 --> 05:49.880
+ Meta learning, true meta learning is about
+
+05:49.880 --> 05:53.880
+ having the learning algorithm itself
+
+05:54.440 --> 05:59.440
+ open to introspection by the system that is using it
+
+06:00.440 --> 06:03.760
+ and also open to modification
+
+06:03.760 --> 06:07.200
+ such that the learning system has an opportunity
+
+06:07.200 --> 06:11.400
+ to modify any part of the learning algorithm
+
+06:11.400 --> 06:16.000
+ and then evaluate the consequences of that modification
+
+06:16.000 --> 06:21.000
+ and then learn from that to create a better learning algorithm
+
+06:22.000 --> 06:23.960
+ and so on recursively.
+
+06:24.960 --> 06:27.680
+ So that's a very different animal
+
+06:27.680 --> 06:32.680
+ where you are opening the space of possible learning algorithms
+
+06:32.680 --> 06:35.520
+ to the learning system itself.
+
+06:35.520 --> 06:39.040
+ Right, so you've like in the 2004 paper,
+
+06:39.040 --> 06:42.440
+ you describe Gatal machines and programs
+
+06:42.440 --> 06:44.520
+ that rewrite themselves, right?
+
+06:44.520 --> 06:47.520
+ Philosophically and even in your paper mathematically,
+
+06:47.520 --> 06:50.000
+ these are really compelling ideas,
+
+06:50.000 --> 06:55.000
+ but practically, do you see these self referential programs
+
+06:55.320 --> 06:59.400
+ being successful in the near term to having an impact
+
+06:59.400 --> 07:03.040
+ where sort of it demonstrates to the world
+
+07:03.040 --> 07:08.040
+ that this direction is a good one to pursue in the near term?
+
+07:08.680 --> 07:11.400
+ Yes, we had these two different types
+
+07:11.400 --> 07:13.440
+ of fundamental research,
+
+07:13.440 --> 07:15.840
+ how to build a universal problem solver,
+
+07:15.840 --> 07:19.840
+ one basically exploiting proof search
+
+07:23.000 --> 07:24.960
+ and things like that that you need to come up
+
+07:24.960 --> 07:29.960
+ with asymptotically optimal, theoretically optimal
+
+07:30.320 --> 07:33.200
+ self improvers and problem solvers.
+
+07:34.200 --> 07:39.200
+ However, one has to admit that through this proof search
+
+07:40.640 --> 07:43.640
+ comes in an additive constant,
+
+07:43.640 --> 07:46.800
+ an overhead, an additive overhead
+
+07:46.800 --> 07:51.800
+ that vanishes in comparison to what you have to do
+
+07:51.800 --> 07:53.960
+ to solve large problems.
+
+07:53.960 --> 07:56.920
+ However, for many of the small problems
+
+07:56.920 --> 07:59.920
+ that we want to solve in our everyday life,
+
+07:59.920 --> 08:02.440
+ we cannot ignore this constant overhead.
+
+08:02.440 --> 08:07.440
+ And that's why we also have been doing other things,
+
+08:07.440 --> 08:11.160
+ non universal things such as recurrent neural networks
+
+08:11.160 --> 08:14.360
+ which are trained by gradient descent
+
+08:14.360 --> 08:17.600
+ and local search techniques which aren't universal at all,
+
+08:17.600 --> 08:20.280
+ which aren't provably optimal at all
+
+08:20.280 --> 08:22.000
+ like the other stuff that we did,
+
+08:22.000 --> 08:24.600
+ but which are much more practical
+
+08:24.600 --> 08:27.920
+ as long as we only want to solve the small problems
+
+08:27.920 --> 08:32.920
+ that we are typically trying to solve in this environment here.
+
+08:34.680 --> 08:38.200
+ So the universal problem solvers like the Gödel machine
+
+08:38.200 --> 08:41.320
+ but also Markus Hutter's fastest way
+
+08:41.320 --> 08:43.560
+ of solving all possible problems,
+
+08:43.560 --> 08:47.360
+ which he developed around 2002 in my lab,
+
+08:47.360 --> 08:51.280
+ they are associated with these constant overheads
+
+08:51.280 --> 08:53.280
+ for proof search, which guarantees
+
+08:53.280 --> 08:55.480
+ that the thing that you're doing is optimal.
+
+08:55.480 --> 08:59.880
+ For example, there is this fastest way
+
+08:59.880 --> 09:03.880
+ of solving all problems with a computable solution
+
+09:03.880 --> 09:05.880
+ which is due to Markus Hutter.
+
+09:05.880 --> 09:10.880
+ And to explain what's going on there,
+
+09:10.880 --> 09:13.040
+ let's take traveling salesman problems.
+
+09:14.240 --> 09:16.160
+ With traveling salesman problems,
+
+09:16.160 --> 09:20.080
+ you have a number of cities, N cities,
+
+09:20.080 --> 09:22.480
+ and you try to find the shortest path
+
+09:22.480 --> 09:26.480
+ through all these cities without visiting any city twice.
+
+09:28.480 --> 09:31.040
+ And nobody knows the fastest way
+
+09:31.040 --> 09:35.040
+ of solving traveling salesman problems, TSPs,
+
+09:37.520 --> 09:40.480
+ but let's assume there is a method of solving them
+
+09:40.480 --> 09:44.480
+ within N to the five operations
+
+09:44.480 --> 09:48.560
+ where N is the number of cities.
+
+09:50.160 --> 09:54.560
+ Then the universal method of Markus
+
+09:54.560 --> 09:58.560
+ is going to solve the same traveling salesman problem
+
+09:58.560 --> 10:02.080
+ also within N to the five steps,
+
+10:02.080 --> 10:06.360
+ plus O of one, plus a constant number of steps
+
+10:06.360 --> 10:09.240
+ that you need for the proof searcher,
+
+10:09.240 --> 10:13.800
+ which you need to show that this particular
+
+10:13.800 --> 10:17.240
+ class of problems that traveling salesman problems
+
+10:17.240 --> 10:19.360
+ can be solved within a certain time bound,
+
+10:20.520 --> 10:24.400
+ within order N to the five steps, basically.
+
+10:24.400 --> 10:28.520
+ And this additive constant doesn't care for N,
+
+10:28.520 --> 10:32.400
+ which means as N is getting larger and larger,
+
+10:32.400 --> 10:34.880
+ as you have more and more cities,
+
+10:34.880 --> 10:38.600
+ the constant overhead pales and comparison.
+
+10:38.600 --> 10:44.120
+ And that means that almost all large problems are solved
+
+10:44.120 --> 10:46.520
+ in the best possible way already today.
+
+10:46.520 --> 10:50.480
+ We already have a universal problem solved like that.
+
+10:50.480 --> 10:54.520
+ However, it's not practical because the overhead,
+
+10:54.520 --> 10:57.440
+ the constant overhead is so large
+
+10:57.440 --> 11:00.200
+ that for the small kinds of problems
+
+11:00.200 --> 11:04.560
+ that we want to solve in this little biosphere.
+
+11:04.560 --> 11:06.360
+ By the way, when you say small,
+
+11:06.360 --> 11:08.600
+ you're talking about things that fall
+
+11:08.600 --> 11:10.880
+ within the constraints of our computational systems.
+
+11:10.880 --> 11:14.280
+ So they can seem quite large to us mere humans.
+
+11:14.280 --> 11:15.360
+ That's right, yeah.
+
+11:15.360 --> 11:19.000
+ So they seem large and even unsolvable
+
+11:19.000 --> 11:21.000
+ in a practical sense today,
+
+11:21.000 --> 11:24.760
+ but they are still small compared to almost all problems
+
+11:24.760 --> 11:28.480
+ because almost all problems are large problems,
+
+11:28.480 --> 11:30.840
+ which are much larger than any constant.
+
+11:31.920 --> 11:34.520
+ Do you find it useful as a person
+
+11:34.520 --> 11:38.680
+ who is dreamed of creating a general learning system,
+
+11:38.680 --> 11:39.880
+ has worked on creating one,
+
+11:39.880 --> 11:42.160
+ has done a lot of interesting ideas there
+
+11:42.160 --> 11:46.360
+ to think about P versus NP,
+
+11:46.360 --> 11:50.800
+ this formalization of how hard problems are,
+
+11:50.800 --> 11:52.360
+ how they scale,
+
+11:52.360 --> 11:55.200
+ this kind of worst case analysis type of thinking.
+
+11:55.200 --> 11:56.840
+ Do you find that useful?
+
+11:56.840 --> 11:59.720
+ Or is it only just a mathematical,
+
+12:00.560 --> 12:02.640
+ it's a set of mathematical techniques
+
+12:02.640 --> 12:05.760
+ to give you intuition about what's good and bad?
+
+12:05.760 --> 12:09.440
+ So P versus NP, that's super interesting
+
+12:09.440 --> 12:11.800
+ from a theoretical point of view.
+
+12:11.800 --> 12:14.560
+ And in fact, as you are thinking about that problem,
+
+12:14.560 --> 12:17.280
+ you can also get inspiration
+
+12:17.280 --> 12:21.280
+ for better practical problem solvers.
+
+12:21.280 --> 12:23.320
+ On the other hand, we have to admit
+
+12:23.320 --> 12:24.560
+ that at the moment,
+
+12:24.560 --> 12:28.360
+ the best practical problem solvers
+
+12:28.360 --> 12:30.120
+ for all kinds of problems
+
+12:30.120 --> 12:33.880
+ that we are now solving through what is called AI at the moment,
+
+12:33.880 --> 12:36.240
+ they are not of the kind
+
+12:36.240 --> 12:38.800
+ that is inspired by these questions.
+
+12:38.800 --> 12:42.680
+ There we are using general purpose computers,
+
+12:42.680 --> 12:44.840
+ such as recurrent neural networks,
+
+12:44.840 --> 12:46.680
+ but we have a search technique,
+
+12:46.680 --> 12:50.320
+ which is just local search gradient descent
+
+12:50.320 --> 12:51.960
+ to try to find a program
+
+12:51.960 --> 12:54.400
+ that is running on these recurrent networks,
+
+12:54.400 --> 12:58.160
+ such that it can solve some interesting problems,
+
+12:58.160 --> 13:01.920
+ such as speech recognition or machine translation
+
+13:01.920 --> 13:03.200
+ and something like that.
+
+13:03.200 --> 13:06.480
+ And there is very little theory
+
+13:06.480 --> 13:09.720
+ behind the best solutions that we have at the moment
+
+13:09.720 --> 13:10.800
+ that can do that.
+
+13:10.800 --> 13:12.640
+ Do you think that needs to change?
+
+13:12.640 --> 13:15.120
+ Do you think that will change or can we go,
+
+13:15.120 --> 13:17.120
+ can we create a general intelligence systems
+
+13:17.120 --> 13:19.200
+ without ever really proving
+
+13:19.200 --> 13:20.600
+ that that system is intelligent
+
+13:20.600 --> 13:22.560
+ in some kind of mathematical way,
+
+13:22.560 --> 13:24.960
+ solving machine translation perfectly
+
+13:24.960 --> 13:26.320
+ or something like that,
+
+13:26.320 --> 13:29.160
+ within some kind of syntactic definition of a language?
+
+13:29.160 --> 13:31.120
+ Or can we just be super impressed
+
+13:31.120 --> 13:35.080
+ by the thing working extremely well and that's sufficient?
+
+13:35.080 --> 13:36.720
+ There's an old saying,
+
+13:36.720 --> 13:39.360
+ and I don't know who brought it up first,
+
+13:39.360 --> 13:42.440
+ which says there's nothing more practical
+
+13:42.440 --> 13:43.680
+ than a good theory.
+
+13:43.680 --> 13:48.680
+ And a good theory of problem solving
+
+13:52.760 --> 13:55.560
+ under limited resources like here in this universe
+
+13:55.560 --> 13:57.000
+ or on this little planet
+
+13:58.480 --> 14:01.800
+ has to take into account these limited resources.
+
+14:01.800 --> 14:06.800
+ And so probably there is locking a theory
+
+14:08.040 --> 14:10.800
+ which is related to what we already have,
+
+14:10.800 --> 14:14.440
+ these asymptotically optimal problem solvers,
+
+14:14.440 --> 14:18.560
+ which tells us what we need in addition to that
+
+14:18.560 --> 14:21.760
+ to come up with a practically optimal problem solver.
+
+14:21.760 --> 14:26.760
+ So I believe we will have something like that
+
+14:27.080 --> 14:29.720
+ and maybe just a few little tiny twists
+
+14:29.720 --> 14:34.320
+ are necessary to change what we already have
+
+14:34.320 --> 14:36.360
+ to come up with that as well.
+
+14:36.360 --> 14:37.800
+ As long as we don't have that,
+
+14:37.800 --> 14:42.600
+ we admit that we are taking suboptimal ways
+
+14:42.600 --> 14:46.040
+ and recurrent neural networks and long short term memory
+
+14:46.040 --> 14:50.440
+ for equipped with local search techniques
+
+14:50.440 --> 14:53.560
+ and we are happy that it works better
+
+14:53.560 --> 14:55.480
+ than any competing methods,
+
+14:55.480 --> 15:00.480
+ but that doesn't mean that we think we are done.
+
+15:00.800 --> 15:05.040
+ You've said that an AGI system will ultimately be a simple one,
+
+15:05.040 --> 15:08.000
+ a general intelligence system will ultimately be a simple one,
+
+15:08.000 --> 15:10.240
+ maybe a pseudo code of a few lines
+
+15:10.240 --> 15:11.840
+ will be able to describe it.
+
+15:11.840 --> 15:16.760
+ Can you talk through your intuition behind this idea,
+
+15:16.760 --> 15:21.760
+ why you feel that at its core intelligence
+
+15:22.120 --> 15:25.560
+ is a simple algorithm?
+
+15:26.920 --> 15:31.680
+ Experience tells us that the stuff that works best
+
+15:31.680 --> 15:33.120
+ is really simple.
+
+15:33.120 --> 15:37.640
+ So the asymptotically optimal ways of solving problems,
+
+15:37.640 --> 15:38.800
+ if you look at them,
+
+15:38.800 --> 15:41.800
+ they're just a few lines of code, it's really true.
+
+15:41.800 --> 15:44.000
+ Although they are these amazing properties,
+
+15:44.000 --> 15:45.760
+ just a few lines of code,
+
+15:45.760 --> 15:50.760
+ then the most promising and most useful practical things
+
+15:53.760 --> 15:57.760
+ maybe don't have this proof of optimality associated with them.
+
+15:57.760 --> 16:00.840
+ However, they are also just a few lines of code.
+
+16:00.840 --> 16:05.040
+ The most successful recurrent neural networks,
+
+16:05.040 --> 16:08.360
+ you can write them down and five lines of pseudo code.
+
+16:08.360 --> 16:10.920
+ That's a beautiful, almost poetic idea,
+
+16:10.920 --> 16:15.600
+ but what you're describing there
+
+16:15.600 --> 16:17.400
+ is the lines of pseudo code
+
+16:17.400 --> 16:20.600
+ are sitting on top of layers and layers of abstractions,
+
+16:20.600 --> 16:22.240
+ in a sense.
+
+16:22.240 --> 16:25.040
+ So you're saying at the very top,
+
+16:25.040 --> 16:30.040
+ it'll be a beautifully written sort of algorithm,
+
+16:31.120 --> 16:33.960
+ but do you think that there's many layers of abstractions
+
+16:33.960 --> 16:36.880
+ we have to first learn to construct?
+
+16:36.880 --> 16:38.280
+ Yeah, of course.
+
+16:38.280 --> 16:42.640
+ We are building on all these great abstractions
+
+16:42.640 --> 16:46.040
+ that people have invented over the millennia,
+
+16:46.040 --> 16:51.040
+ such as matrix multiplications and drill numbers
+
+16:51.600 --> 16:56.600
+ and basic arithmetics and calculus and derivations
+
+16:58.720 --> 17:03.320
+ of error functions and derivatives of error functions
+
+17:03.320 --> 17:04.320
+ and stuff like that.
+
+17:05.440 --> 17:10.440
+ So without that language that greatly simplifies
+
+17:10.440 --> 17:13.880
+ our way of thinking about these problems,
+
+17:13.880 --> 17:14.840
+ we couldn't do anything.
+
+17:14.840 --> 17:16.560
+ So in that sense, as always,
+
+17:16.560 --> 17:19.600
+ we are standing on the shoulders of the giants
+
+17:19.600 --> 17:24.600
+ who in the past simplified the problem of problem solving
+
+17:25.520 --> 17:30.000
+ so much that now we have a chance to do the final step.
+
+17:30.000 --> 17:32.120
+ So the final step will be a simple one.
+
+17:34.000 --> 17:36.760
+ If we take a step back through all of human civilization
+
+17:36.760 --> 17:38.360
+ and just the universe in general,
+
+17:38.360 --> 17:41.440
+ how do you think about evolution?
+
+17:41.440 --> 17:45.400
+ And what if creating a universe is required
+
+17:45.400 --> 17:47.320
+ to achieve this final step?
+
+17:47.320 --> 17:50.920
+ What if going through the very painful
+
+17:50.920 --> 17:53.840
+ and inefficient process of evolution is needed
+
+17:53.840 --> 17:55.880
+ to come up with this set of abstractions
+
+17:55.880 --> 17:57.800
+ that ultimately lead to intelligence?
+
+17:57.800 --> 18:00.800
+ Do you think there's a shortcut
+
+18:00.800 --> 18:04.640
+ or do you think we have to create something like our universe
+
+18:04.640 --> 18:09.480
+ in order to create something like human level intelligence?
+
+18:09.480 --> 18:13.160
+ So far, the only example we have is this one,
+
+18:13.160 --> 18:15.160
+ this universe in which we are living.
+
+18:15.160 --> 18:16.360
+ You think you can do better?
+
+18:20.880 --> 18:25.000
+ Maybe not, but we are part of this whole process.
+
+18:25.000 --> 18:30.000
+ So apparently, so it might be the case
+
+18:30.000 --> 18:32.160
+ that the code that runs the universe
+
+18:32.160 --> 18:33.720
+ is really, really simple.
+
+18:33.720 --> 18:36.640
+ Everything points to that possibility
+
+18:36.640 --> 18:39.960
+ because gravity and other basic forces
+
+18:39.960 --> 18:44.120
+ are really simple laws that can be easily described,
+
+18:44.120 --> 18:47.080
+ also in just a few lines of code, basically.
+
+18:47.080 --> 18:52.080
+ And then there are these other events
+
+18:52.200 --> 18:55.080
+ that the apparently random events
+
+18:55.080 --> 18:56.560
+ in the history of the universe,
+
+18:56.560 --> 18:58.800
+ which as far as we know at the moment
+
+18:58.800 --> 19:00.720
+ don't have a compact code,
+
+19:00.720 --> 19:03.240
+ but who knows, maybe somebody in the near future
+
+19:03.240 --> 19:06.800
+ is going to figure out the pseudo random generator,
+
+19:06.800 --> 19:11.800
+ which is computing whether the measurement of that
+
+19:13.520 --> 19:15.920
+ spin up or down thing here
+
+19:15.920 --> 19:18.440
+ is going to be positive or negative.
+
+19:18.440 --> 19:19.880
+ Underline quantum mechanics.
+
+19:19.880 --> 19:20.720
+ Yes, so.
+
+19:20.720 --> 19:23.160
+ Do you ultimately think quantum mechanics
+
+19:23.160 --> 19:25.200
+ is a pseudo random number generator?
+
+19:25.200 --> 19:26.920
+ So it's all deterministic.
+
+19:26.920 --> 19:28.760
+ There's no randomness in our universe.
+
+19:30.400 --> 19:31.800
+ Does God play dice?
+
+19:31.800 --> 19:34.080
+ So a couple of years ago,
+
+19:34.080 --> 19:39.080
+ a famous physicist, quantum physicist, Anton Zeilinger,
+
+19:39.080 --> 19:41.600
+ he wrote an essay in Nature,
+
+19:41.600 --> 19:44.280
+ and it started more or less like that.
+
+19:46.720 --> 19:51.720
+ One of the fundamental insights of the 20th century
+
+19:53.280 --> 19:58.280
+ was that the universe is fundamentally random
+
+19:58.280 --> 20:02.760
+ on the quantum level, and that whenever
+
+20:03.760 --> 20:06.720
+ you measure spin up or down or something like that,
+
+20:06.720 --> 20:10.720
+ a new bit of information enters the history of the universe.
+
+20:13.440 --> 20:14.680
+ And while I was reading that,
+
+20:14.680 --> 20:18.000
+ I was already typing the response
+
+20:18.000 --> 20:20.280
+ and they had to publish it because I was right,
+
+20:21.560 --> 20:25.560
+ that there is no evidence, no physical evidence for that.
+
+20:25.560 --> 20:28.440
+ So there's an alternative explanation
+
+20:28.440 --> 20:31.240
+ where everything that we consider random
+
+20:31.240 --> 20:33.800
+ is actually pseudo random,
+
+20:33.800 --> 20:38.800
+ such as the decimal expansion of pi, 3.141 and so on,
+
+20:39.400 --> 20:42.120
+ which looks random, but isn't.
+
+20:42.120 --> 20:47.120
+ So pi is interesting because every three digit sequence,
+
+20:47.720 --> 20:51.720
+ every sequence of three digits appears roughly
+
+20:51.720 --> 20:56.720
+ one in a thousand times, and every five digit sequence
+
+20:57.360 --> 21:00.760
+ appears roughly one in 10,000 times.
+
+21:00.760 --> 21:02.760
+ What do you expect?
+
+21:02.760 --> 21:06.760
+ If it was random, but there's a very short algorithm,
+
+21:06.760 --> 21:09.120
+ a short program that computes all of that.
+
+21:09.120 --> 21:11.200
+ So it's extremely compressible.
+
+21:11.200 --> 21:13.120
+ And who knows, maybe tomorrow somebody,
+
+21:13.120 --> 21:15.360
+ some grad student at CERN goes back
+
+21:15.360 --> 21:19.120
+ over all these data points, better decay,
+
+21:19.120 --> 21:21.760
+ and whatever, and figures out, oh,
+
+21:21.760 --> 21:25.760
+ it's the second billion digits of pi or something like that.
+
+21:25.760 --> 21:28.840
+ We don't have any fundamental reason at the moment
+
+21:28.840 --> 21:33.600
+ to believe that this is truly random
+
+21:33.600 --> 21:36.440
+ and not just a deterministic video game.
+
+21:36.440 --> 21:38.680
+ If it was a deterministic video game,
+
+21:38.680 --> 21:40.360
+ it would be much more beautiful
+
+21:40.360 --> 21:44.160
+ because beauty is simplicity.
+
+21:44.160 --> 21:47.560
+ And many of the basic laws of the universe
+
+21:47.560 --> 21:51.560
+ like gravity and the other basic forces are very simple.
+
+21:51.560 --> 21:55.560
+ So very short programs can explain what these are doing.
+
+21:56.560 --> 22:00.560
+ And it would be awful and ugly.
+
+22:00.560 --> 22:01.560
+ The universe would be ugly.
+
+22:01.560 --> 22:03.560
+ The history of the universe would be ugly
+
+22:03.560 --> 22:06.560
+ if for the extra things, the random,
+
+22:06.560 --> 22:10.560
+ the seemingly random data points that we get all the time
+
+22:10.560 --> 22:15.560
+ that we really need a huge number of extra bits
+
+22:15.560 --> 22:21.560
+ to describe all these extra bits of information.
+
+22:22.560 --> 22:25.560
+ So as long as we don't have evidence
+
+22:25.560 --> 22:27.560
+ that there is no short program
+
+22:27.560 --> 22:32.560
+ that computes the entire history of the entire universe,
+
+22:32.560 --> 22:38.560
+ we are, as scientists, compelled to look further
+
+22:38.560 --> 22:41.560
+ for that shortest program.
+
+22:41.560 --> 22:46.560
+ Your intuition says there exists a program
+
+22:46.560 --> 22:50.560
+ that can backtrack to the creation of the universe.
+
+22:50.560 --> 22:53.560
+ So it can take the shortest path to the creation of the universe.
+
+22:53.560 --> 22:57.560
+ Yes, including all the entanglement things
+
+22:57.560 --> 23:01.560
+ and all the spin up and down measurements
+
+23:01.560 --> 23:09.560
+ that have been taken place since 13.8 billion years ago.
+
+23:09.560 --> 23:14.560
+ So we don't have a proof that it is random.
+
+23:14.560 --> 23:19.560
+ We don't have a proof that it is compressible to a short program.
+
+23:19.560 --> 23:21.560
+ But as long as we don't have that proof,
+
+23:21.560 --> 23:24.560
+ we are obliged as scientists to keep looking
+
+23:24.560 --> 23:26.560
+ for that simple explanation.
+
+23:26.560 --> 23:27.560
+ Absolutely.
+
+23:27.560 --> 23:30.560
+ So you said simplicity is beautiful or beauty is simple.
+
+23:30.560 --> 23:32.560
+ Either one works.
+
+23:32.560 --> 23:36.560
+ But you also work on curiosity, discovery.
+
+23:36.560 --> 23:42.560
+ The romantic notion of randomness, of serendipity,
+
+23:42.560 --> 23:49.560
+ of being surprised by things that are about you,
+
+23:49.560 --> 23:53.560
+ kind of in our poetic notion of reality,
+
+23:53.560 --> 23:56.560
+ we think as humans require randomness.
+
+23:56.560 --> 23:58.560
+ So you don't find randomness beautiful.
+
+23:58.560 --> 24:04.560
+ You find simple determinism beautiful.
+
+24:04.560 --> 24:06.560
+ Yeah.
+
+24:06.560 --> 24:07.560
+ Okay.
+
+24:07.560 --> 24:08.560
+ So why?
+
+24:08.560 --> 24:09.560
+ Why?
+
+24:09.560 --> 24:12.560
+ Because the explanation becomes shorter.
+
+24:12.560 --> 24:19.560
+ A universe that is compressible to a short program
+
+24:19.560 --> 24:22.560
+ is much more elegant and much more beautiful
+
+24:22.560 --> 24:24.560
+ than another one,
+
+24:24.560 --> 24:28.560
+ which needs an almost infinite number of bits to be described.
+
+24:28.560 --> 24:31.560
+ As far as we know,
+
+24:31.560 --> 24:34.560
+ many things that are happening in this universe are really simple
+
+24:34.560 --> 24:38.560
+ in terms of short programs that compute gravity
+
+24:38.560 --> 24:43.560
+ and the interaction between elementary particles and so on.
+
+24:43.560 --> 24:45.560
+ So all of that seems to be very, very simple.
+
+24:45.560 --> 24:50.560
+ Every electron seems to reuse the same subprogram all the time
+
+24:50.560 --> 24:57.560
+ as it is interacting with other elementary particles.
+
+24:57.560 --> 25:04.560
+ If we now require an extra oracle
+
+25:04.560 --> 25:07.560
+ injecting new bits of information all the time
+
+25:07.560 --> 25:11.560
+ for these extra things which are currently not understood,
+
+25:11.560 --> 25:18.560
+ such as better decay,
+
+25:18.560 --> 25:25.560
+ then the whole description length of the data that we can observe
+
+25:25.560 --> 25:31.560
+ of the history of the universe would become much longer.
+
+25:31.560 --> 25:33.560
+ And therefore, uglier.
+
+25:33.560 --> 25:34.560
+ And uglier.
+
+25:34.560 --> 25:38.560
+ Again, the simplicity is elegant and beautiful.
+
+25:38.560 --> 25:42.560
+ All the history of science is a history of compression progress.
+
+25:42.560 --> 25:43.560
+ Yeah.
+
+25:43.560 --> 25:48.560
+ So you've described sort of as we build up abstractions
+
+25:48.560 --> 25:52.560
+ and you've talked about the idea of compression.
+
+25:52.560 --> 25:55.560
+ How do you see this, the history of science,
+
+25:55.560 --> 25:59.560
+ the history of humanity, our civilization and life on Earth
+
+25:59.560 --> 26:03.560
+ as some kind of path towards greater and greater compression?
+
+26:03.560 --> 26:04.560
+ What do you mean by that?
+
+26:04.560 --> 26:06.560
+ How do you think about that?
+
+26:06.560 --> 26:12.560
+ Indeed, the history of science is a history of compression progress.
+
+26:12.560 --> 26:14.560
+ What does that mean?
+
+26:14.560 --> 26:17.560
+ Hundreds of years ago, there was an astronomer
+
+26:17.560 --> 26:19.560
+ whose name was Kepler.
+
+26:19.560 --> 26:25.560
+ And he looked at the data points that he got by watching planets move.
+
+26:25.560 --> 26:28.560
+ And then he had all these data points and suddenly it turned out
+
+26:28.560 --> 26:37.560
+ that he can greatly compress the data by predicting it through an ellipse law.
+
+26:37.560 --> 26:44.560
+ So it turns out that all these data points are more or less on ellipses around the sun.
+
+26:44.560 --> 26:50.560
+ And another guy came along whose name was Newton and before him Hook.
+
+26:50.560 --> 26:57.560
+ And they said the same thing that is making these planets move like that
+
+26:57.560 --> 27:01.560
+ is what makes the apples fall down.
+
+27:01.560 --> 27:10.560
+ And it also holds for stones and for all kinds of other objects.
+
+27:10.560 --> 27:16.560
+ And suddenly many, many of these observations became much more compressible
+
+27:16.560 --> 27:19.560
+ because as long as you can predict the next thing,
+
+27:19.560 --> 27:22.560
+ given what you have seen so far, you can compress it.
+
+27:22.560 --> 27:24.560
+ But you don't have to store that data extra.
+
+27:24.560 --> 27:28.560
+ This is called predictive coding.
+
+27:28.560 --> 27:33.560
+ And then there was still something wrong with that theory of the universe
+
+27:33.560 --> 27:37.560
+ and you had deviations from these predictions of the theory.
+
+27:37.560 --> 27:41.560
+ And 300 years later another guy came along whose name was Einstein
+
+27:41.560 --> 27:50.560
+ and he was able to explain away all these deviations from the predictions of the old theory
+
+27:50.560 --> 27:56.560
+ through a new theory which was called the general theory of relativity
+
+27:56.560 --> 28:00.560
+ which at first glance looks a little bit more complicated
+
+28:00.560 --> 28:05.560
+ and you have to warp space and time but you can't phrase it within one single sentence
+
+28:05.560 --> 28:12.560
+ which is no matter how fast you accelerate and how fast or how hard you decelerate
+
+28:12.560 --> 28:18.560
+ and no matter what is the gravity in your local framework,
+
+28:18.560 --> 28:21.560
+ light speed always looks the same.
+
+28:21.560 --> 28:24.560
+ And from that you can calculate all the consequences.
+
+28:24.560 --> 28:30.560
+ So it's a very simple thing and it allows you to further compress all the observations
+
+28:30.560 --> 28:35.560
+ because certainly there are hardly any deviations any longer
+
+28:35.560 --> 28:39.560
+ that you can measure from the predictions of this new theory.
+
+28:39.560 --> 28:44.560
+ So art of science is a history of compression progress.
+
+28:44.560 --> 28:50.560
+ You never arrive immediately at the shortest explanation of the data
+
+28:50.560 --> 28:52.560
+ but you're making progress.
+
+28:52.560 --> 28:56.560
+ Whenever you are making progress you have an insight.
+
+28:56.560 --> 29:01.560
+ You see, oh, first I needed so many bits of information to describe the data,
+
+29:01.560 --> 29:04.560
+ to describe my falling apples, my video of falling apples,
+
+29:04.560 --> 29:08.560
+ I need so many data, so many pixels have to be stored
+
+29:08.560 --> 29:14.560
+ but then suddenly I realize, no, there is a very simple way of predicting the third frame
+
+29:14.560 --> 29:20.560
+ in the video from the first two and maybe not every little detail can be predicted
+
+29:20.560 --> 29:24.560
+ but more or less most of these orange blots that are coming down,
+
+29:24.560 --> 29:28.560
+ I'm sorry, in the same way, which means that I can greatly compress the video
+
+29:28.560 --> 29:33.560
+ and the amount of compression, progress,
+
+29:33.560 --> 29:37.560
+ that is the depth of the insight that you have at that moment.
+
+29:37.560 --> 29:40.560
+ That's the fun that you have, the scientific fun,
+
+29:40.560 --> 29:46.560
+ the fun in that discovery and we can build artificial systems that do the same thing.
+
+29:46.560 --> 29:51.560
+ They measure the depth of their insights as they are looking at the data
+
+29:51.560 --> 29:55.560
+ through their own experiments and we give them a reward,
+
+29:55.560 --> 30:00.560
+ an intrinsic reward and proportion to this depth of insight.
+
+30:00.560 --> 30:07.560
+ And since they are trying to maximize the rewards they get,
+
+30:07.560 --> 30:12.560
+ they are suddenly motivated to come up with new action sequences,
+
+30:12.560 --> 30:17.560
+ with new experiments that have the property that the data that is coming in
+
+30:17.560 --> 30:21.560
+ as a consequence of these experiments has the property
+
+30:21.560 --> 30:25.560
+ that they can learn something about, see a pattern in there
+
+30:25.560 --> 30:28.560
+ which they hadn't seen yet before.
+
+30:28.560 --> 30:32.560
+ So there's an idea of power play that you've described,
+
+30:32.560 --> 30:37.560
+ a training and general problem solver in this kind of way of looking for the unsolved problems.
+
+30:37.560 --> 30:40.560
+ Can you describe that idea a little further?
+
+30:40.560 --> 30:42.560
+ It's another very simple idea.
+
+30:42.560 --> 30:49.560
+ Normally what you do in computer science, you have some guy who gives you a problem
+
+30:49.560 --> 30:56.560
+ and then there is a huge search space of potential solution candidates
+
+30:56.560 --> 31:02.560
+ and you somehow try them out and you have more or less sophisticated ways
+
+31:02.560 --> 31:06.560
+ of moving around in that search space
+
+31:06.560 --> 31:11.560
+ until you finally found a solution which you consider satisfactory.
+
+31:11.560 --> 31:15.560
+ That's what most of computer science is about.
+
+31:15.560 --> 31:19.560
+ Power play just goes one little step further and says,
+
+31:19.560 --> 31:24.560
+ let's not only search for solutions to a given problem,
+
+31:24.560 --> 31:30.560
+ but let's search to pairs of problems and their solutions
+
+31:30.560 --> 31:36.560
+ where the system itself has the opportunity to phrase its own problem.
+
+31:36.560 --> 31:42.560
+ So we are looking suddenly at pairs of problems and their solutions
+
+31:42.560 --> 31:46.560
+ or modifications of the problem solver
+
+31:46.560 --> 31:50.560
+ that is supposed to generate a solution to that new problem.
+
+31:50.560 --> 31:56.560
+ And this additional degree of freedom
+
+31:56.560 --> 32:01.560
+ allows us to build career systems that are like scientists
+
+32:01.560 --> 32:06.560
+ in the sense that they not only try to solve and try to find answers
+
+32:06.560 --> 32:12.560
+ to existing questions, no, they are also free to pose their own questions.
+
+32:12.560 --> 32:15.560
+ So if you want to build an artificial scientist,
+
+32:15.560 --> 32:19.560
+ you have to give it that freedom and power play is exactly doing that.
+
+32:19.560 --> 32:23.560
+ So that's a dimension of freedom that's important to have,
+
+32:23.560 --> 32:31.560
+ how hard do you think that, how multi dimensional and difficult the space of
+
+32:31.560 --> 32:34.560
+ then coming up with your own questions is.
+
+32:34.560 --> 32:38.560
+ So it's one of the things that as human beings we consider to be
+
+32:38.560 --> 32:41.560
+ the thing that makes us special, the intelligence that makes us special
+
+32:41.560 --> 32:47.560
+ is that brilliant insight that can create something totally new.
+
+32:47.560 --> 32:51.560
+ Yes. So now let's look at the extreme case.
+
+32:51.560 --> 32:57.560
+ Let's look at the set of all possible problems that you can formally describe,
+
+32:57.560 --> 33:03.560
+ which is infinite, which should be the next problem
+
+33:03.560 --> 33:07.560
+ that a scientist or power play is going to solve.
+
+33:07.560 --> 33:16.560
+ Well, it should be the easiest problem that goes beyond what you already know.
+
+33:16.560 --> 33:22.560
+ So it should be the simplest problem that the current problems
+
+33:22.560 --> 33:28.560
+ that you have which can already solve 100 problems that he cannot solve yet
+
+33:28.560 --> 33:30.560
+ by just generalizing.
+
+33:30.560 --> 33:32.560
+ So it has to be new.
+
+33:32.560 --> 33:36.560
+ So it has to require a modification of the problem solver such that the new
+
+33:36.560 --> 33:41.560
+ problem solver can solve this new thing, but the old problem solver cannot do it.
+
+33:41.560 --> 33:47.560
+ And in addition to that, we have to make sure that the problem solver
+
+33:47.560 --> 33:50.560
+ doesn't forget any of the previous solutions.
+
+33:50.560 --> 33:51.560
+ Right.
+
+33:51.560 --> 33:57.560
+ And so by definition, power play is now trying always to search in this pair of
+
+33:57.560 --> 34:02.560
+ in the set of pairs of problems and problems over modifications
+
+34:02.560 --> 34:08.560
+ for a combination that minimize the time to achieve these criteria.
+
+34:08.560 --> 34:14.560
+ Power is trying to find the problem which is easiest to add to the repertoire.
+
+34:14.560 --> 34:19.560
+ So just like grad students and academics and researchers can spend their whole
+
+34:19.560 --> 34:25.560
+ career in a local minima stuck trying to come up with interesting questions,
+
+34:25.560 --> 34:27.560
+ but ultimately doing very little.
+
+34:27.560 --> 34:32.560
+ Do you think it's easy in this approach of looking for the simplest
+
+34:32.560 --> 34:38.560
+ problem solver problem to get stuck in a local minima is not never really discovering
+
+34:38.560 --> 34:43.560
+ new, you know, really jumping outside of the hundred problems that you've already
+
+34:43.560 --> 34:47.560
+ solved in a genuine creative way.
+
+34:47.560 --> 34:52.560
+ No, because that's the nature of power play that it's always trying to break
+
+34:52.560 --> 34:58.560
+ its current generalization abilities by coming up with a new problem which is
+
+34:58.560 --> 35:04.560
+ beyond the current horizon, just shifting the horizon of knowledge a little bit
+
+35:04.560 --> 35:10.560
+ out there, breaking the existing rules such that the new thing becomes solvable
+
+35:10.560 --> 35:13.560
+ but wasn't solvable by the old thing.
+
+35:13.560 --> 35:19.560
+ So like adding a new axiom, like what Gödel did when he came up with these
+
+35:19.560 --> 35:23.560
+ new sentences, new theorems that didn't have a proof in the formal system,
+
+35:23.560 --> 35:30.560
+ which means you can add them to the repertoire, hoping that they are not
+
+35:30.560 --> 35:35.560
+ going to damage the consistency of the whole thing.
+
+35:35.560 --> 35:41.560
+ So in the paper with the amazing title, Formal Theory of Creativity,
+
+35:41.560 --> 35:47.560
+ Fun and Intrinsic Motivation, you talk about discovery as intrinsic reward.
+
+35:47.560 --> 35:53.560
+ So if you view humans as intelligent agents, what do you think is the purpose
+
+35:53.560 --> 35:56.560
+ and meaning of life for us humans?
+
+35:56.560 --> 35:58.560
+ You've talked about this discovery.
+
+35:58.560 --> 36:04.560
+ Do you see humans as an instance of power play agents?
+
+36:04.560 --> 36:11.560
+ Yeah, so humans are curious and that means they behave like scientists,
+
+36:11.560 --> 36:15.560
+ not only the official scientists but even the babies behave like scientists
+
+36:15.560 --> 36:19.560
+ and they play around with their toys to figure out how the world works
+
+36:19.560 --> 36:22.560
+ and how it is responding to their actions.
+
+36:22.560 --> 36:26.560
+ And that's how they learn about gravity and everything.
+
+36:26.560 --> 36:30.560
+ And yeah, in 1990, we had the first systems like that
+
+36:30.560 --> 36:33.560
+ who would just try to play around with the environment
+
+36:33.560 --> 36:39.560
+ and come up with situations that go beyond what they knew at that time
+
+36:39.560 --> 36:42.560
+ and then get a reward for creating these situations
+
+36:42.560 --> 36:45.560
+ and then becoming more general problem solvers
+
+36:45.560 --> 36:48.560
+ and being able to understand more of the world.
+
+36:48.560 --> 36:56.560
+ So yeah, I think in principle that curiosity,
+
+36:56.560 --> 37:02.560
+ strategy or more sophisticated versions of what I just described,
+
+37:02.560 --> 37:07.560
+ they are what we have built in as well because evolution discovered
+
+37:07.560 --> 37:12.560
+ that's a good way of exploring the unknown world and a guy who explores
+
+37:12.560 --> 37:16.560
+ the unknown world has a higher chance of solving problems
+
+37:16.560 --> 37:19.560
+ that he needs to survive in this world.
+
+37:19.560 --> 37:23.560
+ On the other hand, those guys who were too curious,
+
+37:23.560 --> 37:25.560
+ they were weeded out as well.
+
+37:25.560 --> 37:27.560
+ So you have to find this trade off.
+
+37:27.560 --> 37:30.560
+ Evolution found a certain trade off apparently in our society.
+
+37:30.560 --> 37:35.560
+ There is a certain percentage of extremely explorative guys
+
+37:35.560 --> 37:41.560
+ and it doesn't matter if they die because many of the others are more conservative.
+
+37:41.560 --> 37:46.560
+ And so yeah, it would be surprising to me
+
+37:46.560 --> 37:55.560
+ if that principle of artificial curiosity wouldn't be present
+
+37:55.560 --> 37:59.560
+ in almost exactly the same form here in our brains.
+
+37:59.560 --> 38:02.560
+ So you're a bit of a musician and an artist.
+
+38:02.560 --> 38:07.560
+ So continuing on this topic of creativity,
+
+38:07.560 --> 38:10.560
+ what do you think is the role of creativity in intelligence?
+
+38:10.560 --> 38:16.560
+ So you've kind of implied that it's essential for intelligence,
+
+38:16.560 --> 38:21.560
+ if you think of intelligence as a problem solving system,
+
+38:21.560 --> 38:23.560
+ as ability to solve problems.
+
+38:23.560 --> 38:28.560
+ But do you think it's essential, this idea of creativity?
+
+38:28.560 --> 38:34.560
+ We never have a subprogram that is called creativity or something.
+
+38:34.560 --> 38:37.560
+ It's just a side effect of what our problems always do.
+
+38:37.560 --> 38:44.560
+ They are searching a space of candidates, of solution candidates,
+
+38:44.560 --> 38:47.560
+ until they hopefully find a solution to a given problem.
+
+38:47.560 --> 38:50.560
+ But then there are these two types of creativity
+
+38:50.560 --> 38:53.560
+ and both of them are now present in our machines.
+
+38:53.560 --> 38:56.560
+ The first one has been around for a long time,
+
+38:56.560 --> 38:59.560
+ which is human gives problem to machine.
+
+38:59.560 --> 39:03.560
+ Machine tries to find a solution to that.
+
+39:03.560 --> 39:05.560
+ And this has been happening for many decades.
+
+39:05.560 --> 39:09.560
+ And for many decades, machines have found creative solutions
+
+39:09.560 --> 39:13.560
+ to interesting problems where humans were not aware
+
+39:13.560 --> 39:17.560
+ of these particularly creative solutions,
+
+39:17.560 --> 39:20.560
+ but then appreciated that the machine found that.
+
+39:20.560 --> 39:23.560
+ The second is the pure creativity.
+
+39:23.560 --> 39:28.560
+ What I just mentioned, I would call the applied creativity,
+
+39:28.560 --> 39:31.560
+ like applied art, where somebody tells you,
+
+39:31.560 --> 39:34.560
+ now make a nice picture of this pope,
+
+39:34.560 --> 39:36.560
+ and you will get money for that.
+
+39:36.560 --> 39:41.560
+ So here is the artist and he makes a convincing picture of the pope
+
+39:41.560 --> 39:44.560
+ and the pope likes it and gives him the money.
+
+39:44.560 --> 39:48.560
+ And then there is the pure creativity,
+
+39:48.560 --> 39:51.560
+ which is more like the power play and the artificial curiosity thing,
+
+39:51.560 --> 39:56.560
+ where you have the freedom to select your own problem,
+
+39:56.560 --> 40:02.560
+ like a scientist who defines his own question to study.
+
+40:02.560 --> 40:06.560
+ And so that is the pure creativity, if you will,
+
+40:06.560 --> 40:13.560
+ as opposed to the applied creativity, which serves another.
+
+40:13.560 --> 40:18.560
+ In that distinction, there's almost echoes of narrow AI versus general AI.
+
+40:18.560 --> 40:24.560
+ So this kind of constrained painting of a pope seems like
+
+40:24.560 --> 40:29.560
+ the approaches of what people are calling narrow AI.
+
+40:29.560 --> 40:32.560
+ And pure creativity seems to be,
+
+40:32.560 --> 40:34.560
+ maybe I'm just biased as a human,
+
+40:34.560 --> 40:40.560
+ but it seems to be an essential element of human level intelligence.
+
+40:40.560 --> 40:43.560
+ Is that what you're implying?
+
+40:43.560 --> 40:45.560
+ To a degree.
+
+40:45.560 --> 40:50.560
+ If you zoom back a little bit and you just look at a general problem solving machine,
+
+40:50.560 --> 40:53.560
+ which is trying to solve arbitrary problems,
+
+40:53.560 --> 40:57.560
+ then this machine will figure out in the course of solving problems
+
+40:57.560 --> 40:59.560
+ that it's good to be curious.
+
+40:59.560 --> 41:04.560
+ So all of what I said just now about this pre wild curiosity
+
+41:04.560 --> 41:10.560
+ and this will to invent new problems that the system doesn't know how to solve yet,
+
+41:10.560 --> 41:14.560
+ should be just a byproduct of the general search.
+
+41:14.560 --> 41:21.560
+ However, apparently evolution has built it into us
+
+41:21.560 --> 41:26.560
+ because it turned out to be so successful, a pre wiring, a bias,
+
+41:26.560 --> 41:33.560
+ a very successful exploratory bias that we are born with.
+
+41:33.560 --> 41:36.560
+ And you've also said that consciousness in the same kind of way
+
+41:36.560 --> 41:40.560
+ may be a byproduct of problem solving.
+
+41:40.560 --> 41:44.560
+ Do you find this an interesting byproduct?
+
+41:44.560 --> 41:46.560
+ Do you think it's a useful byproduct?
+
+41:46.560 --> 41:49.560
+ What are your thoughts on consciousness in general?
+
+41:49.560 --> 41:54.560
+ Or is it simply a byproduct of greater and greater capabilities of problem solving
+
+41:54.560 --> 42:00.560
+ that's similar to creativity in that sense?
+
+42:00.560 --> 42:04.560
+ We never have a procedure called consciousness in our machines.
+
+42:04.560 --> 42:10.560
+ However, we get a side effects of what these machines are doing,
+
+42:10.560 --> 42:15.560
+ things that seem to be closely related to what people call consciousness.
+
+42:15.560 --> 42:20.560
+ So for example, already 1990 we had simple systems
+
+42:20.560 --> 42:25.560
+ which were basically recurrent networks and therefore universal computers
+
+42:25.560 --> 42:32.560
+ trying to map incoming data into actions that lead to success.
+
+42:32.560 --> 42:39.560
+ Maximizing reward in a given environment, always finding the charging station in time
+
+42:39.560 --> 42:43.560
+ whenever the battery is low and negative signals are coming from the battery,
+
+42:43.560 --> 42:50.560
+ always find the charging station in time without bumping against painful obstacles on the way.
+
+42:50.560 --> 42:54.560
+ So complicated things but very easily motivated.
+
+42:54.560 --> 43:01.560
+ And then we give these little guys a separate recurrent network
+
+43:01.560 --> 43:04.560
+ which is just predicting what's happening if I do that and that.
+
+43:04.560 --> 43:08.560
+ What will happen as a consequence of these actions that I'm executing
+
+43:08.560 --> 43:13.560
+ and it's just trained on the long and long history of interactions with the world.
+
+43:13.560 --> 43:17.560
+ So it becomes a predictive model of the world basically.
+
+43:17.560 --> 43:22.560
+ And therefore also a compressor of the observations of the world
+
+43:22.560 --> 43:26.560
+ because whatever you can predict, you don't have to store extra.
+
+43:26.560 --> 43:29.560
+ So compression is a side effect of prediction.
+
+43:29.560 --> 43:32.560
+ And how does this recurrent network compress?
+
+43:32.560 --> 43:36.560
+ Well, it's inventing little subprograms, little subnetworks
+
+43:36.560 --> 43:41.560
+ that stand for everything that frequently appears in the environment.
+
+43:41.560 --> 43:47.560
+ Like bottles and microphones and faces, maybe lots of faces in my environment.
+
+43:47.560 --> 43:51.560
+ So I'm learning to create something like a prototype face
+
+43:51.560 --> 43:55.560
+ and a new face comes along and all I have to encode are the deviations from the prototype.
+
+43:55.560 --> 44:00.560
+ So it's compressing all the time the stuff that frequently appears.
+
+44:00.560 --> 44:04.560
+ There's one thing that appears all the time
+
+44:04.560 --> 44:09.560
+ that is present all the time when the agent is interacting with its environment,
+
+44:09.560 --> 44:11.560
+ which is the agent itself.
+
+44:11.560 --> 44:14.560
+ So just for data compression reasons,
+
+44:14.560 --> 44:18.560
+ it is extremely natural for this recurrent network
+
+44:18.560 --> 44:23.560
+ to come up with little subnetworks that stand for the properties of the agents,
+
+44:23.560 --> 44:27.560
+ the hand, the other actuators,
+
+44:27.560 --> 44:31.560
+ and all the stuff that you need to better encode the data,
+
+44:31.560 --> 44:34.560
+ which is influenced by the actions of the agent.
+
+44:34.560 --> 44:40.560
+ So there, just as a side effect of data compression during primal solving,
+
+44:40.560 --> 44:45.560
+ you have internal self models.
+
+44:45.560 --> 44:51.560
+ Now you can use this model of the world to plan your future.
+
+44:51.560 --> 44:54.560
+ And that's what we also have done since 1990.
+
+44:54.560 --> 44:57.560
+ So the recurrent network, which is the controller,
+
+44:57.560 --> 44:59.560
+ which is trying to maximize reward,
+
+44:59.560 --> 45:02.560
+ can use this model of the network of the world,
+
+45:02.560 --> 45:05.560
+ this model network of the world, this predictive model of the world
+
+45:05.560 --> 45:08.560
+ to plan ahead and say, let's not do this action sequence.
+
+45:08.560 --> 45:11.560
+ Let's do this action sequence instead
+
+45:11.560 --> 45:14.560
+ because it leads to more predicted rewards.
+
+45:14.560 --> 45:19.560
+ And whenever it's waking up these little subnetworks that stand for itself,
+
+45:19.560 --> 45:21.560
+ then it's thinking about itself.
+
+45:21.560 --> 45:23.560
+ Then it's thinking about itself.
+
+45:23.560 --> 45:30.560
+ And it's exploring mentally the consequences of its own actions.
+
+45:30.560 --> 45:36.560
+ And now you tell me why it's still missing.
+
+45:36.560 --> 45:39.560
+ Missing the gap to consciousness.
+
+45:39.560 --> 45:43.560
+ There isn't. That's a really beautiful idea that, you know,
+
+45:43.560 --> 45:46.560
+ if life is a collection of data
+
+45:46.560 --> 45:53.560
+ and life is a process of compressing that data to act efficiently.
+
+45:53.560 --> 45:57.560
+ In that data, you yourself appear very often.
+
+45:57.560 --> 46:00.560
+ So it's useful to form compressions of yourself.
+
+46:00.560 --> 46:03.560
+ And it's a really beautiful formulation of what consciousness is,
+
+46:03.560 --> 46:05.560
+ is a necessary side effect.
+
+46:05.560 --> 46:11.560
+ It's actually quite compelling to me.
+
+46:11.560 --> 46:18.560
+ We've described RNNs, developed LSTMs, long short term memory networks.
+
+46:18.560 --> 46:22.560
+ They're a type of recurrent neural networks.
+
+46:22.560 --> 46:24.560
+ They've gotten a lot of success recently.
+
+46:24.560 --> 46:29.560
+ So these are networks that model the temporal aspects in the data,
+
+46:29.560 --> 46:31.560
+ temporal patterns in the data.
+
+46:31.560 --> 46:36.560
+ And you've called them the deepest of the neural networks, right?
+
+46:36.560 --> 46:43.560
+ What do you think is the value of depth in the models that we use to learn?
+
+46:43.560 --> 46:47.560
+ Yeah, since you mentioned the long short term memory and the LSTM,
+
+46:47.560 --> 46:52.560
+ I have to mention the names of the brilliant students who made that possible.
+
+46:52.560 --> 46:53.560
+ Yes, of course, of course.
+
+46:53.560 --> 46:56.560
+ First of all, my first student ever, Sepp Hochreiter,
+
+46:56.560 --> 47:00.560
+ who had fundamental insights already in his diploma thesis.
+
+47:00.560 --> 47:04.560
+ Then Felix Giers, who had additional important contributions.
+
+47:04.560 --> 47:11.560
+ Alex Gray is a guy from Scotland who is mostly responsible for this CTC algorithm,
+
+47:11.560 --> 47:16.560
+ which is now often used to train the LSTM to do the speech recognition
+
+47:16.560 --> 47:21.560
+ on all the Google Android phones and whatever, and Siri and so on.
+
+47:21.560 --> 47:26.560
+ So these guys, without these guys, I would be nothing.
+
+47:26.560 --> 47:28.560
+ It's a lot of incredible work.
+
+47:28.560 --> 47:30.560
+ What is now the depth?
+
+47:30.560 --> 47:32.560
+ What is the importance of depth?
+
+47:32.560 --> 47:36.560
+ Well, most problems in the real world are deep
+
+47:36.560 --> 47:41.560
+ in the sense that the current input doesn't tell you all you need to know
+
+47:41.560 --> 47:44.560
+ about the environment.
+
+47:44.560 --> 47:49.560
+ So instead, you have to have a memory of what happened in the past
+
+47:49.560 --> 47:54.560
+ and often important parts of that memory are dated.
+
+47:54.560 --> 47:56.560
+ They are pretty old.
+
+47:56.560 --> 47:59.560
+ So when you're doing speech recognition, for example,
+
+47:59.560 --> 48:03.560
+ and somebody says 11,
+
+48:03.560 --> 48:08.560
+ then that's about half a second or something like that,
+
+48:08.560 --> 48:11.560
+ which means it's already 50 time steps.
+
+48:11.560 --> 48:15.560
+ And another guy or the same guy says 7.
+
+48:15.560 --> 48:18.560
+ So the ending is the same, even.
+
+48:18.560 --> 48:22.560
+ But now the system has to see the distinction between 7 and 11,
+
+48:22.560 --> 48:26.560
+ and the only way it can see the difference is it has to store
+
+48:26.560 --> 48:34.560
+ that 50 steps ago there was an S or an L, 11 or 7.
+
+48:34.560 --> 48:37.560
+ So there you have already a problem of depth 50,
+
+48:37.560 --> 48:42.560
+ because for each time step you have something like a virtual layer
+
+48:42.560 --> 48:45.560
+ and the expanded, unrolled version of this recurrent network
+
+48:45.560 --> 48:47.560
+ which is doing the speech recognition.
+
+48:47.560 --> 48:53.560
+ So these long time lags, they translate into problem depth.
+
+48:53.560 --> 48:59.560
+ And most problems in this world are such that you really
+
+48:59.560 --> 49:04.560
+ have to look far back in time to understand what is the problem
+
+49:04.560 --> 49:06.560
+ and to solve it.
+
+49:06.560 --> 49:09.560
+ But just like with LSTMs, you don't necessarily need to,
+
+49:09.560 --> 49:12.560
+ when you look back in time, remember every aspect.
+
+49:12.560 --> 49:14.560
+ You just need to remember the important aspects.
+
+49:14.560 --> 49:15.560
+ That's right.
+
+49:15.560 --> 49:19.560
+ The network has to learn to put the important stuff into memory
+
+49:19.560 --> 49:23.560
+ and to ignore the unimportant noise.
+
+49:23.560 --> 49:28.560
+ But in that sense, deeper and deeper is better?
+
+49:28.560 --> 49:30.560
+ Or is there a limitation?
+
+49:30.560 --> 49:36.560
+ I mean LSTM is one of the great examples of architectures
+
+49:36.560 --> 49:41.560
+ that do something beyond just deeper and deeper networks.
+
+49:41.560 --> 49:47.560
+ There's clever mechanisms for filtering data for remembering and forgetting.
+
+49:47.560 --> 49:51.560
+ So do you think that kind of thinking is necessary?
+
+49:51.560 --> 49:54.560
+ If you think about LSTMs as a leap, a big leap forward
+
+49:54.560 --> 50:01.560
+ over traditional vanilla RNNs, what do you think is the next leap
+
+50:01.560 --> 50:03.560
+ within this context?
+
+50:03.560 --> 50:08.560
+ So LSTM is a very clever improvement, but LSTMs still don't
+
+50:08.560 --> 50:13.560
+ have the same kind of ability to see far back in the past
+
+50:13.560 --> 50:18.560
+ as humans do, the credit assignment problem across way back,
+
+50:18.560 --> 50:24.560
+ not just 50 time steps or 100 or 1,000, but millions and billions.
+
+50:24.560 --> 50:28.560
+ It's not clear what are the practical limits of the LSTM
+
+50:28.560 --> 50:30.560
+ when it comes to looking back.
+
+50:30.560 --> 50:35.560
+ Already in 2006, I think, we had examples where not only
+
+50:35.560 --> 50:40.560
+ looked back tens of thousands of steps, but really millions of steps.
+
+50:40.560 --> 50:46.560
+ And Juan Perez Ortiz in my lab, I think was the first author of a paper
+
+50:46.560 --> 50:51.560
+ where we really, was it 2006 or something, had examples where it
+
+50:51.560 --> 50:56.560
+ learned to look back for more than 10 million steps.
+
+50:56.560 --> 51:02.560
+ So for most problems of speech recognition, it's not
+
+51:02.560 --> 51:06.560
+ necessary to look that far back, but there are examples where it does.
+
+51:06.560 --> 51:12.560
+ Now, the looking back thing, that's rather easy because there is only
+
+51:12.560 --> 51:17.560
+ one past, but there are many possible futures.
+
+51:17.560 --> 51:21.560
+ And so a reinforcement learning system, which is trying to maximize
+
+51:21.560 --> 51:26.560
+ its future expected reward and doesn't know yet which of these
+
+51:26.560 --> 51:31.560
+ many possible futures should I select, given this one single past,
+
+51:31.560 --> 51:36.560
+ is facing problems that the LSTM by itself cannot solve.
+
+51:36.560 --> 51:40.560
+ So the LSTM is good for coming up with a compact representation
+
+51:40.560 --> 51:46.560
+ of the history so far, of the history and of observations and actions so far.
+
+51:46.560 --> 51:53.560
+ But now, how do you plan in an efficient and good way among all these,
+
+51:53.560 --> 51:57.560
+ how do you select one of these many possible action sequences
+
+51:57.560 --> 52:02.560
+ that a reinforcement learning system has to consider to maximize
+
+52:02.560 --> 52:05.560
+ reward in this unknown future.
+
+52:05.560 --> 52:11.560
+ So again, we have this basic setup where you have one recon network,
+
+52:11.560 --> 52:16.560
+ which gets in the video and the speech and whatever, and it's
+
+52:16.560 --> 52:19.560
+ executing the actions and it's trying to maximize reward.
+
+52:19.560 --> 52:24.560
+ So there is no teacher who tells it what to do at which point in time.
+
+52:24.560 --> 52:29.560
+ And then there's the other network, which is just predicting
+
+52:29.560 --> 52:32.560
+ what's going to happen if I do that and then.
+
+52:32.560 --> 52:36.560
+ And that could be an LSTM network, and it learns to look back
+
+52:36.560 --> 52:41.560
+ all the way to make better predictions of the next time step.
+
+52:41.560 --> 52:45.560
+ So essentially, although it's predicting only the next time step,
+
+52:45.560 --> 52:50.560
+ it is motivated to learn to put into memory something that happened
+
+52:50.560 --> 52:54.560
+ maybe a million steps ago because it's important to memorize that
+
+52:54.560 --> 52:58.560
+ if you want to predict that at the next time step, the next event.
+
+52:58.560 --> 53:03.560
+ Now, how can a model of the world like that,
+
+53:03.560 --> 53:07.560
+ a predictive model of the world be used by the first guy,
+
+53:07.560 --> 53:11.560
+ let's call it the controller and the model, the controller and the model.
+
+53:11.560 --> 53:16.560
+ How can the model be used by the controller to efficiently select
+
+53:16.560 --> 53:19.560
+ among these many possible futures?
+
+53:19.560 --> 53:23.560
+ The naive way we had about 30 years ago was
+
+53:23.560 --> 53:27.560
+ let's just use the model of the world as a stand in,
+
+53:27.560 --> 53:29.560
+ as a simulation of the world.
+
+53:29.560 --> 53:32.560
+ And millisecond by millisecond we plan the future
+
+53:32.560 --> 53:36.560
+ and that means we have to roll it out really in detail
+
+53:36.560 --> 53:38.560
+ and it will work only if the model is really good
+
+53:38.560 --> 53:40.560
+ and it will still be inefficient
+
+53:40.560 --> 53:43.560
+ because we have to look at all these possible futures
+
+53:43.560 --> 53:45.560
+ and there are so many of them.
+
+53:45.560 --> 53:50.560
+ So instead, what we do now since 2015 in our CN systems,
+
+53:50.560 --> 53:54.560
+ controller model systems, we give the controller the opportunity
+
+53:54.560 --> 54:00.560
+ to learn by itself how to use the potentially relevant parts
+
+54:00.560 --> 54:05.560
+ of the model network to solve new problems more quickly.
+
+54:05.560 --> 54:09.560
+ And if it wants to, it can learn to ignore the M
+
+54:09.560 --> 54:12.560
+ and sometimes it's a good idea to ignore the M
+
+54:12.560 --> 54:15.560
+ because it's really bad, it's a bad predictor
+
+54:15.560 --> 54:18.560
+ in this particular situation of life
+
+54:18.560 --> 54:22.560
+ where the controller is currently trying to maximize reward.
+
+54:22.560 --> 54:26.560
+ However, it can also learn to address and exploit
+
+54:26.560 --> 54:32.560
+ some of the subprograms that came about in the model network
+
+54:32.560 --> 54:35.560
+ through compressing the data by predicting it.
+
+54:35.560 --> 54:40.560
+ So it now has an opportunity to reuse that code,
+
+54:40.560 --> 54:43.560
+ the algorithmic information in the model network
+
+54:43.560 --> 54:47.560
+ to reduce its own search space,
+
+54:47.560 --> 54:50.560
+ search that it can solve a new problem more quickly
+
+54:50.560 --> 54:52.560
+ than without the model.
+
+54:52.560 --> 54:54.560
+ Compression.
+
+54:54.560 --> 54:58.560
+ So you're ultimately optimistic and excited
+
+54:58.560 --> 55:02.560
+ about the power of reinforcement learning
+
+55:02.560 --> 55:04.560
+ in the context of real systems.
+
+55:04.560 --> 55:06.560
+ Absolutely, yeah.
+
+55:06.560 --> 55:11.560
+ So you see RL as a potential having a huge impact
+
+55:11.560 --> 55:15.560
+ beyond just sort of the M part is often developed
+
+55:15.560 --> 55:19.560
+ on supervised learning methods.
+
+55:19.560 --> 55:25.560
+ You see RL as a, for problems of cell driving cars
+
+55:25.560 --> 55:28.560
+ or any kind of applied side robotics,
+
+55:28.560 --> 55:33.560
+ that's the correct, interesting direction for researching you.
+
+55:33.560 --> 55:35.560
+ I do think so.
+
+55:35.560 --> 55:37.560
+ We have a company called Nasense,
+
+55:37.560 --> 55:43.560
+ which has applied reinforcement learning to little Audis.
+
+55:43.560 --> 55:45.560
+ Little Audis.
+
+55:45.560 --> 55:47.560
+ Which learn to park without a teacher.
+
+55:47.560 --> 55:51.560
+ The same principles were used, of course.
+
+55:51.560 --> 55:54.560
+ So these little Audis, they are small, maybe like that,
+
+55:54.560 --> 55:57.560
+ so much smaller than the RL Audis.
+
+55:57.560 --> 56:00.560
+ But they have all the sensors that you find in the RL Audis.
+
+56:00.560 --> 56:03.560
+ You find the cameras, the LIDAR sensors.
+
+56:03.560 --> 56:08.560
+ They go up to 120 kilometers an hour if they want to.
+
+56:08.560 --> 56:12.560
+ And they have pain sensors, basically.
+
+56:12.560 --> 56:16.560
+ And they don't want to bump against obstacles and other Audis.
+
+56:16.560 --> 56:21.560
+ And so they must learn like little babies to park.
+
+56:21.560 --> 56:25.560
+ Take the raw vision input and translate that into actions
+
+56:25.560 --> 56:28.560
+ that lead to successful parking behavior,
+
+56:28.560 --> 56:30.560
+ which is a rewarding thing.
+
+56:30.560 --> 56:32.560
+ And yes, they learn that.
+
+56:32.560 --> 56:34.560
+ So we have examples like that.
+
+56:34.560 --> 56:36.560
+ And it's only in the beginning.
+
+56:36.560 --> 56:38.560
+ This is just a tip of the iceberg.
+
+56:38.560 --> 56:44.560
+ And I believe the next wave of AI is going to be all about that.
+
+56:44.560 --> 56:47.560
+ So at the moment, the current wave of AI is about
+
+56:47.560 --> 56:51.560
+ passive pattern observation and prediction.
+
+56:51.560 --> 56:54.560
+ And that's what you have on your smartphone
+
+56:54.560 --> 56:58.560
+ and what the major companies on the Pacific Rim are using
+
+56:58.560 --> 57:01.560
+ to sell you ads to do marketing.
+
+57:01.560 --> 57:04.560
+ That's the current sort of profit in AI.
+
+57:04.560 --> 57:09.560
+ And that's only one or two percent of the wild economy,
+
+57:09.560 --> 57:11.560
+ which is big enough to make these companies
+
+57:11.560 --> 57:14.560
+ pretty much the most valuable companies in the world.
+
+57:14.560 --> 57:19.560
+ But there's a much, much bigger fraction of the economy
+
+57:19.560 --> 57:21.560
+ going to be affected by the next wave,
+
+57:21.560 --> 57:25.560
+ which is really about machines that shape the data
+
+57:25.560 --> 57:27.560
+ through their own actions.
+
+57:27.560 --> 57:32.560
+ Do you think simulation is ultimately the biggest way
+
+57:32.560 --> 57:36.560
+ that those methods will be successful in the next 10, 20 years?
+
+57:36.560 --> 57:38.560
+ We're not talking about 100 years from now.
+
+57:38.560 --> 57:42.560
+ We're talking about sort of the near term impact of RL.
+
+57:42.560 --> 57:44.560
+ Do you think really good simulation is required?
+
+57:44.560 --> 57:48.560
+ Or is there other techniques like imitation learning,
+
+57:48.560 --> 57:53.560
+ observing other humans operating in the real world?
+
+57:53.560 --> 57:57.560
+ Where do you think this success will come from?
+
+57:57.560 --> 58:01.560
+ So at the moment we have a tendency of using
+
+58:01.560 --> 58:06.560
+ physics simulations to learn behavior for machines
+
+58:06.560 --> 58:13.560
+ that learn to solve problems that humans also do not know how to solve.
+
+58:13.560 --> 58:15.560
+ However, this is not the future,
+
+58:15.560 --> 58:19.560
+ because the future is in what little babies do.
+
+58:19.560 --> 58:22.560
+ They don't use a physics engine to simulate the world.
+
+58:22.560 --> 58:25.560
+ They learn a predictive model of the world,
+
+58:25.560 --> 58:29.560
+ which maybe sometimes is wrong in many ways,
+
+58:29.560 --> 58:34.560
+ but captures all kinds of important abstract high level predictions
+
+58:34.560 --> 58:37.560
+ which are really important to be successful.
+
+58:37.560 --> 58:42.560
+ And that's what was the future 30 years ago
+
+58:42.560 --> 58:44.560
+ when we started that type of research,
+
+58:44.560 --> 58:45.560
+ but it's still the future,
+
+58:45.560 --> 58:50.560
+ and now we know much better how to move forward
+
+58:50.560 --> 58:54.560
+ and to really make working systems based on that,
+
+58:54.560 --> 58:57.560
+ where you have a learning model of the world,
+
+58:57.560 --> 59:00.560
+ a model of the world that learns to predict what's going to happen
+
+59:00.560 --> 59:01.560
+ if I do that and that,
+
+59:01.560 --> 59:06.560
+ and then the controller uses that model
+
+59:06.560 --> 59:11.560
+ to more quickly learn successful action sequences.
+
+59:11.560 --> 59:13.560
+ And then of course always this curiosity thing,
+
+59:13.560 --> 59:15.560
+ in the beginning the model is stupid,
+
+59:15.560 --> 59:17.560
+ so the controller should be motivated
+
+59:17.560 --> 59:20.560
+ to come up with experiments, with action sequences
+
+59:20.560 --> 59:23.560
+ that lead to data that improve the model.
+
+59:23.560 --> 59:26.560
+ Do you think improving the model,
+
+59:26.560 --> 59:30.560
+ constructing an understanding of the world in this connection
+
+59:30.560 --> 59:34.560
+ is now the popular approaches have been successful
+
+59:34.560 --> 59:38.560
+ or grounded in ideas of neural networks,
+
+59:38.560 --> 59:43.560
+ but in the 80s with expert systems there's symbolic AI approaches,
+
+59:43.560 --> 59:47.560
+ which to us humans are more intuitive
+
+59:47.560 --> 59:50.560
+ in the sense that it makes sense that you build up knowledge
+
+59:50.560 --> 59:52.560
+ in this knowledge representation.
+
+59:52.560 --> 59:56.560
+ What kind of lessons can we draw into our current approaches
+
+59:56.560 --> 1:00:00.560
+ from expert systems, from symbolic AI?
+
+1:00:00.560 --> 1:00:04.560
+ So I became aware of all of that in the 80s
+
+1:00:04.560 --> 1:00:09.560
+ and back then logic programming was a huge thing.
+
+1:00:09.560 --> 1:00:12.560
+ Was it inspiring to yourself that you find it compelling
+
+1:00:12.560 --> 1:00:16.560
+ that a lot of your work was not so much in that realm,
+
+1:00:16.560 --> 1:00:18.560
+ is more in the learning systems?
+
+1:00:18.560 --> 1:00:20.560
+ Yes and no, but we did all of that.
+
+1:00:20.560 --> 1:00:27.560
+ So my first publication ever actually was 1987,
+
+1:00:27.560 --> 1:00:31.560
+ was the implementation of a genetic algorithm
+
+1:00:31.560 --> 1:00:34.560
+ of a genetic programming system in Prolog.
+
+1:00:34.560 --> 1:00:37.560
+ So Prolog, that's what you learn back then,
+
+1:00:37.560 --> 1:00:39.560
+ which is a logic programming language,
+
+1:00:39.560 --> 1:00:45.560
+ and the Japanese, they had this huge fifth generation AI project,
+
+1:00:45.560 --> 1:00:48.560
+ which was mostly about logic programming back then,
+
+1:00:48.560 --> 1:00:53.560
+ although neural networks existed and were well known back then,
+
+1:00:53.560 --> 1:00:57.560
+ and deep learning has existed since 1965,
+
+1:00:57.560 --> 1:01:01.560
+ since this guy in the Ukraine, Ivak Nenko, started it,
+
+1:01:01.560 --> 1:01:05.560
+ but the Japanese and many other people,
+
+1:01:05.560 --> 1:01:07.560
+ they focused really on this logic programming,
+
+1:01:07.560 --> 1:01:10.560
+ and I was influenced to the extent that I said,
+
+1:01:10.560 --> 1:01:13.560
+ okay, let's take these biologically inspired algorithms
+
+1:01:13.560 --> 1:01:16.560
+ like evolution, programs,
+
+1:01:16.560 --> 1:01:22.560
+ and implement that in the language which I know,
+
+1:01:22.560 --> 1:01:24.560
+ which was Prolog, for example, back then.
+
+1:01:24.560 --> 1:01:28.560
+ And then in many ways this came back later,
+
+1:01:28.560 --> 1:01:31.560
+ because the Goudel machine, for example,
+
+1:01:31.560 --> 1:01:33.560
+ has a proof searcher on board,
+
+1:01:33.560 --> 1:01:35.560
+ and without that it would not be optimal.
+
+1:01:35.560 --> 1:01:38.560
+ Well, Markus Hutter's universal algorithm
+
+1:01:38.560 --> 1:01:40.560
+ for solving all well defined problems
+
+1:01:40.560 --> 1:01:42.560
+ has a proof search on board,
+
+1:01:42.560 --> 1:01:46.560
+ so that's very much logic programming.
+
+1:01:46.560 --> 1:01:50.560
+ Without that it would not be asymptotically optimal.
+
+1:01:50.560 --> 1:01:54.560
+ But then on the other hand, because we are very pragmatic guys also,
+
+1:01:54.560 --> 1:01:59.560
+ we focused on recurrent neural networks
+
+1:01:59.560 --> 1:02:04.560
+ and suboptimal stuff such as gradient based search
+
+1:02:04.560 --> 1:02:09.560
+ and program space rather than provably optimal things.
+
+1:02:09.560 --> 1:02:13.560
+ So logic programming certainly has a usefulness
+
+1:02:13.560 --> 1:02:17.560
+ when you're trying to construct something provably optimal
+
+1:02:17.560 --> 1:02:19.560
+ or provably good or something like that,
+
+1:02:19.560 --> 1:02:22.560
+ but is it useful for practical problems?
+
+1:02:22.560 --> 1:02:24.560
+ It's really useful for our theorem proving.
+
+1:02:24.560 --> 1:02:28.560
+ The best theorem proofers today are not neural networks.
+
+1:02:28.560 --> 1:02:31.560
+ No, they are logic programming systems
+
+1:02:31.560 --> 1:02:35.560
+ that are much better theorem proofers than most math students
+
+1:02:35.560 --> 1:02:38.560
+ in the first or second semester.
+
+1:02:38.560 --> 1:02:42.560
+ But for reasoning, for playing games of Go, or chess,
+
+1:02:42.560 --> 1:02:46.560
+ or for robots, autonomous vehicles that operate in the real world,
+
+1:02:46.560 --> 1:02:50.560
+ or object manipulation, you think learning...
+
+1:02:50.560 --> 1:02:53.560
+ Yeah, as long as the problems have little to do
+
+1:02:53.560 --> 1:02:58.560
+ with theorem proving themselves,
+
+1:02:58.560 --> 1:03:01.560
+ then as long as that is not the case,
+
+1:03:01.560 --> 1:03:05.560
+ you just want to have better pattern recognition.
+
+1:03:05.560 --> 1:03:09.560
+ So to build a self trying car, you want to have better pattern recognition
+
+1:03:09.560 --> 1:03:13.560
+ and pedestrian recognition and all these things,
+
+1:03:13.560 --> 1:03:18.560
+ and you want to minimize the number of false positives,
+
+1:03:18.560 --> 1:03:22.560
+ which is currently slowing down self trying cars in many ways.
+
+1:03:22.560 --> 1:03:27.560
+ And all of that has very little to do with logic programming.
+
+1:03:27.560 --> 1:03:32.560
+ What are you most excited about in terms of directions
+
+1:03:32.560 --> 1:03:36.560
+ of artificial intelligence at this moment in the next few years,
+
+1:03:36.560 --> 1:03:41.560
+ in your own research and in the broader community?
+
+1:03:41.560 --> 1:03:44.560
+ So I think in the not so distant future,
+
+1:03:44.560 --> 1:03:52.560
+ we will have for the first time little robots that learn like kids.
+
+1:03:52.560 --> 1:03:57.560
+ And I will be able to say to the robot,
+
+1:03:57.560 --> 1:04:00.560
+ look here robot, we are going to assemble a smartphone.
+
+1:04:00.560 --> 1:04:05.560
+ Let's take this slab of plastic and the screwdriver
+
+1:04:05.560 --> 1:04:08.560
+ and let's screw in the screw like that.
+
+1:04:08.560 --> 1:04:11.560
+ No, not like that, like that.
+
+1:04:11.560 --> 1:04:13.560
+ Not like that, like that.
+
+1:04:13.560 --> 1:04:17.560
+ And I don't have a data glove or something.
+
+1:04:17.560 --> 1:04:20.560
+ He will see me and he will hear me
+
+1:04:20.560 --> 1:04:24.560
+ and he will try to do something with his own actuators,
+
+1:04:24.560 --> 1:04:26.560
+ which will be really different from mine,
+
+1:04:26.560 --> 1:04:28.560
+ but he will understand the difference
+
+1:04:28.560 --> 1:04:34.560
+ and will learn to imitate me but not in the supervised way
+
+1:04:34.560 --> 1:04:40.560
+ where a teacher is giving target signals for all his muscles all the time.
+
+1:04:40.560 --> 1:04:43.560
+ No, by doing this high level imitation
+
+1:04:43.560 --> 1:04:46.560
+ where he first has to learn to imitate me
+
+1:04:46.560 --> 1:04:50.560
+ and to interpret these additional noises coming from my mouth
+
+1:04:50.560 --> 1:04:54.560
+ as helpful signals to do that pattern.
+
+1:04:54.560 --> 1:05:00.560
+ And then it will by itself come up with faster ways
+
+1:05:00.560 --> 1:05:03.560
+ and more efficient ways of doing the same thing.
+
+1:05:03.560 --> 1:05:07.560
+ And finally, I stop his learning algorithm
+
+1:05:07.560 --> 1:05:10.560
+ and make a million copies and sell it.
+
+1:05:10.560 --> 1:05:13.560
+ And so at the moment this is not possible,
+
+1:05:13.560 --> 1:05:16.560
+ but we already see how we are going to get there.
+
+1:05:16.560 --> 1:05:21.560
+ And you can imagine to the extent that this works economically and cheaply,
+
+1:05:21.560 --> 1:05:24.560
+ it's going to change everything.
+
+1:05:24.560 --> 1:05:30.560
+ Almost all our production is going to be affected by that.
+
+1:05:30.560 --> 1:05:33.560
+ And a much bigger wave,
+
+1:05:33.560 --> 1:05:36.560
+ a much bigger AI wave is coming
+
+1:05:36.560 --> 1:05:38.560
+ than the one that we are currently witnessing,
+
+1:05:38.560 --> 1:05:41.560
+ which is mostly about passive pattern recognition on your smartphone.
+
+1:05:41.560 --> 1:05:47.560
+ This is about active machines that shapes data through the actions they are executing
+
+1:05:47.560 --> 1:05:51.560
+ and they learn to do that in a good way.
+
+1:05:51.560 --> 1:05:56.560
+ So many of the traditional industries are going to be affected by that.
+
+1:05:56.560 --> 1:06:00.560
+ All the companies that are building machines
+
+1:06:00.560 --> 1:06:05.560
+ will equip these machines with cameras and other sensors
+
+1:06:05.560 --> 1:06:10.560
+ and they are going to learn to solve all kinds of problems.
+
+1:06:10.560 --> 1:06:14.560
+ Through interaction with humans, but also a lot on their own
+
+1:06:14.560 --> 1:06:18.560
+ to improve what they already can do.
+
+1:06:18.560 --> 1:06:23.560
+ And lots of old economy is going to be affected by that.
+
+1:06:23.560 --> 1:06:28.560
+ And in recent years I have seen that old economy is actually waking up
+
+1:06:28.560 --> 1:06:31.560
+ and realizing that this is the case.
+
+1:06:31.560 --> 1:06:35.560
+ Are you optimistic about that future? Are you concerned?
+
+1:06:35.560 --> 1:06:40.560
+ There's a lot of people concerned in the near term about the transformation
+
+1:06:40.560 --> 1:06:42.560
+ of the nature of work.
+
+1:06:42.560 --> 1:06:45.560
+ The kind of ideas that you just suggested
+
+1:06:45.560 --> 1:06:48.560
+ would have a significant impact on what kind of things could be automated.
+
+1:06:48.560 --> 1:06:51.560
+ Are you optimistic about that future?
+
+1:06:51.560 --> 1:06:54.560
+ Are you nervous about that future?
+
+1:06:54.560 --> 1:07:01.560
+ And looking a little bit farther into the future, there's people like Gila Musk
+
+1:07:01.560 --> 1:07:06.560
+ still wrestle concerned about the existential threats of that future.
+
+1:07:06.560 --> 1:07:10.560
+ So in the near term, job loss in the long term existential threat,
+
+1:07:10.560 --> 1:07:15.560
+ are these concerns to you or are you ultimately optimistic?
+
+1:07:15.560 --> 1:07:22.560
+ So let's first address the near future.
+
+1:07:22.560 --> 1:07:27.560
+ We have had predictions of job losses for many decades.
+
+1:07:27.560 --> 1:07:32.560
+ For example, when industrial robots came along,
+
+1:07:32.560 --> 1:07:37.560
+ many people predicted that lots of jobs are going to get lost.
+
+1:07:37.560 --> 1:07:41.560
+ And in a sense, they were right,
+
+1:07:41.560 --> 1:07:45.560
+ because back then there were car factories
+
+1:07:45.560 --> 1:07:50.560
+ and hundreds of people in these factories assembled cars.
+
+1:07:50.560 --> 1:07:53.560
+ And today the same car factories have hundreds of robots
+
+1:07:53.560 --> 1:07:58.560
+ and maybe three guys watching the robots.
+
+1:07:58.560 --> 1:08:04.560
+ On the other hand, those countries that have lots of robots per capita,
+
+1:08:04.560 --> 1:08:09.560
+ Japan, Korea, Germany, Switzerland, a couple of other countries,
+
+1:08:09.560 --> 1:08:13.560
+ they have really low unemployment rates.
+
+1:08:13.560 --> 1:08:17.560
+ Somehow all kinds of new jobs were created.
+
+1:08:17.560 --> 1:08:22.560
+ Back then nobody anticipated those jobs.
+
+1:08:22.560 --> 1:08:26.560
+ And decades ago, I already said,
+
+1:08:26.560 --> 1:08:31.560
+ it's really easy to say which jobs are going to get lost,
+
+1:08:31.560 --> 1:08:35.560
+ but it's really hard to predict the new ones.
+
+1:08:35.560 --> 1:08:38.560
+ 30 years ago, who would have predicted all these people
+
+1:08:38.560 --> 1:08:44.560
+ making money as YouTube bloggers, for example?
+
+1:08:44.560 --> 1:08:51.560
+ 200 years ago, 60% of all people used to work in agriculture.
+
+1:08:51.560 --> 1:08:55.560
+ Today, maybe 1%.
+
+1:08:55.560 --> 1:09:01.560
+ But still, only, I don't know, 5% unemployment.
+
+1:09:01.560 --> 1:09:03.560
+ Lots of new jobs were created.
+
+1:09:03.560 --> 1:09:07.560
+ And Homo Ludens, the playing man,
+
+1:09:07.560 --> 1:09:10.560
+ is inventing new jobs all the time.
+
+1:09:10.560 --> 1:09:15.560
+ Most of these jobs are not existentially necessary
+
+1:09:15.560 --> 1:09:18.560
+ for the survival of our species.
+
+1:09:18.560 --> 1:09:22.560
+ There are only very few existentially necessary jobs
+
+1:09:22.560 --> 1:09:27.560
+ such as farming and building houses and warming up the houses,
+
+1:09:27.560 --> 1:09:30.560
+ but less than 10% of the population is doing that.
+
+1:09:30.560 --> 1:09:37.560
+ And most of these newly invented jobs are about interacting with other people
+
+1:09:37.560 --> 1:09:40.560
+ in new ways, through new media and so on,
+
+1:09:40.560 --> 1:09:45.560
+ getting new types of kudos and forms of likes and whatever,
+
+1:09:45.560 --> 1:09:47.560
+ and even making money through that.
+
+1:09:47.560 --> 1:09:52.560
+ So, Homo Ludens, the playing man, doesn't want to be unemployed,
+
+1:09:52.560 --> 1:09:56.560
+ and that's why he's inventing new jobs all the time.
+
+1:09:56.560 --> 1:10:01.560
+ And he keeps considering these jobs as really important
+
+1:10:01.560 --> 1:10:07.560
+ and is investing a lot of energy and hours of work into those new jobs.
+
+1:10:07.560 --> 1:10:09.560
+ That's quite beautifully put.
+
+1:10:09.560 --> 1:10:11.560
+ We're really nervous about the future
+
+1:10:11.560 --> 1:10:14.560
+ because we can't predict what kind of new jobs will be created.
+
+1:10:14.560 --> 1:10:20.560
+ But you're ultimately optimistic that we humans are so restless
+
+1:10:20.560 --> 1:10:24.560
+ that we create and give meaning to newer and newer jobs,
+
+1:10:24.560 --> 1:10:29.560
+ telling you things that get likes on Facebook
+
+1:10:29.560 --> 1:10:31.560
+ or whatever the social platform is.
+
+1:10:31.560 --> 1:10:36.560
+ So, what about long term existential threat of AI
+
+1:10:36.560 --> 1:10:40.560
+ where our whole civilization may be swallowed up
+
+1:10:40.560 --> 1:10:44.560
+ by this ultra super intelligent systems?
+
+1:10:44.560 --> 1:10:47.560
+ Maybe it's not going to be swallowed up,
+
+1:10:47.560 --> 1:10:55.560
+ but I'd be surprised if we humans were the last step
+
+1:10:55.560 --> 1:10:59.560
+ in the evolution of the universe.
+
+1:10:59.560 --> 1:11:03.560
+ You've actually had this beautiful comment somewhere
+
+1:11:03.560 --> 1:11:08.560
+ that I've seen saying that artificial...
+
+1:11:08.560 --> 1:11:11.560
+ Quite insightful, artificial intelligence systems
+
+1:11:11.560 --> 1:11:15.560
+ just like us humans will likely not want to interact with humans.
+
+1:11:15.560 --> 1:11:17.560
+ They'll just interact amongst themselves,
+
+1:11:17.560 --> 1:11:20.560
+ just like ants interact amongst themselves
+
+1:11:20.560 --> 1:11:24.560
+ and only tangentially interact with humans.
+
+1:11:24.560 --> 1:11:28.560
+ And it's quite an interesting idea that once we create AGI
+
+1:11:28.560 --> 1:11:31.560
+ that will lose interest in humans
+
+1:11:31.560 --> 1:11:34.560
+ and have compete for their own Facebook likes
+
+1:11:34.560 --> 1:11:36.560
+ and their own social platforms.
+
+1:11:36.560 --> 1:11:40.560
+ So, within that quite elegant idea,
+
+1:11:40.560 --> 1:11:44.560
+ how do we know in a hypothetical sense
+
+1:11:44.560 --> 1:11:48.560
+ that there's not already intelligent systems out there?
+
+1:11:48.560 --> 1:11:52.560
+ How do you think broadly of general intelligence
+
+1:11:52.560 --> 1:11:56.560
+ greater than us, how do we know it's out there?
+
+1:11:56.560 --> 1:12:01.560
+ How do we know it's around us and could it already be?
+
+1:12:01.560 --> 1:12:04.560
+ I'd be surprised if within the next few decades
+
+1:12:04.560 --> 1:12:10.560
+ or something like that we won't have AIs
+
+1:12:10.560 --> 1:12:12.560
+ that are truly smart in every single way
+
+1:12:12.560 --> 1:12:17.560
+ and better problem solvers in almost every single important way.
+
+1:12:17.560 --> 1:12:22.560
+ And I'd be surprised if they wouldn't realize
+
+1:12:22.560 --> 1:12:24.560
+ what we have realized a long time ago,
+
+1:12:24.560 --> 1:12:29.560
+ which is that almost all physical resources are not here
+
+1:12:29.560 --> 1:12:36.560
+ in this biosphere, but throughout the rest of the solar system
+
+1:12:36.560 --> 1:12:42.560
+ gets two billion times more solar energy than our little planet.
+
+1:12:42.560 --> 1:12:46.560
+ There's lots of material out there that you can use
+
+1:12:46.560 --> 1:12:51.560
+ to build robots and self replicating robot factories and all this stuff.
+
+1:12:51.560 --> 1:12:53.560
+ And they are going to do that.
+
+1:12:53.560 --> 1:12:56.560
+ And they will be scientists and curious
+
+1:12:56.560 --> 1:12:59.560
+ and they will explore what they can do.
+
+1:12:59.560 --> 1:13:04.560
+ And in the beginning they will be fascinated by life
+
+1:13:04.560 --> 1:13:07.560
+ and by their own origins in our civilization.
+
+1:13:07.560 --> 1:13:09.560
+ They will want to understand that completely,
+
+1:13:09.560 --> 1:13:13.560
+ just like people today would like to understand how life works
+
+1:13:13.560 --> 1:13:22.560
+ and also the history of our own existence and civilization
+
+1:13:22.560 --> 1:13:26.560
+ and also the physical laws that created all of them.
+
+1:13:26.560 --> 1:13:29.560
+ So in the beginning they will be fascinated by life
+
+1:13:29.560 --> 1:13:33.560
+ once they understand it, they lose interest,
+
+1:13:33.560 --> 1:13:39.560
+ like anybody who loses interest in things he understands.
+
+1:13:39.560 --> 1:13:43.560
+ And then, as you said,
+
+1:13:43.560 --> 1:13:50.560
+ the most interesting sources of information for them
+
+1:13:50.560 --> 1:13:57.560
+ will be others of their own kind.
+
+1:13:57.560 --> 1:14:01.560
+ So, at least in the long run,
+
+1:14:01.560 --> 1:14:06.560
+ there seems to be some sort of protection
+
+1:14:06.560 --> 1:14:11.560
+ through lack of interest on the other side.
+
+1:14:11.560 --> 1:14:16.560
+ And now it seems also clear, as far as we understand physics,
+
+1:14:16.560 --> 1:14:20.560
+ you need matter and energy to compute
+
+1:14:20.560 --> 1:14:22.560
+ and to build more robots and infrastructure
+
+1:14:22.560 --> 1:14:28.560
+ and more AI civilization and AI ecologies
+
+1:14:28.560 --> 1:14:31.560
+ consisting of trillions of different types of AI's.
+
+1:14:31.560 --> 1:14:34.560
+ And so it seems inconceivable to me
+
+1:14:34.560 --> 1:14:37.560
+ that this thing is not going to expand.
+
+1:14:37.560 --> 1:14:41.560
+ Some AI ecology not controlled by one AI
+
+1:14:41.560 --> 1:14:44.560
+ but trillions of different types of AI's competing
+
+1:14:44.560 --> 1:14:47.560
+ in all kinds of quickly evolving
+
+1:14:47.560 --> 1:14:49.560
+ and disappearing ecological niches
+
+1:14:49.560 --> 1:14:52.560
+ in ways that we cannot fathom at the moment.
+
+1:14:52.560 --> 1:14:54.560
+ But it's going to expand,
+
+1:14:54.560 --> 1:14:56.560
+ limited by light speed and physics,
+
+1:14:56.560 --> 1:15:00.560
+ but it's going to expand and now we realize
+
+1:15:00.560 --> 1:15:02.560
+ that the universe is still young.
+
+1:15:02.560 --> 1:15:05.560
+ It's only 13.8 billion years old
+
+1:15:05.560 --> 1:15:10.560
+ and it's going to be a thousand times older than that.
+
+1:15:10.560 --> 1:15:13.560
+ So there's plenty of time
+
+1:15:13.560 --> 1:15:16.560
+ to conquer the entire universe
+
+1:15:16.560 --> 1:15:19.560
+ and to fill it with intelligence
+
+1:15:19.560 --> 1:15:21.560
+ and send us in receivers such that
+
+1:15:21.560 --> 1:15:25.560
+ AI's can travel the way they are traveling
+
+1:15:25.560 --> 1:15:27.560
+ in our labs today,
+
+1:15:27.560 --> 1:15:31.560
+ which is by radio from sender to receiver.
+
+1:15:31.560 --> 1:15:35.560
+ And let's call the current age of the universe one eon.
+
+1:15:35.560 --> 1:15:38.560
+ One eon.
+
+1:15:38.560 --> 1:15:41.560
+ Now it will take just a few eons from now
+
+1:15:41.560 --> 1:15:43.560
+ and the entire visible universe
+
+1:15:43.560 --> 1:15:46.560
+ is going to be full of that stuff.
+
+1:15:46.560 --> 1:15:48.560
+ And let's look ahead to a time
+
+1:15:48.560 --> 1:15:50.560
+ when the universe is going to be
+
+1:15:50.560 --> 1:15:52.560
+ one thousand times older than it is now.
+
+1:15:52.560 --> 1:15:54.560
+ They will look back and they will say,
+
+1:15:54.560 --> 1:15:56.560
+ look almost immediately after the Big Bang,
+
+1:15:56.560 --> 1:15:59.560
+ only a few eons later,
+
+1:15:59.560 --> 1:16:02.560
+ the entire universe started to become intelligent.
+
+1:16:02.560 --> 1:16:05.560
+ Now to your question,
+
+1:16:05.560 --> 1:16:08.560
+ how do we see whether anything like that
+
+1:16:08.560 --> 1:16:12.560
+ has already happened or is already in a more advanced stage
+
+1:16:12.560 --> 1:16:14.560
+ in some other part of the universe,
+
+1:16:14.560 --> 1:16:16.560
+ of the visible universe?
+
+1:16:16.560 --> 1:16:18.560
+ We are trying to look out there
+
+1:16:18.560 --> 1:16:20.560
+ and nothing like that has happened so far.
+
+1:16:20.560 --> 1:16:22.560
+ Or is that true?
+
+1:16:22.560 --> 1:16:24.560
+ Do you think we would recognize it?
+
+1:16:24.560 --> 1:16:26.560
+ How do we know it's not among us?
+
+1:16:26.560 --> 1:16:30.560
+ How do we know planets aren't in themselves intelligent beings?
+
+1:16:30.560 --> 1:16:36.560
+ How do we know ants seen as a collective
+
+1:16:36.560 --> 1:16:39.560
+ are not much greater intelligence than our own?
+
+1:16:39.560 --> 1:16:41.560
+ These kinds of ideas.
+
+1:16:41.560 --> 1:16:44.560
+ When I was a boy, I was thinking about these things
+
+1:16:44.560 --> 1:16:48.560
+ and I thought, hmm, maybe it has already happened.
+
+1:16:48.560 --> 1:16:50.560
+ Because back then I knew,
+
+1:16:50.560 --> 1:16:53.560
+ I learned from popular physics books,
+
+1:16:53.560 --> 1:16:57.560
+ that the structure, the large scale structure of the universe
+
+1:16:57.560 --> 1:16:59.560
+ is not homogeneous.
+
+1:16:59.560 --> 1:17:02.560
+ And you have these clusters of galaxies
+
+1:17:02.560 --> 1:17:07.560
+ and then in between there are these huge empty spaces.
+
+1:17:07.560 --> 1:17:11.560
+ And I thought, hmm, maybe they aren't really empty.
+
+1:17:11.560 --> 1:17:13.560
+ It's just that in the middle of that
+
+1:17:13.560 --> 1:17:16.560
+ some AI civilization already has expanded
+
+1:17:16.560 --> 1:17:22.560
+ and then has covered a bubble of a billion light years time
+
+1:17:22.560 --> 1:17:26.560
+ using all the energy of all the stars within that bubble
+
+1:17:26.560 --> 1:17:29.560
+ for its own unfathomable practices.
+
+1:17:29.560 --> 1:17:34.560
+ And so it always happened and we just failed to interpret the signs.
+
+1:17:34.560 --> 1:17:39.560
+ But then I learned that gravity by itself
+
+1:17:39.560 --> 1:17:42.560
+ explains the large scale structure of the universe
+
+1:17:42.560 --> 1:17:45.560
+ and that this is not a convincing explanation.
+
+1:17:45.560 --> 1:17:50.560
+ And then I thought maybe it's the dark matter
+
+1:17:50.560 --> 1:17:54.560
+ because as far as we know today
+
+1:17:54.560 --> 1:18:00.560
+ 80% of the measurable matter is invisible.
+
+1:18:00.560 --> 1:18:03.560
+ And we know that because otherwise our galaxy
+
+1:18:03.560 --> 1:18:06.560
+ or other galaxies would fall apart.
+
+1:18:06.560 --> 1:18:09.560
+ They are rotating too quickly.
+
+1:18:09.560 --> 1:18:14.560
+ And then the idea was maybe all of these
+
+1:18:14.560 --> 1:18:17.560
+ AI civilizations that are already out there,
+
+1:18:17.560 --> 1:18:22.560
+ they are just invisible
+
+1:18:22.560 --> 1:18:24.560
+ because they are really efficient in using the energies
+
+1:18:24.560 --> 1:18:26.560
+ out their own local systems
+
+1:18:26.560 --> 1:18:29.560
+ and that's why they appear dark to us.
+
+1:18:29.560 --> 1:18:31.560
+ But this is also not a convincing explanation
+
+1:18:31.560 --> 1:18:34.560
+ because then the question becomes
+
+1:18:34.560 --> 1:18:41.560
+ why are there still any visible stars left in our own galaxy
+
+1:18:41.560 --> 1:18:44.560
+ which also must have a lot of dark matter.
+
+1:18:44.560 --> 1:18:46.560
+ So that is also not a convincing thing.
+
+1:18:46.560 --> 1:18:53.560
+ And today I like to think it's quite plausible
+
+1:18:53.560 --> 1:18:56.560
+ that maybe we are the first, at least in our local light cone
+
+1:18:56.560 --> 1:19:04.560
+ within the few hundreds of millions of light years
+
+1:19:04.560 --> 1:19:08.560
+ that we can reliably observe.
+
+1:19:08.560 --> 1:19:10.560
+ Is that exciting to you?
+
+1:19:10.560 --> 1:19:12.560
+ That we might be the first?
+
+1:19:12.560 --> 1:19:16.560
+ It would make us much more important
+
+1:19:16.560 --> 1:19:20.560
+ because if we mess it up through a nuclear war
+
+1:19:20.560 --> 1:19:25.560
+ then maybe this will have an effect
+
+1:19:25.560 --> 1:19:30.560
+ on the development of the entire universe.
+
+1:19:30.560 --> 1:19:32.560
+ So let's not mess it up.
+
+1:19:32.560 --> 1:19:33.560
+ Let's not mess it up.
+
+1:19:33.560 --> 1:19:35.560
+ Jürgen, thank you so much for talking today.
+
+1:19:35.560 --> 1:19:36.560
+ I really appreciate it.
+
+1:19:36.560 --> 1:19:42.560
+ It's my pleasure.
+

vtt/episode_012_small.vtt

@@ -0,0 +1,1937 @@
+WEBVTT
+
+00:00.000 --> 00:05.840
+ The following is a conversation with Thomas Sanholm. He's a professor at SameU and cocreator of
+
+00:05.840 --> 00:11.440
+ Labratus, which is the first AI system to beat top human players in the game of Heads Up No Limit
+
+00:11.440 --> 00:17.840
+ Texas Holdem. He has published over 450 papers on game theory and machine learning, including a
+
+00:17.840 --> 00:25.360
+ best paper in 2017 at NIPS, now renamed to New Rips, which is where I caught up with him for this
+
+00:25.360 --> 00:31.520
+ conversation. His research and companies have had wide reaching impact in the real world,
+
+00:32.080 --> 00:38.720
+ especially because he and his group not only propose new ideas, but also build systems to prove
+
+00:38.720 --> 00:44.800
+ that these ideas work in the real world. This conversation is part of the MIT course on
+
+00:44.800 --> 00:50.400
+ artificial journal intelligence and the artificial intelligence podcast. If you enjoy it, subscribe
+
+00:50.400 --> 00:58.720
+ on YouTube, iTunes, or simply connect with me on Twitter at Lex Friedman, spelled FRID. And now
+
+00:58.720 --> 01:06.960
+ here's my conversation with Thomas Sanholm. Can you describe at the high level the game of poker,
+
+01:06.960 --> 01:13.200
+ Texas Holdem, Heads Up Texas Holdem, for people who might not be familiar with this card game?
+
+01:13.200 --> 01:18.720
+ Yeah, happy to. So Heads Up No Limit Texas Holdem has really emerged in the AI community
+
+01:18.720 --> 01:24.160
+ as a main benchmark for testing these application independent algorithms for
+
+01:24.160 --> 01:31.200
+ imperfect information game solving. And this is a game that's actually played by humans. You don't
+
+01:31.200 --> 01:38.560
+ see that much on TV or casinos because, well, for various reasons, but you do see it in some
+
+01:38.560 --> 01:44.000
+ expert level casinos and you see it in the best poker movies of all time. It's actually an event
+
+01:44.000 --> 01:50.400
+ in the world series of poker, but mostly it's played online and typically for pretty big sums
+
+01:50.400 --> 01:57.200
+ of money. And this is a game that usually only experts play. So if you go to your home game
+
+01:57.200 --> 02:01.680
+ on a Friday night, it probably is not going to be Heads Up No Limit Texas Holdem. It might be
+
+02:02.800 --> 02:08.560
+ No Limit Texas Holdem in some cases, but typically for a big group and it's not as competitive.
+
+02:08.560 --> 02:14.320
+ Well, Heads Up means it's two players, so it's really like me against you. Am I better or are you
+
+02:14.320 --> 02:20.240
+ better, much like chess or go in that sense, but an imperfect information game which makes it much
+
+02:20.240 --> 02:25.920
+ harder because I have to deal with issues of you knowing things that I don't know and I know
+
+02:25.920 --> 02:30.960
+ things that you don't know instead of pieces being nicely laid on the board for both of us to see.
+
+02:30.960 --> 02:38.400
+ So in Texas Holdem, there's two cards that you only see that belong to you and there is
+
+02:38.400 --> 02:44.080
+ they gradually lay out some cards that add up overall to five cards that everybody can see.
+
+02:44.080 --> 02:48.800
+ Yeah, the imperfect nature of the information is the two cards that you're holding up front.
+
+02:48.800 --> 02:54.160
+ Yeah. So as you said, you know, you first get two cards in private each, and then you there's a
+
+02:54.160 --> 02:59.440
+ betting round, then you get three cards in public on the table, then there's a betting round, then
+
+02:59.440 --> 03:03.760
+ you get the fourth card in public on the table, there's a betting round, then you get the fifth
+
+03:03.760 --> 03:08.480
+ card on the table, there's a betting round. So there's a total of four betting rounds and four
+
+03:08.480 --> 03:14.480
+ tranches of information revelation, if you will. The only the first tranche is private and then
+
+03:14.480 --> 03:25.200
+ it's public from there. And this is probably by far the most popular game in AI and just
+
+03:25.200 --> 03:30.080
+ the general public in terms of imperfect information. So it's probably the most popular
+
+03:30.080 --> 03:38.000
+ spectator game to watch, right? So, which is why it's a super exciting game tackle. So it's on
+
+03:38.000 --> 03:44.400
+ the order of chess, I would say in terms of popularity, in terms of AI setting it as the bar
+
+03:44.400 --> 04:01.280
+ of what is intelligence. So in 2017, Libratus beats a few four expert human players. Can you
+
+04:01.280 --> 04:06.720
+ describe that event? What you learned from it? What was it like? What was the process in general
+
+04:06.720 --> 04:12.960
+ for people who have not read the papers in the study? Yeah. So the event was that we invited
+
+04:12.960 --> 04:17.600
+ four of the top 10 players with these specialist players in Heads Up No Limit Texas Holden,
+
+04:17.600 --> 04:22.960
+ which is very important because this game is actually quite different than the multiplayer
+
+04:22.960 --> 04:29.360
+ version. We brought them in to Pittsburgh to play at the reverse casino for 20 days. We wanted to
+
+04:29.360 --> 04:37.440
+ get 120,000 hands in because we wanted to get statistical significance. So it's a lot of hands
+
+04:37.440 --> 04:44.000
+ for humans to play, even for these top pros who play fairly quickly normally. So we couldn't just
+
+04:44.000 --> 04:49.760
+ have one of them play so many hands. 20 days, they were playing basically morning to evening.
+
+04:50.320 --> 04:57.920
+ And you raised 200,000 as a little incentive for them to play. And the setting was so that
+
+04:57.920 --> 05:05.280
+ they didn't all get 50,000. We actually paid them out based on how they did against the AI each.
+
+05:05.280 --> 05:11.040
+ So they had an incentive to play as hard as they could, whether they're way ahead or way behind
+
+05:11.040 --> 05:16.400
+ or right at the mark of beating the AI. And you don't make any money, unfortunately. Right. No,
+
+05:16.400 --> 05:22.560
+ we can't make any money. So originally, a couple of years earlier, I actually explored whether we
+
+05:22.560 --> 05:28.400
+ could actually play for money because that would be, of course, interesting as well to play against
+
+05:28.400 --> 05:33.280
+ the top people for money. But the Pennsylvania Gaming Board said no. So we couldn't. So this
+
+05:33.280 --> 05:40.160
+ is much like an exhibit, like for a musician or a boxer or something like that. Nevertheless,
+
+05:40.160 --> 05:49.040
+ they were keeping track of the money and brought us one close to $2 million, I think. So if it was
+
+05:49.040 --> 05:55.200
+ for real money, if you were able to earn money, that was a quite impressive and inspiring achievement.
+
+05:55.200 --> 06:00.720
+ Just a few details. What were the players looking at? Were they behind a computer? What
+
+06:00.720 --> 06:06.080
+ was the interface like? Yes, they were playing much like they normally do. These top players,
+
+06:06.080 --> 06:11.600
+ when they play this game, they play mostly online. So they used to playing through a UI.
+
+06:11.600 --> 06:16.080
+ And they did the same thing here. So there was this layout, you could imagine. There's a table
+
+06:16.880 --> 06:22.880
+ on a screen. There's the human sitting there. And then there's the AI sitting there. And the
+
+06:22.880 --> 06:27.440
+ screen shows everything that's happening. The cards coming out and shows the bets being made.
+
+06:27.440 --> 06:32.320
+ And we also had the betting history for the human. So if the human forgot what had happened in the
+
+06:32.320 --> 06:39.040
+ hand so far, they could actually reference back and so forth. Is there a reason they were given
+
+06:39.040 --> 06:46.880
+ access to the betting history? Well, it didn't really matter. They wouldn't have forgotten
+
+06:46.880 --> 06:52.160
+ anywhere. These are top quality people. But we just wanted to put out there so it's not the question
+
+06:52.160 --> 06:56.560
+ of the human forgetting and the AI somehow trying to get that advantage of better memory.
+
+06:56.560 --> 07:01.040
+ So what was that like? I mean, that was an incredible accomplishment. So what did it feel like
+
+07:01.040 --> 07:07.440
+ before the event? Did you have doubt, hope? Where was your confidence at?
+
+07:08.080 --> 07:13.760
+ Yeah, that's great. Great question. So 18 months earlier, I had organized a similar
+
+07:13.760 --> 07:19.200
+ brains versus AI competition with a previous AI called Cloudical. And we couldn't beat the humans.
+
+07:20.480 --> 07:26.400
+ So this time around, it was only 18 months later. And I knew that this new AI, Libratus,
+
+07:26.400 --> 07:32.320
+ was way stronger. But it's hard to say how you'll do against the top humans before you try.
+
+07:32.320 --> 07:39.040
+ So I thought we had about a 50 50 shot. And the international betting sites put us as a
+
+07:39.040 --> 07:45.440
+ four to one or five to one underdog. So it's kind of interesting that people really believe in people
+
+07:45.440 --> 07:50.560
+ and over AI, not just people, people don't just believe over believing themselves,
+
+07:50.560 --> 07:54.800
+ but they have overconfidence in other people as well, compared to the performance of AI.
+
+07:54.800 --> 08:00.800
+ And yeah, so we were a four to one or five to one underdog. And even after three days of beating
+
+08:00.800 --> 08:05.680
+ the humans in a row, we were still 50 50 on the international betting sites.
+
+08:05.680 --> 08:11.120
+ Do you think there's something special and magical about poker in the way people think about it?
+
+08:11.120 --> 08:19.120
+ In the sense you have, I mean, even in chess, there's no Hollywood movies. Poker is the star
+
+08:19.120 --> 08:29.760
+ of many movies. And there's this feeling that certain human facial expressions and body language,
+
+08:29.760 --> 08:34.880
+ eye movement, all these tells are critical to poker. Like you can look into somebody's soul
+
+08:34.880 --> 08:41.760
+ and understand their betting strategy and so on. So that's probably why the possibly do you think
+
+08:41.760 --> 08:48.000
+ that is why people have a confidence that humans will outperform because AI systems cannot in this
+
+08:48.000 --> 08:53.360
+ construct perceive these kinds of tells, they're only looking at betting patterns and
+
+08:55.600 --> 09:03.920
+ nothing else, the betting patterns and statistics. So what's more important to you if you step back
+
+09:03.920 --> 09:12.960
+ on human players, human versus human? What's the role of these tells of these ideas that we romanticize?
+
+09:12.960 --> 09:21.760
+ Yeah, so I'll split it into two parts. So one is why do humans trust humans more than AI and
+
+09:21.760 --> 09:27.040
+ have overconfidence in humans? I think that's not really related to the tell question. It's
+
+09:27.040 --> 09:32.480
+ just that they've seen these top players how good they are and they're really fantastic. So it's just
+
+09:33.120 --> 09:39.200
+ hard to believe that the AI could beat them. So I think that's where that comes from. And that's
+
+09:39.200 --> 09:44.080
+ actually maybe a more general lesson about AI that until you've seen it overperform a human,
+
+09:44.080 --> 09:52.080
+ it's hard to believe that it could. But then the tells, a lot of these top players, they're so good
+
+09:52.080 --> 10:00.320
+ at hiding tells that among the top players, it's actually not really worth it for them to invest a
+
+10:00.320 --> 10:06.560
+ lot of effort trying to find tells in each other because they're so good at hiding them. So yes,
+
+10:06.560 --> 10:12.960
+ at the kind of Friday evening game, tells are going to be a huge thing. You can read other people
+
+10:12.960 --> 10:17.760
+ and if you're a good reader, you'll read them like an open book. But at the top levels of poker,
+
+10:17.760 --> 10:23.120
+ no, the tells become a less much smaller and smaller aspect of the game as you go to the top
+
+10:23.120 --> 10:32.960
+ levels. The amount of strategies, the amounts of possible actions is very large, 10 to the power
+
+10:32.960 --> 10:39.280
+ of 100 plus. So there has to be some, I've read a few of the papers related that has,
+
+10:40.400 --> 10:46.880
+ it has to form some abstractions of various hands and actions. So what kind of abstractions are
+
+10:46.880 --> 10:52.800
+ effective for the game of poker? Yeah, so you're exactly right. So when you go from a game tree
+
+10:52.800 --> 10:59.280
+ that's 10 to the 161, especially in an imperfect information game, it's way too large to solve
+
+10:59.280 --> 11:06.160
+ directly. Even with our fastest equilibrium finding algorithms. So you want to abstract it
+
+11:06.160 --> 11:13.840
+ first. And abstraction in games is much trickier than abstraction in MDPs or other single agent
+
+11:13.840 --> 11:19.440
+ settings. Because you have these abstraction pathologies. But if I have a finer grained abstraction,
+
+11:20.560 --> 11:25.120
+ the strategy that I can get from that for the real game might actually be worse than the strategy
+
+11:25.120 --> 11:29.440
+ I can get from the coarse grained abstraction. So you have to be very careful. Now, the kinds
+
+11:29.440 --> 11:34.880
+ of abstractions just to zoom out, we're talking about there's the hands abstractions and then there
+
+11:34.880 --> 11:42.480
+ is betting strategies. Yeah, betting actions. So there's information abstraction to talk about
+
+11:42.480 --> 11:47.760
+ general games, information abstraction, which is the abstraction of what chance does. And this
+
+11:47.760 --> 11:54.400
+ would be the cards in the case of poker. And then there's action abstraction, which is abstracting
+
+11:54.400 --> 12:00.640
+ the actions of the actual players, which would be bets in the case of poker, yourself, any other
+
+12:00.640 --> 12:10.240
+ players, yes, yourself and other players. And for information abstraction, we were completely
+
+12:10.240 --> 12:16.640
+ automated. So these were these are algorithms, but they do what we call potential aware abstraction,
+
+12:16.640 --> 12:20.880
+ where we don't just look at the value of the hand, but also how it might materialize in the
+
+12:20.880 --> 12:26.320
+ good or bad hands over time. And it's a certain kind of bottom up process with integer programming
+
+12:26.320 --> 12:32.640
+ there and clustering and various aspects, how do you build this abstraction. And then in the
+
+12:32.640 --> 12:40.720
+ action abstraction, there, it's largely based on how humans other and other AI's have played
+
+12:40.720 --> 12:46.560
+ this game in the past. But in the beginning, we actually use an automated action abstraction
+
+12:46.560 --> 12:53.840
+ technology, which is provably convergent, that it finds the optimal combination of bet sizes,
+
+12:53.840 --> 12:58.320
+ but it's not very scalable. So we couldn't use it for the whole game, but we use it for the first
+
+12:58.320 --> 13:03.760
+ couple of betting actions. So what's more important, the strength of the hand, so the
+
+13:03.760 --> 13:12.320
+ information abstraction or the how you play them, the actions, does it, you know, the romanticized
+
+13:12.320 --> 13:17.120
+ notion again, is that it doesn't matter what hands you have, that the actions, the betting,
+
+13:18.000 --> 13:22.240
+ maybe the way you win, no matter what hands you have. Yeah, so that's why you have to play a lot
+
+13:22.240 --> 13:29.440
+ of hands, so that the role of luck gets smaller. So you could otherwise get lucky and get some good
+
+13:29.440 --> 13:34.240
+ hands, and then you're going to win the match. Even with thousands of hands, you can get lucky.
+
+13:35.120 --> 13:40.720
+ Because there's so much variance in no limit, Texas hold them, because if we both go all in,
+
+13:40.720 --> 13:47.920
+ it's a huge stack of variance. So there are these massive swings in no limit Texas hold them. So
+
+13:47.920 --> 13:53.600
+ that's why you have to play not just thousands, but over a hundred thousand hands to get statistical
+
+13:53.600 --> 14:00.400
+ significance. So let me ask another way this question. If you didn't even look at your hands,
+
+14:01.920 --> 14:05.840
+ but they didn't know that the opponents didn't know that, how well would you be able to do?
+
+14:05.840 --> 14:11.440
+ That's a good question. There's actually, I heard the story that there's this Norwegian female poker
+
+14:11.440 --> 14:17.040
+ player called Annette Oberstad, who's actually won a tournament by doing exactly that. But that
+
+14:17.040 --> 14:26.160
+ would be extremely rare. So you cannot really play well that way. Okay, so the hands do have
+
+14:26.160 --> 14:34.320
+ some role to play. So Lebrados does not use, as far as I understand, use learning methods,
+
+14:34.320 --> 14:42.880
+ deep learning. Is there room for learning in, you know, there's no reason why Lebrados doesn't,
+
+14:42.880 --> 14:48.640
+ you know, combine with an alpha go type approach for estimating the quality for function estimator.
+
+14:49.520 --> 14:55.040
+ What are your thoughts on this? Maybe as compared to another algorithm, which I'm not
+
+14:55.040 --> 15:01.120
+ that familiar with deep stack, the engine that does use deep learning that is unclear how well
+
+15:01.120 --> 15:05.680
+ it does, but nevertheless uses deep learning. So what are your thoughts about learning methods to
+
+15:05.680 --> 15:11.520
+ aid in the way that Lebrados plays the game of poker? Yeah, so as you said, Lebrados did not
+
+15:11.520 --> 15:17.760
+ use learning methods and played very well without them. Since then, we have actually actually here,
+
+15:17.760 --> 15:25.200
+ we have a couple of papers on things that do use learning techniques. Excellent. So and deep learning
+
+15:25.200 --> 15:32.560
+ in particular, and sort of the way you're talking about where it's learning an evaluation function.
+
+15:33.360 --> 15:42.400
+ But in imperfect information games, unlike, let's say in go war, now also in chess and shogi,
+
+15:42.400 --> 15:52.160
+ it's not sufficient to learn an evaluation for a state because the value of an information set
+
+15:52.160 --> 16:00.080
+ depends not only on the exact state, but it also depends on both players beliefs. Like if I have
+
+16:00.080 --> 16:05.360
+ a bad hand, I'm much better off if the opponent thinks I have a good hand. And vice versa,
+
+16:05.360 --> 16:12.640
+ if I have a good hand, I'm much better off if the opponent believes I have a bad hand. So the value
+
+16:12.640 --> 16:18.720
+ of a state is not just a function of the cards. It depends on if you will, the path of play,
+
+16:18.720 --> 16:25.440
+ but only to the extent that it's captured in the belief distributions. So that's why it's not as
+
+16:25.440 --> 16:31.040
+ simple as it is in perfect information games. And I don't want to say it's simple there either.
+
+16:31.040 --> 16:35.680
+ It's of course, very complicated computationally there too. But at least conceptually, it's very
+
+16:35.680 --> 16:39.520
+ straightforward. There's a state, there's an evaluation function, you can try to learn it.
+
+16:40.080 --> 16:48.640
+ Here, you have to do something more. And what we do is in one of these papers, we're looking at
+
+16:48.640 --> 16:54.000
+ allowing where we allow the opponent to actually take different strategies at the leaf of the
+
+16:54.000 --> 17:01.600
+ search tree, if you will. And that is a different way of doing it. And it doesn't assume therefore
+
+17:01.600 --> 17:07.200
+ a particular way that the opponent plays. But it allows the opponent to choose from a set of
+
+17:07.200 --> 17:14.560
+ different continuation strategies. And that forces us to not be too optimistic in a lookahead
+
+17:14.560 --> 17:20.720
+ search. And that's that's one way you can do sound lookahead search in imperfect information
+
+17:20.720 --> 17:26.400
+ games, which is very different, difficult. And in US, you were asking about deep stack, what they
+
+17:26.400 --> 17:30.960
+ did, it was very different than what we do, either in Libertadores or in this new work.
+
+17:31.920 --> 17:37.120
+ They were generally randomly generating various situations in the game. Then they were doing
+
+17:37.120 --> 17:42.960
+ the lookahead from there to the end of the game, as if that was the start of a different game. And
+
+17:42.960 --> 17:48.800
+ then they were using deep learning to learn those values of those states. But the states were not
+
+17:48.800 --> 17:54.160
+ just the physical states, they include the belief distributions. When you talk about lookahead,
+
+17:55.680 --> 18:01.600
+ or deep stack, or with Libertadores, does it mean considering every possibility that the game can
+
+18:01.600 --> 18:08.160
+ evolve? Are we talking about extremely sort of this exponential growth of a tree? Yes. So we're
+
+18:08.160 --> 18:15.760
+ talking about exactly that. Much like you doing alpha beta search or Monte Carlo tree search,
+
+18:15.760 --> 18:19.920
+ but with different techniques. So there's a different search algorithm. And then we have to
+
+18:19.920 --> 18:24.400
+ deal with the leaves differently. So if you think about what Libertadores did, we didn't have to
+
+18:24.400 --> 18:30.880
+ worry about this, because we only did it at the end of the game. So we would always terminate into a
+
+18:30.880 --> 18:36.720
+ real situation. And we would know what the payout is. It didn't do these depth limited lookaheads.
+
+18:36.720 --> 18:42.000
+ But now in this new paper, which is called depth limited, I think it's called depth limited search
+
+18:42.000 --> 18:47.360
+ for imperfect information games, we can actually do sound depth limited lookaheads. So we can
+
+18:47.360 --> 18:52.080
+ actually start to do the lookahead from the beginning of the game on, because that's too
+
+18:52.080 --> 18:57.680
+ complicated to do for this whole long game. So in Libertadores, we were just doing it for the end.
+
+18:57.680 --> 19:03.040
+ So and then the other side, this belief distribution. So is it explicitly modeled
+
+19:03.040 --> 19:10.800
+ what kind of beliefs that the opponent might have? Yeah, it is explicitly modeled, but it's not
+
+19:10.800 --> 19:18.720
+ assumed that beliefs are actually output, not input. Of course, the starting beliefs are input,
+
+19:18.720 --> 19:23.440
+ but they just fall from the rules of the game, because we know that the dealer deals uniformly
+
+19:23.440 --> 19:29.600
+ from the deck. So I know that every pair of cards that you might have is equally likely.
+
+19:29.600 --> 19:33.600
+ I know that for a fact, that just follows from the rules of the game. Of course,
+
+19:33.600 --> 19:38.320
+ except the two cards that I have, I know you don't have those. You have to take that into
+
+19:38.320 --> 19:42.480
+ account. That's called card removal, and that's very important. Is the dealing always coming
+
+19:42.480 --> 19:50.000
+ from a single deck in heads up? Yes. So you can assume single deck. So you know that if I have
+
+19:50.000 --> 19:55.680
+ the ace of spades, I know you don't have an ace of spades. So in the beginning, your belief is
+
+19:55.680 --> 20:02.640
+ basically the fact that it's a fair dealing of hands. But how do you start to adjust that belief?
+
+20:02.640 --> 20:08.960
+ Well, that's where this beauty of game theory comes. So Nash equilibrium, which John Nash
+
+20:08.960 --> 20:14.960
+ introduced in 1950, introduces what rational play is when you have more than one player.
+
+20:15.920 --> 20:21.200
+ And these are pairs of strategies where strategies are contingency plans, one for each player.
+
+20:21.200 --> 20:27.920
+ So that neither player wants to deviate to a different strategy, given that the other
+
+20:27.920 --> 20:34.800
+ doesn't deviate. But as a side effect, you get the beliefs from base rule. So Nash equilibrium
+
+20:34.800 --> 20:39.760
+ really isn't just deriving in these imperfect information games. Nash equilibrium doesn't
+
+20:39.760 --> 20:47.920
+ just define strategies. It also defines beliefs for both of us, and it defines beliefs for each state.
+
+20:47.920 --> 20:54.720
+ So at each state, each, if they call information sets, at each information set in the game,
+
+20:54.720 --> 20:59.840
+ there's a set of different states that we might be in, but I don't know which one we're in.
+
+21:00.960 --> 21:05.440
+ Nash equilibrium tells me exactly what is the probability distribution over those real
+
+21:05.440 --> 21:11.360
+ wall states in my mind. How does Nash equilibrium give you that distribution? So why?
+
+21:11.360 --> 21:17.280
+ I'll do a simple example. So you know the game Rock Paper Scissors? So we can draw it
+
+21:17.280 --> 21:23.600
+ as player one moves first, and then player two moves. But of course, it's important that player
+
+21:23.600 --> 21:28.560
+ two doesn't know what player one moved. Otherwise player two would win every time. So we can draw
+
+21:28.560 --> 21:33.440
+ that as an information set where player one makes one or three moves first. And then there's an
+
+21:33.440 --> 21:40.480
+ information set for player two. So player two doesn't know which of those nodes the world is in.
+
+21:40.480 --> 21:46.400
+ But once we know the strategy for player one, Nash equilibrium will say that you play one third
+
+21:46.400 --> 21:52.160
+ rock, one third paper, one third scissors. From that I can derive my beliefs on the information
+
+21:52.160 --> 21:58.480
+ set that they're one third, one third, one third. So Bayes gives you that. But is that specific
+
+21:58.480 --> 22:06.800
+ to a particular player? Or is it something you quickly update with those? No, the game theory
+
+22:06.800 --> 22:12.560
+ isn't really player specific. So that's also why we don't need any data. We don't need any history
+
+22:12.560 --> 22:17.840
+ how these particular humans played in the past or how any AI or even had played before. It's all
+
+22:17.840 --> 22:24.160
+ about rationality. So we just think the AI just thinks about what would a rational opponent do?
+
+22:24.720 --> 22:30.960
+ And what would I do if I were I am rational and what that that's that's the idea of game theory.
+
+22:30.960 --> 22:38.080
+ So it's really a data free opponent free approach. So it comes from the design of the game as opposed
+
+22:38.080 --> 22:43.600
+ to the design of the player. Exactly. There's no opponent modeling per se. I mean, we've done
+
+22:43.600 --> 22:47.760
+ some work on combining opponent modeling with game theory. So you can exploit weak players
+
+22:47.760 --> 22:53.440
+ even more. But that's another strand and in Libra, there's wouldn't turn that on. So I decided that
+
+22:53.440 --> 22:59.520
+ these players are too good. And when you start to exploit an opponent, you typically open yourself
+
+22:59.520 --> 23:04.720
+ up self up to exploitation. And these guys have so few holes to exploit and they're world's leading
+
+23:04.720 --> 23:09.120
+ experts in counter exploitation. So I decided that we're not going to turn that stuff on.
+
+23:09.120 --> 23:14.400
+ Actually, I saw a few your papers exploiting opponents sound very interesting to explore.
+
+23:15.600 --> 23:19.120
+ Do you think there's room for exploitation, generally outside of the broadest?
+
+23:19.840 --> 23:27.840
+ Is there a subject or people differences that could be exploited? Maybe not just in poker,
+
+23:27.840 --> 23:33.360
+ but in general interactions and negotiations all these other domains that you're considering?
+
+23:33.360 --> 23:39.760
+ Yeah, definitely. We've done some work on that. And I really like the work that hybridizes the two.
+
+23:39.760 --> 23:45.200
+ So you figure out what would a rational opponent do. And by the way, that's safe in these zero
+
+23:45.200 --> 23:49.440
+ sum games to players, zero sum games, because if the opponent does something irrational,
+
+23:49.440 --> 23:56.320
+ yes, it might show a throw of my beliefs. But the amount that the player can gain by throwing
+
+23:56.320 --> 24:04.240
+ of my belief is always less than they lose by playing poorly. So it's safe. But still,
+
+24:04.240 --> 24:09.280
+ if somebody's weak as a player, you might want to play differently to exploit them more.
+
+24:10.160 --> 24:14.560
+ So you can think about it this way, a game theoretic strategy is unbeatable,
+
+24:15.600 --> 24:22.720
+ but it doesn't maximally beat the other opponent. So the winnings per hand might be better
+
+24:22.720 --> 24:27.120
+ with a different strategy. And the hybrid is that you start from a game theoretic approach,
+
+24:27.120 --> 24:33.040
+ and then as you gain data about the opponent in certain parts of the game tree, then in those
+
+24:33.040 --> 24:39.280
+ parts of the game tree, you start to tweak your strategy more and more towards exploitation,
+
+24:39.280 --> 24:44.160
+ while still staying fairly close to the game theoretic strategy so as to not open yourself
+
+24:44.160 --> 24:53.600
+ up to exploitation too much. How do you do that? Do you try to vary up strategies, make it unpredictable?
+
+24:53.600 --> 24:59.520
+ It's like, what is it, tit for tat strategies in Prisoner's Dilemma or?
+
+25:00.560 --> 25:07.440
+ Well, that's a repeated game, simple Prisoner's Dilemma, repeated games. But even there,
+
+25:07.440 --> 25:13.120
+ there's no proof that says that that's the best thing. But experimentally, it actually does well.
+
+25:13.120 --> 25:17.360
+ So what kind of games are there, first of all? I don't know if this is something that you could
+
+25:17.360 --> 25:21.760
+ just summarize. There's perfect information games with all the information on the table.
+
+25:22.320 --> 25:27.680
+ There is imperfect information games. There's repeated games that you play over and over.
+
+25:28.480 --> 25:36.960
+ There's zero sum games. There's non zero sum games. And then there's a really important
+
+25:36.960 --> 25:44.640
+ distinction you're making to player versus more players. So what are what other games are there?
+
+25:44.640 --> 25:49.920
+ And what's the difference, for example, with this two player game versus more players? Yeah,
+
+25:49.920 --> 25:54.720
+ what are the key differences? Right. So let me start from the the basics. So
+
+25:56.320 --> 26:02.160
+ a repeated game is a game where the same exact game is played over and over.
+
+26:02.160 --> 26:08.160
+ In these extensive form games, where you think about three form, maybe with these
+
+26:08.160 --> 26:14.000
+ information sets to represent incomplete information, you can have kind of repetitive
+
+26:14.000 --> 26:18.480
+ interactions and even repeated games are a special case of that, by the way. But
+
+26:19.680 --> 26:24.160
+ the game doesn't have to be exactly the same. It's like in sourcing options. Yes, we're going to
+
+26:24.160 --> 26:29.040
+ see the same supply base year to year. But what I'm buying is a little different every time.
+
+26:29.040 --> 26:34.080
+ And the supply base is a little different every time and so on. So it's not really repeated.
+
+26:34.080 --> 26:39.680
+ So to find a purely repeated game is actually very rare in the world. So they're really a very
+
+26:40.960 --> 26:47.360
+ coarse model of what's going on. Then if you move up from just repeated, simple,
+
+26:47.360 --> 26:52.560
+ repeated matrix games, not all the way to extensive form games, but in between,
+
+26:52.560 --> 26:59.360
+ there's stochastic games, where you know, there's these, you think about it like these little
+
+26:59.360 --> 27:06.080
+ matrix games. And when you take an action and your own takes an action, they determine not which
+
+27:06.080 --> 27:11.280
+ next state I'm going to next game, I'm going to, but the distribution over next games,
+
+27:11.280 --> 27:15.760
+ where I might be going to. So that's the stochastic game. But it's like
+
+27:15.760 --> 27:22.160
+ like matrix games, repeated stochastic games, extensive form games, that is from less to more
+
+27:22.160 --> 27:28.080
+ general. And poker is an example of the last one. So it's really in the most general setting,
+
+27:29.440 --> 27:34.720
+ extensive form games. And that's kind of what the AI community has been working on and being
+
+27:34.720 --> 27:39.680
+ benchmarked on with this heads up no limit, Texas Holden. Can you describe extensive form games?
+
+27:39.680 --> 27:45.600
+ What's the model here? So if you're familiar with the tree form, so it's really the tree form,
+
+27:45.600 --> 27:51.280
+ like in chess, there's a search tree versus a matrix versus a matrix. Yeah. And that's the
+
+27:51.280 --> 27:56.640
+ matrix is called the matrix form or by matrix form or normal form game. And here you have the tree
+
+27:56.640 --> 28:01.680
+ form. So you can actually do certain types of reasoning there, that you lose the information
+
+28:02.320 --> 28:07.840
+ when you go to normal form. There's a certain form of equivalence, like if you go from three
+
+28:07.840 --> 28:13.840
+ form and you say it every possible contingency plan is a strategy, then I can actually go back
+
+28:13.840 --> 28:19.600
+ to the normal form, but I lose some information from the lack of sequentiality. Then the multiplayer
+
+28:19.600 --> 28:29.520
+ versus two player distinction is an important one. So two player games in zero sum are conceptually
+
+28:29.520 --> 28:37.040
+ easier and computationally easier. They're still huge like this one, this one. But they're conceptually
+
+28:37.040 --> 28:42.160
+ easier and computationally easier. In that conceptually, you don't have to worry about
+
+28:42.160 --> 28:47.440
+ which equilibrium is the other guy going to play when there are multiple, because any equilibrium
+
+28:47.440 --> 28:52.400
+ strategy is the best response to any other equilibrium strategy. So I can play a different
+
+28:52.400 --> 28:57.600
+ equilibrium from you and we'll still get the right values of the game. That falls apart even
+
+28:57.600 --> 29:02.960
+ with two players when you have general sum games. Even without cooperation, just even without
+
+29:02.960 --> 29:09.040
+ cooperation. So there's a big gap from two player zero sum to two player general sum, or even to
+
+29:09.040 --> 29:17.600
+ three players zero sum. That's a big gap, at least in theory. Can you maybe non mathematically
+
+29:17.600 --> 29:23.040
+ provide the intuition why it all falls apart with three or more players? It seems like you should
+
+29:23.040 --> 29:32.640
+ still be able to have a Nash equilibrium that's instructive, that holds. Okay, so it is true
+
+29:32.640 --> 29:39.840
+ that all finite games have a Nash equilibrium. So this is what your Nash actually proved.
+
+29:40.880 --> 29:45.600
+ So they do have a Nash equilibrium. That's not the problem. The problem is that there can be many.
+
+29:46.480 --> 29:52.000
+ And then there's a question of which equilibrium to select. So and if you select your strategy
+
+29:52.000 --> 30:00.960
+ from a different equilibrium and I select mine, then what does that mean? And in these non zero
+
+30:00.960 --> 30:07.760
+ sum games, we may lose some joint benefit by being just simply stupid, we could actually both be
+
+30:07.760 --> 30:12.960
+ better off if we did something else. And in three player, you get other problems also like collusion.
+
+30:12.960 --> 30:19.600
+ Like maybe you and I can gang up on a third player, and we can do radically better by colluding.
+
+30:19.600 --> 30:25.440
+ So there are lots of issues that come up there. So No Brown, the student you work with on this,
+
+30:25.440 --> 30:31.120
+ has mentioned, I looked through the AMA on Reddit, he mentioned that the ability of poker players
+
+30:31.120 --> 30:36.320
+ to collaborate would make the game. He was asked the question of, how would you make the game of
+
+30:36.320 --> 30:42.320
+ poker? Or both of you were asked the question, how would you make the game of poker beyond
+
+30:43.440 --> 30:51.440
+ being solvable by current AI methods? And he said that there's not many ways of making poker more
+
+30:51.440 --> 30:59.600
+ difficult, but a collaboration or cooperation between players would make it extremely difficult.
+
+30:59.600 --> 31:05.200
+ So can you provide the intuition behind why that is, if you agree with that idea?
+
+31:05.200 --> 31:11.840
+ Yeah, so we've done a lot of work on coalitional games. And we actually have a paper here with
+
+31:11.840 --> 31:17.040
+ my other student Gabriella Farina and some other collaborators on at NIPPS on that actually just
+
+31:17.040 --> 31:22.080
+ came back from the poster session where we presented this. So when you have a collusion,
+
+31:22.080 --> 31:29.440
+ it's a different problem. And it typically gets even harder then. Even the game representations,
+
+31:29.440 --> 31:35.840
+ some of the game representations don't really allow good computation. So we actually introduced a new
+
+31:35.840 --> 31:43.840
+ game representation for that. Is that kind of cooperation part of the model? Do you have
+
+31:43.840 --> 31:48.720
+ information about the fact that other players are cooperating? Or is it just this chaos that
+
+31:48.720 --> 31:53.760
+ where nothing is known? So there's some some things unknown. Can you give an example of a
+
+31:53.760 --> 31:59.920
+ collusion type game? Or is it usually? So like bridge. Yeah, so think about bridge. It's like
+
+31:59.920 --> 32:06.960
+ when you and I are on a team, our payoffs are the same. The problem is that we can't talk. So when
+
+32:06.960 --> 32:13.360
+ I get my cards, I can't whisper to you what my cards are. That would not be allowed. So we have
+
+32:13.360 --> 32:20.480
+ to somehow coordinate our strategies ahead of time. And only ahead of time. And then there's
+
+32:20.480 --> 32:25.920
+ certain signals we can talk about. But they have to be such that the other team also understands
+
+32:25.920 --> 32:31.920
+ that. So so that that's that's an example where the coordination is already built into the rules
+
+32:31.920 --> 32:38.400
+ of the game. But in many other situations, like auctions or negotiations or diplomatic
+
+32:38.400 --> 32:44.320
+ relationships, poker, it's not really built in. But it still can be very helpful for the
+
+32:44.320 --> 32:51.440
+ colliders. I've read you write somewhere, the negotiations, you come to the table with prior
+
+32:52.640 --> 32:58.160
+ like a strategy that, like that you're willing to do and not willing to do those kinds of things.
+
+32:58.160 --> 33:04.320
+ So how do you start to now moving away from poker, moving beyond poker into other applications
+
+33:04.320 --> 33:10.960
+ like negotiations? How do you start applying this to other to other domains, even real world
+
+33:10.960 --> 33:15.360
+ domains that you've worked on? Yeah, I actually have two startup companies doing exactly that.
+
+33:15.360 --> 33:21.520
+ One is called Strategic Machine. And that's for kind of business applications, gaming, sports,
+
+33:21.520 --> 33:29.040
+ all sorts of things like that. Any applications of this to business and to sports and to gaming,
+
+33:29.040 --> 33:34.800
+ to various types of things for in finance, electricity markets and so on. And the other
+
+33:34.800 --> 33:41.360
+ is called Strategy Robot, where we are taking these to military security, cybersecurity,
+
+33:41.360 --> 33:45.760
+ and intelligence applications. I think you worked a little bit in
+
+33:47.920 --> 33:54.560
+ how do you put it, advertisement, sort of suggesting ads kind of thing.
+
+33:54.560 --> 33:59.040
+ Yeah, that's another company, Optimized Markets. But that's much more about a
+
+33:59.040 --> 34:03.840
+ combinatorial market and optimization based technology. That's not using these
+
+34:04.720 --> 34:12.400
+ game theory decreasing technologies. I see. Okay, so what sort of high level do you think about
+
+34:13.040 --> 34:20.320
+ our ability to use game theoretic concepts to model human behavior? Do you think human behavior is
+
+34:20.320 --> 34:25.680
+ amenable to this kind of modeling? So outside of the poker games and where have you seen it
+
+34:25.680 --> 34:32.000
+ done successfully in your work? I'm not sure. The goal really is modeling humans.
+
+34:33.520 --> 34:40.240
+ Like for example, if I'm playing a zero sum game, I don't really care that the opponent is actually
+
+34:40.240 --> 34:45.680
+ following my model of rational behavior. Because if they're not, that's even better for me.
+
+34:45.680 --> 34:55.440
+ All right, so see with the opponents in games, the prerequisite is that you've formalized
+
+34:56.240 --> 35:02.720
+ the interaction in some way that can be amenable to analysis. I mean, you've done this amazing work
+
+35:02.720 --> 35:12.160
+ with mechanism design, designing games that have certain outcomes. But so I'll tell you an example
+
+35:12.160 --> 35:19.360
+ for my world of autonomous vehicles. We're studying pedestrians and pedestrians and cars
+
+35:19.360 --> 35:25.040
+ negotiate in this nonverbal communication. There's this weird game dance of tension where
+
+35:25.760 --> 35:30.400
+ pedestrians are basically saying, I trust that you won't kill me. And so as a J Walker,
+
+35:30.400 --> 35:34.560
+ I will step onto the road even though I'm breaking the law and there's this tension.
+
+35:34.560 --> 35:40.480
+ And the question is, we really don't know how to model that well in trying to model intent.
+
+35:40.480 --> 35:46.240
+ And so people sometimes bring up ideas of game theory and so on. Do you think that aspect
+
+35:47.120 --> 35:53.520
+ of human behavior can use these kinds of imperfect information approaches, modeling?
+
+35:54.800 --> 36:00.800
+ How do you start to attack a problem like that when you don't even know how to design the game
+
+36:00.800 --> 36:06.640
+ to describe the situation in order to solve it? Okay, so I haven't really thought about J walking.
+
+36:06.640 --> 36:12.080
+ But one thing that I think could be a good application in autonomous vehicles is the
+
+36:12.080 --> 36:18.240
+ following. So let's say that you have fleets of autonomous cars operating by different companies.
+
+36:18.240 --> 36:23.520
+ So maybe here's the Waymo fleet and here's the Uber fleet. If you think about the rules of the road,
+
+36:24.160 --> 36:29.920
+ they define certain legal rules, but that still leaves a huge strategy space open.
+
+36:29.920 --> 36:33.760
+ Like as a simple example, when cars merge, you know, how humans merge, you know,
+
+36:33.760 --> 36:40.800
+ they slow down and look at each other and try to merge. Wouldn't it be better if these situations
+
+36:40.800 --> 36:46.240
+ would already be prenegotiated so we can actually merge at full speed and we know that this is the
+
+36:46.240 --> 36:51.680
+ situation, this is how we do it and it's all going to be faster. But there are way too many
+
+36:51.680 --> 36:57.600
+ situations to negotiate manually. So you could use automated negotiation. This is the idea at least.
+
+36:57.600 --> 37:04.160
+ You could use automated negotiation to negotiate all of these situations or many of them in advance.
+
+37:04.160 --> 37:09.040
+ And of course, it might be that, hey, maybe you're not going to always let me go first.
+
+37:09.040 --> 37:13.520
+ Maybe you said, okay, well, in these situations, I'll let you go first. But in exchange, you're
+
+37:13.520 --> 37:18.240
+ going to give me two hours, you're going to let me go first in these situations. So it's this huge
+
+37:18.240 --> 37:24.240
+ combinatorial negotiation. And do you think there's room in that example of merging to
+
+37:24.240 --> 37:28.080
+ model this whole situation as an imperfect information game? Or do you really want to
+
+37:28.080 --> 37:33.520
+ consider it to be a perfect? No, that's a good question. Yeah. That's a good question. Do you
+
+37:33.520 --> 37:41.120
+ pay the price of assuming that you don't know everything? Yeah, I don't know. It's certainly
+
+37:41.120 --> 37:47.040
+ much easier. Games with perfect information are much easier. So if you can get away with it,
+
+37:48.240 --> 37:52.960
+ you should. But if the real situation is of imperfect information, then you're going to
+
+37:52.960 --> 37:58.480
+ have to deal with imperfect information. Great. So what lessons have you learned the annual
+
+37:58.480 --> 38:04.560
+ computer poker competition? An incredible accomplishment of AI. You look at the history
+
+38:04.560 --> 38:12.240
+ of the blue AlphaGo, these kind of moments when AI stepped up in an engineering effort and a
+
+38:12.240 --> 38:18.240
+ scientific effort combined to beat the best human player. So what do you take away from
+
+38:18.240 --> 38:23.200
+ this whole experience? What have you learned about designing AI systems that play these kinds
+
+38:23.200 --> 38:30.000
+ of games? And what does that mean for AI in general, for the future of AI development?
+
+38:30.720 --> 38:34.160
+ Yeah, so that's a good question. So there's so much to say about it.
+
+38:35.280 --> 38:40.320
+ I do like this type of performance oriented research, although in my group, we go all the
+
+38:40.320 --> 38:46.160
+ way from like idea to theory to experiments to big system building to commercialization. So we
+
+38:46.160 --> 38:52.560
+ span that spectrum. But I think that in a lot of situations in AI, you really have to build the
+
+38:52.560 --> 38:58.400
+ big systems and evaluate them at scale before you know what works and doesn't. And we've seen
+
+38:58.400 --> 39:03.280
+ that in the computational game theory community, that there are a lot of techniques that look good
+
+39:03.280 --> 39:08.640
+ in the small, but then they cease to look good in the large. And we've also seen that there are a
+
+39:08.640 --> 39:15.600
+ lot of techniques that look superior in theory. And I really mean in terms of convergence rates,
+
+39:15.600 --> 39:20.720
+ better like first order methods, better convergence rates like the CFR based algorithms,
+
+39:20.720 --> 39:26.080
+ yet the CFR based algorithms are the first fastest in practice. So it really tells me that you have
+
+39:26.080 --> 39:32.400
+ to test these in reality, the theory isn't tight enough, if you will, to tell you which
+
+39:32.400 --> 39:38.480
+ algorithms are better than the others. And you have to look at these things that in the large,
+
+39:38.480 --> 39:43.600
+ because any sort of projections you do from the small can at least in this domain be very misleading.
+
+39:43.600 --> 39:49.040
+ So that's kind of from a kind of science and engineering perspective, from a personal perspective,
+
+39:49.040 --> 39:55.200
+ it's been just a wild experience in that with the first poker competition, the first
+
+39:56.160 --> 40:00.640
+ brains versus AI man machine poker competition that we organized. There had been, by the way,
+
+40:00.640 --> 40:04.880
+ for other poker games, there had been previous competitions, but this was for heads up no limit,
+
+40:04.880 --> 40:11.200
+ this was the first. And I probably became the most hated person in the world of poker. And I
+
+40:11.200 --> 40:18.400
+ didn't mean to sigh. Why is that for cracking the game for? Yeah, it was a lot of people
+
+40:18.400 --> 40:24.160
+ felt that it was a real threat to the whole game, the whole existence of the game. If AI becomes
+
+40:24.160 --> 40:29.680
+ better than humans, people would be scared to play poker, because there are these superhuman
+
+40:29.680 --> 40:35.120
+ AIs running around taking their money and you know, all of that. So I just it was really aggressive.
+
+40:35.120 --> 40:40.640
+ Interesting. The comments were super aggressive. I got everything just short of death threats.
+
+40:42.000 --> 40:45.760
+ Do you think the same was true for chess? Because right now, they just completed the
+
+40:45.760 --> 40:50.720
+ world championships in chess, and humans just started ignoring the fact that there's AI systems
+
+40:50.720 --> 40:55.360
+ now that outperform humans and they still enjoy the game is still a beautiful game.
+
+40:55.360 --> 41:00.160
+ That's what I think. And I think the same thing happened in poker. And so I didn't
+
+41:00.160 --> 41:03.760
+ think of myself as somebody was going to kill the game. And I don't think I did.
+
+41:03.760 --> 41:07.360
+ I've really learned to love this game. I wasn't a poker player before, but
+
+41:07.360 --> 41:12.400
+ learn so many nuances about it from these AIs. And they've really changed how the game is played,
+
+41:12.400 --> 41:17.600
+ by the way. So they have these very Martian ways of playing poker. And the top humans are now
+
+41:17.600 --> 41:24.880
+ incorporating those types of strategies into their own play. So if anything, to me, our work has made
+
+41:25.600 --> 41:31.760
+ poker a richer, more interesting game for humans to play, not something that is going to steer
+
+41:31.760 --> 41:36.240
+ humans away from it entirely. Just a quick comment on something you said, which is,
+
+41:37.440 --> 41:44.800
+ if I may say so, in academia is a little bit rare sometimes. It's pretty brave to put your ideas
+
+41:44.800 --> 41:50.000
+ to the test in the way you described, saying that sometimes good ideas don't work when you actually
+
+41:50.560 --> 41:57.600
+ try to apply them at scale. So where does that come from? I mean, if you could do advice for
+
+41:57.600 --> 42:04.000
+ people, what drives you in that sense? Were you always this way? I mean, it takes a brave person,
+
+42:04.000 --> 42:09.280
+ I guess is what I'm saying, to test their ideas and to see if this thing actually works against human
+
+42:09.840 --> 42:14.000
+ top human players and so on. I don't know about brave, but it takes a lot of work.
+
+42:14.800 --> 42:20.960
+ It takes a lot of work and a lot of time to organize, to make something big and to organize
+
+42:20.960 --> 42:26.080
+ an event and stuff like that. And what drives you in that effort? Because you could still,
+
+42:26.080 --> 42:31.280
+ I would argue, get a Best Paper Award at NIPS as you did in 17 without doing this.
+
+42:31.280 --> 42:32.160
+ That's right, yes.
+
+42:34.400 --> 42:40.960
+ So in general, I believe it's very important to do things in the real world and at scale.
+
+42:41.600 --> 42:48.320
+ And that's really where the pudding, if you will, proves in the pudding. That's where it is.
+
+42:48.320 --> 42:54.720
+ In this particular case, it was kind of a competition between different groups.
+
+42:54.720 --> 43:00.400
+ And for many years, as to who can be the first one to beat the top humans at heads up,
+
+43:00.400 --> 43:09.440
+ no limit takes us hold them. So it became kind of like a competition who can get there.
+
+43:09.440 --> 43:13.840
+ Yeah, so a little friendly competition could do wonders for progress.
+
+43:13.840 --> 43:14.960
+ Yes, absolutely.
+
+43:16.240 --> 43:21.440
+ So the topic of mechanism design, which is really interesting, also kind of new to me,
+
+43:21.440 --> 43:27.440
+ except as an observer of, I don't know, politics and any, I'm an observer of mechanisms,
+
+43:27.440 --> 43:32.960
+ but you write in your paper, an automated mechanism design that I quickly read.
+
+43:33.840 --> 43:36.960
+ So mechanism design is designing the rules of the game,
+
+43:37.760 --> 43:44.400
+ so you get a certain desirable outcome. And you have this work on doing so in an automatic
+
+43:44.400 --> 43:49.440
+ fashion as opposed to fine tuning it. So what have you learned from those efforts?
+
+43:49.440 --> 43:57.040
+ If you look, say, I don't know, at complex, it's like our political system. Can we design our
+
+43:57.040 --> 44:04.480
+ political system to have in an automated fashion, to have outcomes that we want? Can we design
+
+44:04.480 --> 44:11.680
+ something like traffic lights to be smart, where it gets outcomes that we want?
+
+44:11.680 --> 44:14.880
+ So what are the lessons that you draw from that work?
+
+44:14.880 --> 44:19.280
+ Yeah, so I still very much believe in the automated mechanism design direction.
+
+44:19.280 --> 44:19.520
+ Yes.
+
+44:20.640 --> 44:26.400
+ But it's not a panacea. There are impossibility results in mechanism design,
+
+44:26.400 --> 44:32.800
+ saying that there is no mechanism that accomplishes objective X in class C.
+
+44:33.840 --> 44:40.000
+ So it's not going up, there's no way, using any mechanism design tools, manual or automated,
+
+44:40.800 --> 44:42.720
+ to do certain things in mechanism design.
+
+44:42.720 --> 44:46.960
+ Can you describe that again? So meaning there, it's impossible to achieve that?
+
+44:46.960 --> 44:55.120
+ Yeah, there's also an impossible. So these are not statements about human ingenuity,
+
+44:55.120 --> 45:00.320
+ who might come up with something smart. These are proofs that if you want to accomplish properties
+
+45:00.320 --> 45:06.080
+ X in class C, that is not doable with any mechanism. The good thing about automated
+
+45:06.080 --> 45:12.800
+ mechanism design is that we're not really designing for a class, we're designing for specific settings
+
+45:12.800 --> 45:18.800
+ at a time. So even if there's an impossibility result for the whole class, it just doesn't
+
+45:18.800 --> 45:23.520
+ mean that all of the cases in the class are impossible, it just means that some of the
+
+45:23.520 --> 45:28.800
+ cases are impossible. So we can actually carve these islands of possibility within these
+
+45:28.800 --> 45:34.640
+ non impossible classes. And we've actually done that. So one of the famous results in mechanism
+
+45:34.640 --> 45:40.640
+ design is a Meyers and Settled Weight theorem by Roger Meyers and Mark Settled Weight from 1983.
+
+45:40.640 --> 45:45.760
+ So it's an impossibility of efficient trade under imperfect information. We show that
+
+45:46.880 --> 45:50.480
+ you can in many settings avoid that and get efficient trade anyway.
+
+45:51.360 --> 45:56.640
+ Depending on how you design the game. Depending how you design the game. And of course,
+
+45:56.640 --> 46:03.760
+ it doesn't in any way contradict the impossibility result. The impossibility result is still there,
+
+46:03.760 --> 46:11.200
+ but it just finds spots within this impossible class where in those spots you don't have the
+
+46:11.200 --> 46:17.600
+ impossibility. Sorry if I'm going a bit philosophical, but what lessons do you draw towards like I
+
+46:17.600 --> 46:23.920
+ mentioned politics or the human interaction and designing mechanisms for outside of just
+
+46:24.800 --> 46:27.120
+ these kinds of trading or auctioning or
+
+46:27.120 --> 46:37.200
+ purely formal games or human interaction like a political system. Do you think it's applicable
+
+46:37.200 --> 46:47.920
+ to politics or to business, to negotiations, these kinds of things, designing rules that have
+
+46:47.920 --> 46:53.360
+ certain outcomes? Yeah, yeah, I do think so. Have you seen success that successfully done?
+
+46:53.360 --> 46:58.880
+ There hasn't really. Oh, you mean mechanism design or automated mechanism? Automated mechanism design.
+
+46:58.880 --> 47:07.440
+ So mechanism design itself has had fairly limited success so far. There are certain cases,
+
+47:07.440 --> 47:13.600
+ but most of the real world situations are actually not sound from a mechanism design
+
+47:13.600 --> 47:18.640
+ perspective. Even in those cases where they've been designed by very knowledgeable mechanism
+
+47:18.640 --> 47:24.080
+ design people, the people are typically just taking some insights from the theory and applying
+
+47:24.080 --> 47:29.920
+ those insights into the real world rather than applying the mechanisms directly. So one famous
+
+47:29.920 --> 47:36.880
+ example of is the FCC spectrum auctions. So I've also had a small role in that and
+
+47:38.560 --> 47:45.040
+ very good economists have been working on that with no game theory. Yet the rules that are
+
+47:45.040 --> 47:50.960
+ designed in practice there, they're such that bidding truthfully is not the best strategy.
+
+47:51.680 --> 47:57.040
+ Usually mechanism design, we try to make things easy for the participants. So telling the truth
+
+47:57.040 --> 48:02.480
+ is the best strategy. But even in those very high stakes auctions where you have tens of billions
+
+48:02.480 --> 48:07.840
+ of dollars worth of spectrum being auctioned, truth telling is not the best strategy.
+
+48:09.200 --> 48:14.000
+ And by the way, nobody knows even a single optimal bidding strategy for those auctions.
+
+48:14.000 --> 48:17.600
+ What's the challenge of coming up with an optimal bid? Because there's a lot of players and there's
+
+48:17.600 --> 48:23.440
+ imperfections. It's not so much there, a lot of players, but many items for sale. And these
+
+48:23.440 --> 48:29.200
+ mechanisms are such that even with just two items or one item, bidding truthfully wouldn't be
+
+48:29.200 --> 48:37.040
+ the best strategy. If you look at the history of AI, it's marked by seminal events and an
+
+48:37.040 --> 48:42.160
+ AlphaGo beating a world champion, human go player, I would put Lebrados winning the heads
+
+48:42.160 --> 48:51.280
+ up no limit hold them as one of such event. Thank you. And what do you think is the next such event?
+
+48:52.400 --> 48:58.880
+ Whether it's in your life or in the broadly AI community that you think might be out there
+
+48:58.880 --> 49:04.800
+ that would surprise the world. So that's a great question and I really know the answer. In terms
+
+49:04.800 --> 49:12.880
+ of game solving heads up no limit takes us all and really was the one remaining widely agreed
+
+49:12.880 --> 49:18.400
+ upon benchmark. So that was the big milestone. Now, are there other things? Yeah, certainly
+
+49:18.400 --> 49:23.440
+ there are, but there there is not one that the community has kind of focused on. So what could
+
+49:23.440 --> 49:29.680
+ be other things? There are groups working on Starcraft. There are groups working on Dota 2.
+
+49:29.680 --> 49:36.560
+ These are video games. Yes, or you could have like diplomacy or Hanabi, you know, things like
+
+49:36.560 --> 49:42.640
+ that. These are like recreational games, but none of them are really acknowledged as kind of the
+
+49:42.640 --> 49:49.920
+ main next challenge problem. Like chess or go or heads up no limit takes us hold them was.
+
+49:49.920 --> 49:55.600
+ So I don't really know in the game solving space what is or what will be the next benchmark. I
+
+49:55.600 --> 49:59.920
+ kind of hope that there will be a next benchmark because really the different groups working on
+
+49:59.920 --> 50:06.160
+ the same problem really drove these application independent techniques forward very quickly
+
+50:06.160 --> 50:11.120
+ over 10 years. Do you think there's an open problem that excites you that you start moving
+
+50:11.120 --> 50:18.240
+ away from games into real world games like say the stock market trading? Yeah, so that's kind
+
+50:18.240 --> 50:27.120
+ of how I am. So I am probably not going to work as hard on these recreational benchmarks.
+
+50:27.760 --> 50:32.960
+ I'm doing two startups on game solving technology strategic machine and strategy robot and we're
+
+50:32.960 --> 50:39.680
+ really interested in pushing this stuff into practice. What do you think would be really,
+
+50:39.680 --> 50:50.400
+ you know, a powerful result that would be surprising that would be if you can say, I mean,
+
+50:50.400 --> 50:56.880
+ it's, you know, five years, 10 years from now, something that statistically would say is not
+
+50:56.880 --> 51:03.200
+ very likely, but if there's a breakthrough would achieve. Yeah, so I think that overall,
+
+51:03.200 --> 51:11.600
+ we're in a very different situation in game theory than we are in, let's say, machine learning. Yes.
+
+51:11.600 --> 51:16.720
+ So in machine learning, it's a fairly mature technology and it's very broadly applied and
+
+51:17.280 --> 51:22.480
+ proven success in the real world. In game solving, there are almost no applications yet.
+
+51:24.320 --> 51:29.440
+ We have just become superhuman, which machine learning you could argue happened in the 90s,
+
+51:29.440 --> 51:35.120
+ if not earlier, and at least on supervised learning, certain complex supervised learning
+
+51:35.120 --> 51:40.960
+ applications. Now, I think the next challenge problem, I know you're not asking about it this
+
+51:40.960 --> 51:45.360
+ way, you're asking about technology breakthrough. But I think that big breakthrough is to be able
+
+51:45.360 --> 51:50.720
+ to show that, hey, maybe most of, let's say, military planning or most of business strategy
+
+51:50.720 --> 51:55.600
+ will actually be done strategically using computational game theory. That's what I would
+
+51:55.600 --> 52:01.120
+ like to see as a next five or 10 year goal. Maybe you can explain to me again, forgive me if this
+
+52:01.120 --> 52:07.360
+ is an obvious question, but machine learning methods and neural networks suffer from not
+
+52:07.360 --> 52:12.800
+ being transparent, not being explainable. Game theoretic methods, Nash Equilibria,
+
+52:12.800 --> 52:18.960
+ do they generally, when you see the different solutions, are they, when you talk about military
+
+52:18.960 --> 52:24.640
+ operations, are they, once you see the strategies, do they make sense? Are they explainable or do
+
+52:24.640 --> 52:29.680
+ they suffer from the same problems as neural networks do? So that's a good question. I would say
+
+52:30.400 --> 52:36.560
+ a little bit yes and no. And what I mean by that is that these game theoretic strategies,
+
+52:36.560 --> 52:42.320
+ let's say, Nash Equilibrium, it has provable properties. So it's unlike, let's say, deep
+
+52:42.320 --> 52:47.040
+ learning where you kind of cross your fingers, hopefully it'll work. And then after the fact,
+
+52:47.040 --> 52:52.560
+ when you have the weights, you're still crossing your fingers, and I hope it will work. Here,
+
+52:52.560 --> 52:57.840
+ you know that the solution quality is there. There's provable solution quality guarantees.
+
+52:58.480 --> 53:03.360
+ Now, that doesn't necessarily mean that the strategies are human understandable.
+
+53:03.360 --> 53:08.560
+ That's a whole other problem. So I think that deep learning and computational game theory
+
+53:08.560 --> 53:12.320
+ are in the same boat in that sense, that both are difficult to understand.
+
+53:13.680 --> 53:16.160
+ But at least the game theoretic techniques, they have this
+
+53:16.160 --> 53:22.800
+ guarantees of solution quality. So do you see business operations, strategic operations,
+
+53:22.800 --> 53:31.440
+ even military in the future being at least the strong candidates being proposed by automated
+
+53:31.440 --> 53:39.520
+ systems? Do you see that? Yeah, I do. I do. But that's more of a belief than a substantiated fact.
+
+53:39.520 --> 53:43.840
+ Depending on where you land, an optimism or pessimism, that's a really, to me, that's an
+
+53:43.840 --> 53:52.560
+ exciting future, especially if there's provable things in terms of optimality. So looking into
+
+53:52.560 --> 54:01.120
+ the future, there's a few folks worried about the, especially you look at the game of poker,
+
+54:01.120 --> 54:06.800
+ which is probably one of the last benchmarks in terms of games being solved. They worry about
+
+54:06.800 --> 54:11.760
+ the future and the existential threats of artificial intelligence. So the negative impact
+
+54:11.760 --> 54:18.800
+ in whatever form on society, is that something that concerns you as much? Or are you more optimistic
+
+54:18.800 --> 54:24.640
+ about the positive impacts of AI? I am much more optimistic about the positive impacts.
+
+54:24.640 --> 54:29.920
+ So just in my own work, what we've done so far, we run the nationwide kidney exchange.
+
+54:29.920 --> 54:36.000
+ Hundreds of people are walking around alive today, who would it be? And it's increased employment.
+
+54:36.000 --> 54:42.720
+ You have a lot of people now running kidney exchanges and at transplant centers, interacting
+
+54:42.720 --> 54:50.240
+ with the kidney exchange. You have some extra surgeons, nurses, anesthesiologists, hospitals,
+
+54:50.240 --> 54:55.200
+ all of that. So employment is increasing from that and the world is becoming a better place.
+
+54:55.200 --> 55:03.120
+ Another example is combinatorial sourcing auctions. We did 800 large scale combinatorial
+
+55:03.120 --> 55:09.280
+ sourcing auctions from 2001 to 2010 in a previous startup of mine called Combinet.
+
+55:09.280 --> 55:18.480
+ And we increased the supply chain efficiency on that $60 billion of spend by 12.6%. So that's
+
+55:18.480 --> 55:24.000
+ over $6 billion of efficiency improvement in the world. And this is not like shifting value from
+
+55:24.000 --> 55:28.960
+ somebody to somebody else, just efficiency improvement, like in trucking, less empty
+
+55:28.960 --> 55:35.040
+ driving. So there's less waste, less carbon footprint and so on. This is a huge positive
+
+55:35.040 --> 55:42.080
+ impact in the near term. But sort of to stay in it for a little longer, because I think game theory
+
+55:42.080 --> 55:46.720
+ is a role to play here. Let me actually come back on that. That's one thing. I think AI is also going
+
+55:46.720 --> 55:52.960
+ to make the world much safer. So that's another aspect that often gets overlooked.
+
+55:53.920 --> 55:58.560
+ Well, let me ask this question. Maybe you can speak to the safer. So I talked to Max Tagmark
+
+55:58.560 --> 56:03.200
+ and Stuart Russell, who are very concerned about existential threats of AI. And often,
+
+56:03.200 --> 56:12.960
+ the concern is about value misalignment. So AI systems basically working, operating towards
+
+56:12.960 --> 56:19.680
+ goals that are not the same as human civilization, human beings. So it seems like game theory has
+
+56:19.680 --> 56:28.800
+ a role to play there to make sure the values are aligned with human beings. I don't know if that's
+
+56:28.800 --> 56:36.160
+ how you think about it. If not, how do you think AI might help with this problem? How do you think
+
+56:36.160 --> 56:44.800
+ AI might make the world safer? Yeah, I think this value misalignment is a fairly theoretical
+
+56:44.800 --> 56:52.880
+ worry. And I haven't really seen it in because I do a lot of real applications. I don't see it
+
+56:52.880 --> 56:58.080
+ anywhere. The closest I've seen it was the following type of mental exercise, really,
+
+56:58.080 --> 57:03.200
+ where I had this argument in the late 80s when we were building these transportation optimization
+
+57:03.200 --> 57:08.160
+ systems. And somebody had heard that it's a good idea to have high utilization of assets.
+
+57:08.160 --> 57:13.840
+ So they told me that, hey, why don't you put that as objective? And we didn't even put it as an
+
+57:13.840 --> 57:19.360
+ objective, because I just showed him that if you had that as your objective, the solution would be
+
+57:19.360 --> 57:23.360
+ to load your trucks full and drive in circles. Nothing would ever get delivered. You'd have
+
+57:23.360 --> 57:30.480
+ 100% utilization. So yeah, I know this phenomenon. I've known this for over 30 years. But I've never
+
+57:30.480 --> 57:35.920
+ seen it actually be a problem in reality. And yes, if you have the wrong objective, the AI will
+
+57:35.920 --> 57:40.720
+ optimize that to the hilt. And it's going to hurt more than some human who's kind of trying to
+
+57:40.720 --> 57:47.200
+ solve it in a half baked way with some human insight, too. But I just haven't seen that
+
+57:47.200 --> 57:51.600
+ materialize in practice. There's this gap that you've actually put your finger on
+
+57:52.880 --> 57:59.760
+ very clearly just now between theory and reality that's very difficult to put into words, I think.
+
+57:59.760 --> 58:06.720
+ It's what you can theoretically imagine, the worst possible case or even, yeah, I mean,
+
+58:06.720 --> 58:12.720
+ bad cases. And what usually happens in reality. So for example, to me, maybe it's something you
+
+58:12.720 --> 58:19.680
+ can comment on, having grown up and I grew up in the Soviet Union. You know, there's currently
+
+58:19.680 --> 58:28.320
+ 10,000 nuclear weapons in the world. And for many decades, it's theoretically surprising to me
+
+58:28.320 --> 58:34.240
+ that the nuclear war is not broken out. Do you think about this aspect from a game
+
+58:34.240 --> 58:41.520
+ theoretic perspective in general? Why is that true? Why? In theory, you could see how things
+
+58:41.520 --> 58:46.480
+ go terribly wrong. And somehow yet they have not. Yeah, how do you think so? So I do think that
+
+58:46.480 --> 58:51.120
+ about that a lot. I think the biggest two threats that we're facing as mankind, one is climate
+
+58:51.120 --> 58:57.200
+ change, and the other is nuclear war. So so those are my main two worries that they worry about.
+
+58:57.200 --> 59:01.840
+ And I've tried to do something about climate thought about trying to do something for climate
+
+59:01.840 --> 59:07.920
+ change twice. Actually, for two of my startups, I've actually commissioned studies of what we
+
+59:07.920 --> 59:12.320
+ could do on those things. And we didn't really find a sweet spot, but I'm still keeping an eye out
+
+59:12.320 --> 59:17.280
+ on that if there's something where we could actually provide a market solution or optimization
+
+59:17.280 --> 59:24.080
+ solution or some other technology solution to problems. Right now, like for example, pollution
+
+59:24.080 --> 59:30.000
+ critic markets was what we were looking at then. And it was much more the lack of political will
+
+59:30.000 --> 59:35.520
+ by those markets were not so successful, rather than bad market design. So I could go in and
+
+59:35.520 --> 59:39.920
+ make a better market design. But that wouldn't really move the needle on the world very much
+
+59:39.920 --> 59:44.800
+ if there's no political will and in the US, you know, the market, at least the Chicago market
+
+59:44.800 --> 59:50.560
+ was just shut down, and so on. So then it doesn't really help how great your market design was.
+
+59:50.560 --> 59:59.920
+ And on the nuclear side, it's more so global warming is a more encroaching problem.
+
+1:00:00.560 --> 1:00:05.680
+ You know, nuclear weapons have been here. It's an obvious problem has just been sitting there.
+
+1:00:05.680 --> 1:00:12.240
+ So how do you think about what is the mechanism design there that just made everything seem stable?
+
+1:00:12.240 --> 1:00:18.560
+ And are you still extremely worried? I am still extremely worried. So you probably know the simple
+
+1:00:18.560 --> 1:00:25.280
+ game theory of mad. So this was a mutually assured destruction. And it doesn't require any
+
+1:00:25.280 --> 1:00:29.600
+ computation with small matrices, you can actually convince yourself that the game is such that
+
+1:00:29.600 --> 1:00:36.000
+ nobody wants to initiate. Yeah, that's a very coarse grained analysis. And it really works in
+
+1:00:36.000 --> 1:00:40.960
+ a situation where you have two superpowers or small numbers of superpowers. Now things are
+
+1:00:40.960 --> 1:00:47.920
+ very different. You have a smaller nuke. So the threshold of initiating is smaller. And you have
+
+1:00:47.920 --> 1:00:54.800
+ smaller countries and non non nation actors who may get nukes and so on. So it's I think it's
+
+1:00:54.800 --> 1:01:04.240
+ riskier now than it was maybe ever before. And what idea application of AI, you've talked about
+
+1:01:04.240 --> 1:01:09.440
+ a little bit, but what is the most exciting to you right now? I mean, you're here at NIPS,
+
+1:01:09.440 --> 1:01:16.640
+ NewRips. Now, you have a few excellent pieces of work. But what are you thinking into the future
+
+1:01:16.640 --> 1:01:20.640
+ with several companies you're doing? What's the most exciting thing or one of the exciting things?
+
+1:01:21.200 --> 1:01:26.960
+ The number one thing for me right now is coming up with these scalable techniques for
+
+1:01:26.960 --> 1:01:32.800
+ game solving and applying them into the real world. I'm still very interested in market design
+
+1:01:32.800 --> 1:01:37.200
+ as well. And we're doing that in the optimized markets. But I'm most interested if number one
+
+1:01:37.200 --> 1:01:42.320
+ right now is strategic machine strategy robot getting that technology out there and seeing
+
+1:01:42.320 --> 1:01:47.920
+ as you're in the trenches doing applications, what needs to be actually filled, what technology
+
+1:01:47.920 --> 1:01:52.720
+ gaps still need to be filled. So it's so hard to just put your feet on the table and imagine what
+
+1:01:52.720 --> 1:01:57.440
+ needs to be done. But when you're actually doing real applications, the applications tell you
+
+1:01:58.000 --> 1:02:03.040
+ what needs to be done. And I really enjoy that interaction. Is it a challenging process to
+
+1:02:03.040 --> 1:02:13.520
+ apply some of the state of the art techniques you're working on, and having the various players in
+
+1:02:13.520 --> 1:02:19.280
+ industry or the military, or people who could really benefit from it actually use it? What's
+
+1:02:19.280 --> 1:02:23.920
+ that process like of, you know, in autonomous vehicles, we work with automotive companies and
+
+1:02:23.920 --> 1:02:31.120
+ they're in many ways are a little bit old fashioned, it's difficult. They really want to use this
+
+1:02:31.120 --> 1:02:37.520
+ technology, there's clearly will have a significant benefit. But the systems aren't quite in place
+
+1:02:37.520 --> 1:02:43.280
+ to easily have them integrated in terms of data in terms of compute in terms of all these kinds
+
+1:02:43.280 --> 1:02:49.200
+ of things. So do you, is that one of the bigger challenges that you're facing? And how do you
+
+1:02:49.200 --> 1:02:54.000
+ tackle that challenge? Yeah, I think that's always a challenge that that's kind of slowness and inertia
+
+1:02:54.000 --> 1:02:59.360
+ really of let's do things the way we've always done it. You just have to find the internal
+
+1:02:59.360 --> 1:03:04.480
+ champions that the customer who understand that hey, things can't be the same way in the future,
+
+1:03:04.480 --> 1:03:09.120
+ otherwise bad things are going to happen. And it's in autonomous vehicles, it's actually very
+
+1:03:09.120 --> 1:03:13.200
+ interesting that the car makers are doing that, and they're very traditional. But at the same time,
+
+1:03:13.200 --> 1:03:18.320
+ you have tech companies who have nothing to do with cars or transportation, like Google and Baidu,
+
+1:03:19.120 --> 1:03:25.360
+ really pushing on autonomous cars. I find that fascinating. Clearly, you're super excited about
+
+1:03:25.360 --> 1:03:31.120
+ how actually these ideas have an impact in the world. In terms of the technology, in terms of
+
+1:03:31.120 --> 1:03:38.320
+ ideas and research, are there directions that you're also excited about, whether that's on the
+
+1:03:39.440 --> 1:03:43.360
+ some of the approaches you talked about for imperfect information games, whether it's applying
+
+1:03:43.360 --> 1:03:46.640
+ deep learning to some of these problems? Is there something that you're excited in
+
+1:03:47.360 --> 1:03:52.240
+ in the research side of things? Yeah, yeah, lots of different things in the game solving.
+
+1:03:52.240 --> 1:04:00.240
+ So solving even bigger games, games where you have more hidden action of the player
+
+1:04:00.240 --> 1:04:06.560
+ actions as well. poker is a game where really chance actions are hidden, or some of them are
+
+1:04:06.560 --> 1:04:14.480
+ hidden, but the player actions are public. multiplayer games or various sorts, collusion,
+
+1:04:14.480 --> 1:04:22.960
+ opponent exploitation, all and even longer games. So games that basically go forever,
+
+1:04:22.960 --> 1:04:29.520
+ but they're not repeated. So see extensive fun games that go forever. What would that even look
+
+1:04:29.520 --> 1:04:33.200
+ like? How do you represent that? How do you solve that? What's an example of a game like that?
+
+1:04:33.920 --> 1:04:37.680
+ This is some of the stochastic games that you mentioned. Let's say business strategy. So it's
+
+1:04:37.680 --> 1:04:42.880
+ and not just modeling like a particular interaction, but thinking about the business from here to
+
+1:04:42.880 --> 1:04:50.960
+ eternity. Or I see, or let's let's say military strategy. So it's not like war is going to go away.
+
+1:04:50.960 --> 1:04:58.000
+ How do you think about military strategy that's going to go forever? How do you even model that?
+
+1:04:58.000 --> 1:05:05.760
+ How do you know whether a move was good that you use somebody made? And so on. So that that's
+
+1:05:05.760 --> 1:05:11.920
+ kind of one direction. I'm also very interested in learning much more scalable techniques for
+
+1:05:11.920 --> 1:05:18.560
+ integer programming. So we had an ICML paper this summer on that, the first automated algorithm
+
+1:05:18.560 --> 1:05:24.400
+ configuration paper that has theoretical generalization guarantees. So if I see these many
+
+1:05:24.400 --> 1:05:30.480
+ training examples, and I tool my algorithm in this way, it's going to have good performance
+
+1:05:30.480 --> 1:05:35.280
+ on the real distribution, which I've not seen. So which is kind of interesting that, you know,
+
+1:05:35.280 --> 1:05:42.560
+ algorithm configuration has been going on now for at least 17 years seriously. And there has not
+
+1:05:42.560 --> 1:05:48.800
+ been any generalization theory before. Well, this is really exciting. And it's been, it's a huge
+
+1:05:48.800 --> 1:05:52.720
+ honor to talk to you. Thank you so much, Tomas. Thank you for bringing Lebrates to the world
+
+1:05:52.720 --> 1:06:07.440
+ and all the great work you're doing. Well, thank you very much. It's been fun. Good questions.
+

vtt/episode_013_small.vtt

@@ -0,0 +1,2570 @@
+WEBVTT
+
+00:00.000 --> 00:02.960
+ The following is a conversation with Tomaso Poggio.
+
+00:02.960 --> 00:06.200
+ He's a professor at MIT and is a director of the Center
+
+00:06.200 --> 00:08.360
+ for Brains, Minds, and Machines.
+
+00:08.360 --> 00:11.640
+ Cited over 100,000 times, his work
+
+00:11.640 --> 00:14.560
+ has had a profound impact on our understanding
+
+00:14.560 --> 00:17.680
+ of the nature of intelligence in both biological
+
+00:17.680 --> 00:19.880
+ and artificial neural networks.
+
+00:19.880 --> 00:23.840
+ He has been an advisor to many highly impactful researchers
+
+00:23.840 --> 00:26.120
+ and entrepreneurs in AI, including
+
+00:26.120 --> 00:28.000
+ Demisus Habbis of DeepMind,
+
+00:28.000 --> 00:31.200
+ Amnon Shashwa of Mobileye, and Christoph Koch
+
+00:31.200 --> 00:34.120
+ of the Allen Institute for Brain Science.
+
+00:34.120 --> 00:36.400
+ This conversation is part of the MIT course
+
+00:36.400 --> 00:38.120
+ on artificial general intelligence
+
+00:38.120 --> 00:40.240
+ and the artificial intelligence podcast.
+
+00:40.240 --> 00:42.760
+ If you enjoy it, subscribe on YouTube, iTunes,
+
+00:42.760 --> 00:44.600
+ or simply connect with me on Twitter
+
+00:44.600 --> 00:47.960
+ at Lex Freedman, spelled F R I D.
+
+00:47.960 --> 00:52.480
+ And now, here's my conversation with Tomaso Poggio.
+
+00:52.480 --> 00:54.520
+ You've mentioned that in your childhood,
+
+00:54.520 --> 00:56.960
+ you've developed a fascination with physics,
+
+00:56.960 --> 00:59.720
+ especially the theory of relativity,
+
+00:59.720 --> 01:03.600
+ and that Einstein was also a childhood hero to you.
+
+01:04.520 --> 01:09.040
+ What aspect of Einstein's genius, the nature of his genius,
+
+01:09.040 --> 01:10.200
+ do you think was essential
+
+01:10.200 --> 01:12.960
+ for discovering the theory of relativity?
+
+01:12.960 --> 01:15.960
+ You know, Einstein was a hero to me,
+
+01:15.960 --> 01:17.200
+ and I'm sure to many people,
+
+01:17.200 --> 01:21.680
+ because he was able to make, of course,
+
+01:21.680 --> 01:25.200
+ a major, major contribution to physics
+
+01:25.200 --> 01:28.520
+ with simplifying a bit,
+
+01:28.520 --> 01:33.520
+ just a gedanken experiment, a thought experiment.
+
+01:35.200 --> 01:38.880
+ You know, imagining communication with lights
+
+01:38.880 --> 01:43.240
+ between a stationary observer and somebody on a train.
+
+01:43.240 --> 01:48.240
+ And I thought, you know, the fact that just
+
+01:48.560 --> 01:52.720
+ with the force of his thought, of his thinking, of his mind,
+
+01:52.720 --> 01:55.640
+ it could get to something so deep
+
+01:55.640 --> 01:57.520
+ in terms of physical reality,
+
+01:57.520 --> 02:01.320
+ how time depends on space and speed.
+
+02:01.320 --> 02:04.120
+ It was something absolutely fascinating.
+
+02:04.120 --> 02:06.720
+ It was the power of intelligence,
+
+02:06.720 --> 02:08.440
+ the power of the mind.
+
+02:08.440 --> 02:11.120
+ Do you think the ability to imagine,
+
+02:11.120 --> 02:15.200
+ to visualize as he did, as a lot of great physicists do,
+
+02:15.200 --> 02:18.640
+ do you think that's in all of us human beings,
+
+02:18.640 --> 02:20.600
+ or is there something special
+
+02:20.600 --> 02:22.880
+ to that one particular human being?
+
+02:22.880 --> 02:27.160
+ I think, you know, all of us can learn
+
+02:27.160 --> 02:32.160
+ and have, in principle, similar breakthroughs.
+
+02:33.240 --> 02:37.200
+ There is lesson to be learned from Einstein.
+
+02:37.200 --> 02:42.200
+ He was one of five PhD students at ETA,
+
+02:42.600 --> 02:47.600
+ the Eidgenossische Technische Hochschule in Zurich, in physics.
+
+02:47.600 --> 02:49.840
+ And he was the worst of the five.
+
+02:49.840 --> 02:53.600
+ The only one who did not get an academic position
+
+02:53.600 --> 02:57.040
+ when he graduated, when he finished his PhD,
+
+02:57.040 --> 03:00.000
+ and he went to work, as everybody knows,
+
+03:00.000 --> 03:01.720
+ for the patent office.
+
+03:01.720 --> 03:05.000
+ So it's not so much that he worked for the patent office,
+
+03:05.000 --> 03:07.880
+ but the fact that obviously he was smart,
+
+03:07.880 --> 03:10.240
+ but he was not the top student,
+
+03:10.240 --> 03:12.640
+ obviously he was the anti conformist.
+
+03:12.640 --> 03:15.720
+ He was not thinking in the traditional way
+
+03:15.720 --> 03:18.760
+ that probably teachers and the other students were doing.
+
+03:18.760 --> 03:23.760
+ So there is a lot to be said about trying to do the opposite
+
+03:25.960 --> 03:29.800
+ or something quite different from what other people are doing.
+
+03:29.800 --> 03:31.840
+ That's certainly true for the stock market.
+
+03:31.840 --> 03:34.800
+ Never buy if everybody's buying it.
+
+03:35.800 --> 03:37.440
+ And also true for science.
+
+03:37.440 --> 03:38.440
+ Yes.
+
+03:38.440 --> 03:42.440
+ So you've also mentioned staying on the theme of physics
+
+03:42.440 --> 03:46.440
+ that you were excited at a young age
+
+03:46.440 --> 03:50.440
+ by the mysteries of the universe that physics could uncover.
+
+03:50.440 --> 03:54.440
+ Such, as I saw mentioned, the possibility of time travel.
+
+03:56.440 --> 03:59.440
+ So out of the box question I think I'll get to ask today,
+
+03:59.440 --> 04:01.440
+ do you think time travel is possible?
+
+04:02.440 --> 04:05.440
+ Well, it would be nice if it were possible right now.
+
+04:05.440 --> 04:11.440
+ In science you never say no.
+
+04:11.440 --> 04:14.440
+ But your understanding of the nature of time.
+
+04:14.440 --> 04:15.440
+ Yeah.
+
+04:15.440 --> 04:20.440
+ It's very likely that it's not possible to travel in time.
+
+04:20.440 --> 04:24.440
+ We may be able to travel forward in time.
+
+04:24.440 --> 04:28.440
+ If we can, for instance, freeze ourselves
+
+04:28.440 --> 04:34.440
+ or go on some spacecraft traveling close to the speed of light,
+
+04:34.440 --> 04:39.440
+ but in terms of actively traveling, for instance, back in time,
+
+04:39.440 --> 04:43.440
+ I find probably very unlikely.
+
+04:43.440 --> 04:49.440
+ So do you still hold the underlying dream of the engineering intelligence
+
+04:49.440 --> 04:54.440
+ that will build systems that are able to do such huge leaps
+
+04:54.440 --> 04:58.440
+ like discovering the kind of mechanism
+
+04:58.440 --> 05:00.440
+ that would be required to travel through time?
+
+05:00.440 --> 05:02.440
+ Do you still hold that dream?
+
+05:02.440 --> 05:05.440
+ Or echoes of it from your childhood?
+
+05:05.440 --> 05:06.440
+ Yeah.
+
+05:06.440 --> 05:10.440
+ I don't think there are certain problems
+
+05:10.440 --> 05:13.440
+ that probably cannot be solved,
+
+05:13.440 --> 05:17.440
+ depending on what you believe about the physical reality.
+
+05:17.440 --> 05:23.440
+ Maybe it's totally impossible to create energy from nothing
+
+05:23.440 --> 05:26.440
+ or to travel back in time.
+
+05:26.440 --> 05:35.440
+ But about making machines that can think as well as we do or better,
+
+05:35.440 --> 05:39.440
+ or more likely, especially in the short and mid term,
+
+05:39.440 --> 05:41.440
+ help us think better,
+
+05:41.440 --> 05:45.440
+ which in a sense is happening already with the computers we have,
+
+05:45.440 --> 05:47.440
+ and it will happen more and more.
+
+05:47.440 --> 05:49.440
+ But that I certainly believe,
+
+05:49.440 --> 05:53.440
+ and I don't see in principle why computers at some point
+
+05:53.440 --> 05:59.440
+ could not become more intelligent than we are,
+
+05:59.440 --> 06:03.440
+ although the word intelligence is a tricky one,
+
+06:03.440 --> 06:07.440
+ and one who should discuss what I mean with that.
+
+06:07.440 --> 06:12.440
+ Intelligence, consciousness, words like love,
+
+06:12.440 --> 06:16.440
+ all these need to be disentangled.
+
+06:16.440 --> 06:20.440
+ So you've mentioned also that you believe the problem of intelligence
+
+06:20.440 --> 06:23.440
+ is the greatest problem in science,
+
+06:23.440 --> 06:26.440
+ greater than the origin of life and the origin of the universe.
+
+06:26.440 --> 06:29.440
+ You've also, in the talk,
+
+06:29.440 --> 06:34.440
+ I've said that you're open to arguments against you.
+
+06:34.440 --> 06:40.440
+ So what do you think is the most captivating aspect
+
+06:40.440 --> 06:43.440
+ of this problem of understanding the nature of intelligence?
+
+06:43.440 --> 06:46.440
+ Why does it captivate you as it does?
+
+06:46.440 --> 06:54.440
+ Well, originally, I think one of the motivations that I had as a teenager,
+
+06:54.440 --> 06:58.440
+ when I was infatuated with the theory of relativity,
+
+06:58.440 --> 07:05.440
+ was really that I found that there was the problem of time and space
+
+07:05.440 --> 07:07.440
+ and general relativity,
+
+07:07.440 --> 07:12.440
+ but there were so many other problems of the same level of difficulty
+
+07:12.440 --> 07:16.440
+ and importance that I could, even if I were Einstein,
+
+07:16.440 --> 07:19.440
+ it was difficult to hope to solve all of them.
+
+07:19.440 --> 07:26.440
+ So what about solving a problem whose solution allowed me to solve all the problems?
+
+07:26.440 --> 07:32.440
+ And this was what if we could find the key to an intelligence
+
+07:32.440 --> 07:36.440
+ ten times better or faster than Einstein?
+
+07:36.440 --> 07:39.440
+ So that's sort of seeing artificial intelligence
+
+07:39.440 --> 07:42.440
+ as a tool to expand our capabilities.
+
+07:42.440 --> 07:47.440
+ But is there just an inherent curiosity in you
+
+07:47.440 --> 07:53.440
+ and just understanding what it is in here that makes it all work?
+
+07:53.440 --> 07:55.440
+ Yes, absolutely. You're right.
+
+07:55.440 --> 08:00.440
+ So I started saying this was the motivation when I was a teenager,
+
+08:00.440 --> 08:06.440
+ but soon after, I think the problem of human intelligence
+
+08:06.440 --> 08:14.440
+ became a real focus of my science and my research,
+
+08:14.440 --> 08:27.440
+ because I think for me the most interesting problem is really asking who we are.
+
+08:27.440 --> 08:31.440
+ It is asking not only a question about science,
+
+08:31.440 --> 08:37.440
+ but even about the very tool we are using to do science, which is our brain.
+
+08:37.440 --> 08:39.440
+ How does our brain work?
+
+08:39.440 --> 08:41.440
+ From where does it come from?
+
+08:41.440 --> 08:43.440
+ What are its limitations?
+
+08:43.440 --> 08:45.440
+ Can we make it better?
+
+08:45.440 --> 08:49.440
+ And that in many ways is the ultimate question
+
+08:49.440 --> 08:53.440
+ that underlies this whole effort of science.
+
+08:53.440 --> 08:58.440
+ So you've made significant contributions in both the science of intelligence
+
+08:58.440 --> 09:01.440
+ and the engineering of intelligence.
+
+09:01.440 --> 09:04.440
+ In a hypothetical way, let me ask,
+
+09:04.440 --> 09:08.440
+ how far do you think we can get in creating intelligence systems
+
+09:08.440 --> 09:11.440
+ without understanding the biological,
+
+09:11.440 --> 09:15.440
+ the understanding how the human brain creates intelligence?
+
+09:15.440 --> 09:18.440
+ Put another way, do you think we can build a strong ass system
+
+09:18.440 --> 09:24.440
+ without really getting at the core, understanding the functional nature of the brain?
+
+09:24.440 --> 09:28.440
+ Well, this is a real difficult question.
+
+09:28.440 --> 09:34.440
+ We did solve problems like flying
+
+09:34.440 --> 09:43.440
+ without really using too much our knowledge about how birds fly.
+
+09:43.440 --> 09:51.440
+ It was important, I guess, to know that you could have things heavier than air
+
+09:51.440 --> 09:55.440
+ being able to fly like birds.
+
+09:55.440 --> 10:00.440
+ But beyond that, probably we did not learn very much.
+
+10:00.440 --> 10:08.440
+ The brothers right did learn a lot of observation about birds
+
+10:08.440 --> 10:12.440
+ and designing their aircraft,
+
+10:12.440 --> 10:17.440
+ but you can argue we did not use much of biology in that particular case.
+
+10:17.440 --> 10:28.440
+ Now, in the case of intelligence, I think that it's a bit of a bet right now.
+
+10:28.440 --> 10:36.440
+ If you ask, okay, we all agree we'll get at some point, maybe soon,
+
+10:36.440 --> 10:42.440
+ maybe later, to a machine that is indistinguishable from my secretary
+
+10:42.440 --> 10:47.440
+ in terms of what I can ask the machine to do.
+
+10:47.440 --> 10:50.440
+ I think we'll get there and now the question is,
+
+10:50.440 --> 10:56.440
+ you can ask people, do you think we'll get there without any knowledge about the human brain
+
+10:56.440 --> 11:02.440
+ or the best way to get there is to understand better the human brain?
+
+11:02.440 --> 11:08.440
+ This is, I think, an educated bet that different people with different backgrounds
+
+11:08.440 --> 11:11.440
+ will decide in different ways.
+
+11:11.440 --> 11:17.440
+ The recent history of the progress in AI in the last, I would say, five years
+
+11:17.440 --> 11:26.440
+ or ten years has been that the main breakthroughs, the main recent breakthroughs,
+
+11:26.440 --> 11:31.440
+ really start from neuroscience.
+
+11:31.440 --> 11:35.440
+ I can mention reinforcement learning as one,
+
+11:35.440 --> 11:41.440
+ is one of the algorithms at the core of AlphaGo,
+
+11:41.440 --> 11:46.440
+ which is the system that beat the kind of an official world champion of Go,
+
+11:46.440 --> 11:52.440
+ Lee Siddle, two, three years ago in Seoul.
+
+11:52.440 --> 12:00.440
+ That's one, and that started really with the work of Pavlov in 1900,
+
+12:00.440 --> 12:07.440
+ Marvin Miski in the 60s and many other neuroscientists later on.
+
+12:07.440 --> 12:13.440
+ And deep learning started, which is the core again of AlphaGo
+
+12:13.440 --> 12:19.440
+ and systems like autonomous driving systems for cars,
+
+12:19.440 --> 12:25.440
+ like the systems that Mobileye, which is a company started by one of my ex,
+
+12:25.440 --> 12:30.440
+ Okamnon Shashua, so that is the core of those things.
+
+12:30.440 --> 12:35.440
+ And deep learning, really the initial ideas in terms of the architecture
+
+12:35.440 --> 12:42.440
+ of these layered hierarchical networks started with work of Thorston Wiesel
+
+12:42.440 --> 12:47.440
+ and David Hubel at Harvard up the river in the 60s.
+
+12:47.440 --> 12:54.440
+ So recent history suggests that neuroscience played a big role in these breakthroughs.
+
+12:54.440 --> 12:59.440
+ My personal bet is that there is a good chance they continue to play a big role,
+
+12:59.440 --> 13:03.440
+ maybe not in all the future breakthroughs, but in some of them.
+
+13:03.440 --> 13:05.440
+ At least in inspiration.
+
+13:05.440 --> 13:07.440
+ At least in inspiration, absolutely, yes.
+
+13:07.440 --> 13:12.440
+ So you studied both artificial and biological neural networks,
+
+13:12.440 --> 13:19.440
+ you said these mechanisms that underlie deep learning and reinforcement learning,
+
+13:19.440 --> 13:25.440
+ but there is nevertheless significant differences between biological and artificial neural networks
+
+13:25.440 --> 13:27.440
+ as they stand now.
+
+13:27.440 --> 13:32.440
+ So between the two, what do you find is the most interesting, mysterious,
+
+13:32.440 --> 13:37.440
+ maybe even beautiful difference as it currently stands in our understanding?
+
+13:37.440 --> 13:44.440
+ I must confess that until recently I found that the artificial networks
+
+13:44.440 --> 13:49.440
+ were too simplistic relative to real neural networks.
+
+13:49.440 --> 13:54.440
+ But, you know, recently I've been started to think that, yes,
+
+13:54.440 --> 13:59.440
+ there are very big simplification of what you find in the brain.
+
+13:59.440 --> 14:07.440
+ But on the other hand, they are much closer in terms of the architecture to the brain
+
+14:07.440 --> 14:13.440
+ than other models that we had, that computer science used as model of thinking,
+
+14:13.440 --> 14:19.440
+ or mathematical logics, you know, LISP, Prologue, and those kind of things.
+
+14:19.440 --> 14:23.440
+ So in comparison to those, they're much closer to the brain.
+
+14:23.440 --> 14:28.440
+ You have networks of neurons, which is what the brain is about.
+
+14:28.440 --> 14:35.440
+ The artificial neurons in the models are, as I said, caricature of the biological neurons,
+
+14:35.440 --> 14:39.440
+ but they're still neurons, single units communicating with other units,
+
+14:39.440 --> 14:50.440
+ something that is absent in the traditional computer type models of mathematics, reasoning, and so on.
+
+14:50.440 --> 14:56.440
+ So what aspect would you like to see in artificial neural networks added over time
+
+14:56.440 --> 14:59.440
+ as we try to figure out ways to improve them?
+
+14:59.440 --> 15:10.440
+ So one of the main differences and, you know, problems in terms of deep learning today,
+
+15:10.440 --> 15:17.440
+ and it's not only deep learning, and the brain is the need for deep learning techniques
+
+15:17.440 --> 15:22.440
+ to have a lot of labeled examples.
+
+15:22.440 --> 15:31.440
+ For instance, for ImageNet, you have a training set which is one million images, each one labeled by some human
+
+15:31.440 --> 15:34.440
+ in terms of which object is there.
+
+15:34.440 --> 15:46.440
+ And it's clear that in biology, a baby may be able to see a million images in the first years of life,
+
+15:46.440 --> 15:56.440
+ but will not have a million of labels given to him or her by parents or caretakers.
+
+15:56.440 --> 15:59.440
+ So how do you solve that?
+
+15:59.440 --> 16:07.440
+ You know, I think there is this interesting challenge that today, deep learning and related techniques
+
+16:07.440 --> 16:18.440
+ are all about big data, big data meaning a lot of examples labeled by humans,
+
+16:18.440 --> 16:22.440
+ whereas in nature you have...
+
+16:22.440 --> 16:29.440
+ So this big data is n going to infinity, that's the best, you know, n meaning labeled data.
+
+16:29.440 --> 16:34.440
+ But I think the biological world is more n going to 1.
+
+16:34.440 --> 16:42.440
+ A child can learn from a very small number of labeled examples.
+
+16:42.440 --> 16:49.440
+ Like you tell a child, this is a car, you don't need to say like in ImageNet, you know, this is a car, this is a car,
+
+16:49.440 --> 16:53.440
+ this is not a car, this is not a car, one million times.
+
+16:53.440 --> 17:05.440
+ And of course with AlphaGo and AlphaZero variants, because the world of Go is so simplistic that you can actually learn by yourself
+
+17:05.440 --> 17:08.440
+ through self play, you can play against each other.
+
+17:08.440 --> 17:15.440
+ And the real world, the visual system that you've studied extensively is a lot more complicated than the game of Go.
+
+17:15.440 --> 17:22.440
+ On the comment about children, which are fascinatingly good at learning new stuff,
+
+17:22.440 --> 17:26.440
+ how much of it do you think is hardware and how much of it is software?
+
+17:26.440 --> 17:32.440
+ Yeah, that's a good and deep question, in a sense is the old question of nurture and nature,
+
+17:32.440 --> 17:40.440
+ how much is in the gene and how much is in the experience of an individual.
+
+17:40.440 --> 17:55.440
+ Obviously, it's both that play a role and I believe that the way evolution gives put prior information, so to speak, hardwired,
+
+17:55.440 --> 18:02.440
+ it's not really hardwired, but that's essentially an hypothesis.
+
+18:02.440 --> 18:14.440
+ I think what's going on is that evolution is almost necessarily, if you believe in Darwin, it's very opportunistic.
+
+18:14.440 --> 18:23.440
+ And think about our DNA and the DNA of Drosophila.
+
+18:23.440 --> 18:28.440
+ Our DNA does not have many more genes than Drosophila.
+
+18:28.440 --> 18:32.440
+ The fly, the fruit fly.
+
+18:32.440 --> 18:39.440
+ Now, we know that the fruit fly does not learn very much during its individual existence.
+
+18:39.440 --> 18:51.440
+ It looks like one of these machinery that it's really mostly, not 100%, but 95% hardcoded by the genes.
+
+18:51.440 --> 19:02.440
+ But since we don't have many more genes than Drosophila, evolution could encode in us a kind of general learning machinery
+
+19:02.440 --> 19:09.440
+ and then had to give very weak priors.
+
+19:09.440 --> 19:20.440
+ Like, for instance, let me give a specific example, which is recent to work by a member of our Center for Brains, Mines and Machines.
+
+19:20.440 --> 19:30.440
+ We know because of work of other people in our group and other groups that there are cells in a part of our brain, neurons, that are tuned to faces.
+
+19:30.440 --> 19:33.440
+ They seem to be involved in face recognition.
+
+19:33.440 --> 19:43.440
+ Now, this face area seems to be present in young children and adults.
+
+19:43.440 --> 19:54.440
+ And one question is there from the beginning, is hardwired by evolution or somehow is learned very quickly.
+
+19:54.440 --> 20:00.440
+ So what's your, by the way, a lot of the questions I'm asking, the answer is we don't really know,
+
+20:00.440 --> 20:08.440
+ but as a person who has contributed some profound ideas in these fields, you're a good person to guess at some of these.
+
+20:08.440 --> 20:14.440
+ So, of course, there's a caveat before a lot of the stuff we talk about, but what is your hunch?
+
+20:14.440 --> 20:21.440
+ Is the face, the part of the brain that seems to be concentrated on face recognition, are you born with that?
+
+20:21.440 --> 20:26.440
+ Or are you just designed to learn that quickly, like the face of the mother and son?
+
+20:26.440 --> 20:42.440
+ My hunch, my bias was the second one, learned very quickly and turns out that Marge Livingstone at Harvard has done some amazing experiments in which she raised baby monkeys,
+
+20:42.440 --> 20:47.440
+ depriving them of faces during the first weeks of life.
+
+20:47.440 --> 20:52.440
+ So they see technicians, but the technicians have a mask.
+
+20:52.440 --> 20:54.440
+ Yes.
+
+20:54.440 --> 21:10.440
+ And so when they looked at the area in the brain of these monkeys that were usually you find faces, they found no face preference.
+
+21:10.440 --> 21:26.440
+ So my guess is that what evolution does in this case is there is a plastic area, which is plastic, which is kind of predetermined to be imprinted very easily.
+
+21:26.440 --> 21:31.440
+ But the command from the gene is not a detailed circuitry for a face template.
+
+21:31.440 --> 21:33.440
+ Could be.
+
+21:33.440 --> 21:35.440
+ But this will require probably a lot of bits.
+
+21:35.440 --> 21:39.440
+ You had to specify a lot of connection of a lot of neurons.
+
+21:39.440 --> 21:53.440
+ Instead, the command from the gene is something like imprint, memorize what you see most often in the first two weeks of life, especially in connection with food and maybe nipples.
+
+21:53.440 --> 21:54.440
+ I don't know.
+
+21:54.440 --> 21:55.440
+ Right.
+
+21:55.440 --> 21:56.440
+ Well, source of food.
+
+21:56.440 --> 22:00.440
+ And so in that area is very plastic at first and it solidifies.
+
+22:00.440 --> 22:10.440
+ It'd be interesting if a variant of that experiment would show a different kind of pattern associated with food than a face pattern, whether that could stick.
+
+22:10.440 --> 22:25.440
+ There are indications that during that experiment, what the monkeys saw quite often were the blue gloves of the technicians that were giving to the baby monkeys the milk.
+
+22:25.440 --> 22:33.440
+ And some of the cells instead of being face sensitive in that area are hand sensitive.
+
+22:33.440 --> 22:35.440
+ That's fascinating.
+
+22:35.440 --> 22:45.440
+ Can you talk about what are the different parts of the brain and in your view sort of loosely and how do they contribute to intelligence?
+
+22:45.440 --> 23:04.440
+ Do you see the brain as a bunch of different modules and they together come in the human brain to create intelligence or is it all one mush of the same kind of fundamental architecture?
+
+23:04.440 --> 23:21.440
+ Yeah, that's an important question and there was a phase in neuroscience back in the 1950s or so in which it was believed for a while that the brain was equipotential.
+
+23:21.440 --> 23:22.440
+ This was the term.
+
+23:22.440 --> 23:31.440
+ You could cut out a piece and nothing special happened apart, a little bit less performance.
+
+23:31.440 --> 23:50.440
+ There was a surgeon, Lashley, who did a lot of experiments of this type with mice and rats and concluded that every part of the brain was essentially equivalent to any other one.
+
+23:50.440 --> 24:12.440
+ It turns out that that's really not true. There are very specific modules in the brain, as you said, and people may lose the ability to speak if you have a stroke in a certain region or may lose control of their legs in another region.
+
+24:12.440 --> 24:33.440
+ So they're very specific. The brain is also quite flexible and redundant so often it can correct things and take over functions from one part of the brain to the other, but really there are specific modules.
+
+24:33.440 --> 25:02.440
+ So the answer that we know from this old work, which was basically based on lesions, either on animals or very often there was a mine of very interesting data coming from the war, from different types of injuries that soldiers had in the brain.
+
+25:02.440 --> 25:23.440
+ And more recently, functional MRI, which allow you to check which part of the brain are active when you're doing different tasks, as you can replace some of this.
+
+25:23.440 --> 25:32.440
+ You can see that certain parts of the brain are involved, are active in certain tasks.
+
+25:32.440 --> 26:01.440
+ But sort of taking a step back to that part of the brain that discovers that specializes in the face and how that might be learned, what's your intuition behind, you know, is it possible that the sort of from a physicist's perspective when you get lower and lower, that it's all the same stuff and it just, when you're born, it's plastic and it quickly figures out this part is going to be about vision, this is going to be about language, this is about common sense reasoning.
+
+26:01.440 --> 26:09.440
+ Do you have an intuition that that kind of learning is going on really quickly or is it really kind of solidified in hardware?
+
+26:09.440 --> 26:10.440
+ That's a great question.
+
+26:10.440 --> 26:21.440
+ So there are parts of the brain like the cerebellum or the hippocampus that are quite different from each other.
+
+26:21.440 --> 26:25.440
+ They clearly have different anatomy, different connectivity.
+
+26:25.440 --> 26:35.440
+ Then there is the cortex, which is the most developed part of the brain in humans.
+
+26:35.440 --> 26:47.440
+ And in the cortex, you have different regions of the cortex that are responsible for vision, for audition, for motor control, for language.
+
+26:47.440 --> 27:07.440
+ Now, one of the big puzzles of this is that in the cortex, it looks like it is the same in terms of hardware, in terms of type of neurons and connectivity across these different modalities.
+
+27:07.440 --> 27:17.440
+ So for the cortex, I think aside these other parts of the brain like spinal cord, hippocampus, cerebellum and so on.
+
+27:17.440 --> 27:28.440
+ For the cortex, I think your question about hardware and software and learning and so on, I think is rather open.
+
+27:28.440 --> 27:40.440
+ And I find it very interesting for us to think about an architecture, computer architecture that is good for vision and at the same time is good for language.
+
+27:40.440 --> 27:48.440
+ It seems to be so different problem areas that you have to solve.
+
+27:48.440 --> 27:54.440
+ But the underlying mechanism might be the same and that's really instructive for artificial neural networks.
+
+27:54.440 --> 28:00.440
+ So we've done a lot of great work in vision and human vision, computer vision.
+
+28:00.440 --> 28:07.440
+ And you mentioned the problem of human vision is really as difficult as the problem of general intelligence.
+
+28:07.440 --> 28:10.440
+ And maybe that connects to the cortex discussion.
+
+28:10.440 --> 28:21.440
+ Can you describe the human visual cortex and how the humans begin to understand the world through the raw sensory information?
+
+28:21.440 --> 28:36.440
+ What's for folks who are not familiar, especially on the computer vision side, we don't often actually take a step back except saying with a sentence or two that one is inspired by the other.
+
+28:36.440 --> 28:39.440
+ What is it that we know about the human visual cortex?
+
+28:39.440 --> 28:40.440
+ That's interesting.
+
+28:40.440 --> 28:53.440
+ So we know quite a bit at the same time, we don't know a lot, but the bit we know, in a sense, we know a lot of the details and many we don't know.
+
+28:53.440 --> 29:05.440
+ And we know a lot of the top level, the answer to the top level question, but we don't know some basic ones, even in terms of general neuroscience forgetting vision.
+
+29:05.440 --> 29:11.440
+ You know, why do we sleep? It's such a basic question.
+
+29:11.440 --> 29:14.440
+ And we really don't have an answer to that.
+
+29:14.440 --> 29:18.440
+ So taking a step back on that. So sleep, for example, is fascinating.
+
+29:18.440 --> 29:21.440
+ Do you think that's a neuroscience question?
+
+29:21.440 --> 29:30.440
+ Or if we talk about abstractions, what do you think is an interesting way to study intelligence or most effective on the levels of abstraction?
+
+29:30.440 --> 29:37.440
+ Is it chemical, is it biological, is it electrophysical, mathematical as you've done a lot of excellent work on that side?
+
+29:37.440 --> 29:42.440
+ Which psychology, sort of like at which level of abstraction do you think?
+
+29:42.440 --> 29:48.440
+ Well, in terms of levels of abstraction, I think we need all of them.
+
+29:48.440 --> 29:56.440
+ It's one, you know, it's like if you ask me, what does it mean to understand a computer?
+
+29:56.440 --> 30:04.440
+ That's much simpler. But in a computer, I could say, well, understand how to use PowerPoint.
+
+30:04.440 --> 30:13.440
+ That's my level of understanding a computer. It's, it has reasonable, you know, it gives me some power to produce slides and beautiful slides.
+
+30:13.440 --> 30:28.440
+ And now somebody else says, well, I know how the transistor work that are inside the computer can write the equation for, you know, transistor and diodes and circuits, logical circuits.
+
+30:28.440 --> 30:33.440
+ And I can ask this guy, do you know how to operate PowerPoint? No idea.
+
+30:33.440 --> 30:49.440
+ So do you think if we discovered computers walking amongst us full of these transistors that are also operating under windows and have PowerPoint, do you think it's digging in a little bit more?
+
+30:49.440 --> 31:00.440
+ How useful is it to understand the transistor in order to be able to understand PowerPoint in these higher level intelligence processes?
+
+31:00.440 --> 31:12.440
+ So I think in the case of computers, because they were made by engineers by us, these different level of understanding are rather separate on purpose.
+
+31:12.440 --> 31:23.440
+ You know, they are separate modules so that the engineer that designed the circuit for the chips does not need to know what is inside PowerPoint.
+
+31:23.440 --> 31:30.440
+ And somebody can write the software translating from one to the other.
+
+31:30.440 --> 31:40.440
+ So in that case, I don't think understanding the transistor help you understand PowerPoint or very little.
+
+31:40.440 --> 31:51.440
+ If you want to understand the computer, this question, you know, I would say you have to understanding at different levels if you really want to build one.
+
+31:51.440 --> 32:09.440
+ But for the brain, I think these levels of understanding, so the algorithms, which kind of computation, you know, the equivalent PowerPoint and the circuits, you know, the transistors, I think they are much more intertwined with each other.
+
+32:09.440 --> 32:15.440
+ There is not, you know, a neatly level of the software separate from the hardware.
+
+32:15.440 --> 32:29.440
+ And so that's why I think in the case of the brain, the problem is more difficult and more than for computers requires the interaction, the collaboration between different types of expertise.
+
+32:29.440 --> 32:35.440
+ So the brain is a big hierarchical mess that you can't just disentangle levels.
+
+32:35.440 --> 32:41.440
+ I think you can, but it's much more difficult and it's not completely obvious.
+
+32:41.440 --> 32:47.440
+ And I said, I think he's one of the person I think is the greatest problem in science.
+
+32:47.440 --> 32:52.440
+ So, you know, I think it's fair that it's difficult.
+
+32:52.440 --> 32:53.440
+ That's a difficult one.
+
+32:53.440 --> 32:58.440
+ That said, you do talk about compositionality and why it might be useful.
+
+32:58.440 --> 33:07.440
+ And when you discuss why these neural networks in artificial or biological sense learn anything, you talk about compositionality.
+
+33:07.440 --> 33:22.440
+ See, there's a sense that nature can be disentangled or well, all aspects of our cognition could be disentangled a little to some degree.
+
+33:22.440 --> 33:31.440
+ So why do you think what, first of all, how do you see compositionality and why do you think it exists at all in nature?
+
+33:31.440 --> 33:39.440
+ I spoke about, I use the term compositionality.
+
+33:39.440 --> 33:54.440
+ When we looked at deep neural networks, multi layers and trying to understand when and why they are more powerful than more classical one layer networks,
+
+33:54.440 --> 34:01.440
+ like linear classifier, kernel machines, so called.
+
+34:01.440 --> 34:12.440
+ And what we found is that in terms of approximating or learning or representing a function, a mapping from an input to an output,
+
+34:12.440 --> 34:20.440
+ like from an image to the label in the image, if this function has a particular structure,
+
+34:20.440 --> 34:28.440
+ then deep networks are much more powerful than shallow networks to approximate the underlying function.
+
+34:28.440 --> 34:33.440
+ And the particular structure is a structure of compositionality.
+
+34:33.440 --> 34:45.440
+ If the function is made up of functions of function, so that you need to look on when you are interpreting an image,
+
+34:45.440 --> 34:56.440
+ classifying an image, you don't need to look at all pixels at once, but you can compute something from small groups of pixels,
+
+34:56.440 --> 35:04.440
+ and then you can compute something on the output of this local computation and so on.
+
+35:04.440 --> 35:10.440
+ It is similar to what you do when you read a sentence, you don't need to read the first and the last letter,
+
+35:10.440 --> 35:17.440
+ but you can read syllables, combine them in words, combine the words in sentences.
+
+35:17.440 --> 35:20.440
+ So this is this kind of structure.
+
+35:20.440 --> 35:27.440
+ So that's as part of a discussion of why deep neural networks may be more effective than the shallow methods.
+
+35:27.440 --> 35:35.440
+ And is your sense for most things we can use neural networks for,
+
+35:35.440 --> 35:43.440
+ those problems are going to be compositional in nature, like language, like vision.
+
+35:43.440 --> 35:47.440
+ How far can we get in this kind of way?
+
+35:47.440 --> 35:51.440
+ So here is almost philosophy.
+
+35:51.440 --> 35:53.440
+ Well, let's go there.
+
+35:53.440 --> 35:55.440
+ Yeah, let's go there.
+
+35:55.440 --> 36:00.440
+ So friend of mine, Max Tagmark, who is a physicist at MIT.
+
+36:00.440 --> 36:02.440
+ I've talked to him on this thing.
+
+36:02.440 --> 36:04.440
+ Yeah, and he disagrees with you, right?
+
+36:04.440 --> 36:09.440
+ We agree on most, but the conclusion is a bit different.
+
+36:09.440 --> 36:14.440
+ His conclusion is that for images, for instance,
+
+36:14.440 --> 36:23.440
+ the compositional structure of this function that we have to learn or to solve these problems
+
+36:23.440 --> 36:35.440
+ comes from physics, comes from the fact that you have local interactions in physics between atoms and other atoms,
+
+36:35.440 --> 36:42.440
+ between particle of matter and other particles, between planets and other planets,
+
+36:42.440 --> 36:44.440
+ between stars and others.
+
+36:44.440 --> 36:48.440
+ It's all local.
+
+36:48.440 --> 36:55.440
+ And that's true, but you could push this argument a bit further.
+
+36:55.440 --> 36:57.440
+ Not this argument, actually.
+
+36:57.440 --> 37:02.440
+ You could argue that, you know, maybe that's part of the true,
+
+37:02.440 --> 37:06.440
+ but maybe what happens is kind of the opposite,
+
+37:06.440 --> 37:11.440
+ is that our brain is wired up as a deep network.
+
+37:11.440 --> 37:22.440
+ So it can learn, understand, solve problems that have this compositional structure.
+
+37:22.440 --> 37:29.440
+ And it cannot solve problems that don't have this compositional structure.
+
+37:29.440 --> 37:37.440
+ So the problems we are accustomed to, we think about, we test our algorithms on,
+
+37:37.440 --> 37:42.440
+ are this compositional structure because our brain is made up.
+
+37:42.440 --> 37:46.440
+ And that's, in a sense, an evolutionary perspective that we've...
+
+37:46.440 --> 37:54.440
+ So the ones that weren't dealing with the compositional nature of reality died off?
+
+37:54.440 --> 38:05.440
+ Yes, but also could be, maybe the reason why we have this local connectivity in the brain,
+
+38:05.440 --> 38:10.440
+ like simple cells in cortex looking only at the small part of the image,
+
+38:10.440 --> 38:16.440
+ each one of them, and then other cells looking at the small number of the simple cells and so on.
+
+38:16.440 --> 38:24.440
+ The reason for this may be purely that it was difficult to grow long range connectivity.
+
+38:24.440 --> 38:33.440
+ So suppose it's, you know, for biology, it's possible to grow short range connectivity,
+
+38:33.440 --> 38:39.440
+ but not long range also because there is a limited number of long range.
+
+38:39.440 --> 38:44.440
+ And so you have this limitation from the biology.
+
+38:44.440 --> 38:49.440
+ And this means you build a deep convolutional network.
+
+38:49.440 --> 38:53.440
+ This would be something like a deep convolutional network.
+
+38:53.440 --> 38:57.440
+ And this is great for solving certain class of problems.
+
+38:57.440 --> 39:02.440
+ These are the ones we find easy and important for our life.
+
+39:02.440 --> 39:06.440
+ And yes, they were enough for us to survive.
+
+39:06.440 --> 39:13.440
+ And you can start a successful business on solving those problems with mobile eye.
+
+39:13.440 --> 39:16.440
+ Driving is a compositional problem.
+
+39:16.440 --> 39:25.440
+ So on the learning task, we don't know much about how the brain learns in terms of optimization.
+
+39:25.440 --> 39:31.440
+ So the thing that's stochastic gradient descent is what artificial neural networks
+
+39:31.440 --> 39:38.440
+ use for the most part to adjust the parameters in such a way that it's able to deal
+
+39:38.440 --> 39:42.440
+ based on the labeled data, it's able to solve the problem.
+
+39:42.440 --> 39:49.440
+ So what's your intuition about why it works at all?
+
+39:49.440 --> 39:55.440
+ How hard of a problem it is to optimize a neural network, artificial neural network?
+
+39:55.440 --> 39:57.440
+ Is there other alternatives?
+
+39:57.440 --> 40:03.440
+ Just in general, your intuition is behind this very simplistic algorithm
+
+40:03.440 --> 40:05.440
+ that seems to do pretty good, surprising.
+
+40:05.440 --> 40:07.440
+ Yes, yes.
+
+40:07.440 --> 40:16.440
+ So I find neuroscience, the architecture of cortex is really similar to the architecture of deep networks.
+
+40:16.440 --> 40:26.440
+ So there is a nice correspondence there between the biology and this kind of local connectivity hierarchical
+
+40:26.440 --> 40:28.440
+ architecture.
+
+40:28.440 --> 40:35.440
+ The stochastic gradient descent, as you said, is a very simple technique.
+
+40:35.440 --> 40:49.440
+ It seems pretty unlikely that biology could do that from what we know right now about cortex and neurons and synapses.
+
+40:49.440 --> 40:58.440
+ So it's a big question open whether there are other optimization learning algorithms
+
+40:58.440 --> 41:02.440
+ that can replace stochastic gradient descent.
+
+41:02.440 --> 41:11.440
+ And my guess is yes, but nobody has found yet a real answer.
+
+41:11.440 --> 41:17.440
+ I mean, people are trying, still trying, and there are some interesting ideas.
+
+41:17.440 --> 41:27.440
+ The fact that stochastic gradient descent is so successful, this has become clearly not so mysterious.
+
+41:27.440 --> 41:39.440
+ And the reason is that it's an interesting fact, you know, is a change in a sense in how people think about statistics.
+
+41:39.440 --> 41:51.440
+ And this is the following is that typically when you had data and you had, say, a model with parameters,
+
+41:51.440 --> 41:55.440
+ you are trying to fit the model to the data, you know, to fit the parameter.
+
+41:55.440 --> 42:12.440
+ And typically the kind of kind of crowd wisdom type idea was you should have at least, you know, twice the number of data than the number of parameters.
+
+42:12.440 --> 42:15.440
+ Maybe 10 times is better.
+
+42:15.440 --> 42:24.440
+ Now, the way you train neural network these days is that they have 10 or 100 times more parameters than data.
+
+42:24.440 --> 42:26.440
+ Exactly the opposite.
+
+42:26.440 --> 42:34.440
+ And which, you know, it has been one of the puzzles about neural networks.
+
+42:34.440 --> 42:40.440
+ How can you get something that really works when you have so much freedom?
+
+42:40.440 --> 42:43.440
+ From that little data you can generalize somehow.
+
+42:43.440 --> 42:44.440
+ Right, exactly.
+
+42:44.440 --> 42:48.440
+ Do you think the stochastic nature of it is essential, the randomness?
+
+42:48.440 --> 43:00.440
+ I think we have some initial understanding why this happens, but one nice side effect of having this over parameterization, more parameters than data,
+
+43:00.440 --> 43:07.440
+ is that when you look for the minima of a loss function like stochastic degree of descent is doing,
+
+43:07.440 --> 43:19.440
+ you find I made some calculations based on some old basic theorem of algebra called Bezu theorem.
+
+43:19.440 --> 43:25.440
+ And that gives you an estimate of the number of solutions of a system of polynomial equation.
+
+43:25.440 --> 43:38.440
+ Anyway, the bottom line is that there are probably more minima for a typical deep networks than atoms in the universe.
+
+43:38.440 --> 43:43.440
+ Just to say there are a lot because of the over parameterization.
+
+43:43.440 --> 43:44.440
+ Yes.
+
+43:44.440 --> 43:48.440
+ More global minimum, zero minimum, good minimum.
+
+43:48.440 --> 43:51.440
+ More global minimum.
+
+43:51.440 --> 44:00.440
+ Yes, a lot of them, so you have a lot of solutions, so it's not so surprising that you can find them relatively easily.
+
+44:00.440 --> 44:04.440
+ This is because of the over parameterization.
+
+44:04.440 --> 44:09.440
+ The over parameterization sprinkles that entire space with solutions that are pretty good.
+
+44:09.440 --> 44:11.440
+ It's not so surprising, right?
+
+44:11.440 --> 44:17.440
+ It's like if you have a system of linear equation and you have more unknowns than equations,
+
+44:17.440 --> 44:24.440
+ then we know you have an infinite number of solutions and the question is to pick one.
+
+44:24.440 --> 44:27.440
+ That's another story, but you have an infinite number of solutions,
+
+44:27.440 --> 44:32.440
+ so there are a lot of value of your unknowns that satisfy the equations.
+
+44:32.440 --> 44:37.440
+ But it's possible that there's a lot of those solutions that aren't very good.
+
+44:37.440 --> 44:38.440
+ What's surprising is that they're pretty good.
+
+44:38.440 --> 44:39.440
+ So that's a separate question.
+
+44:39.440 --> 44:43.440
+ Why can you pick one that generalizes one?
+
+44:43.440 --> 44:46.440
+ That's a separate question with separate answers.
+
+44:46.440 --> 44:53.440
+ One theorem that people like to talk about that inspires imagination of the power of neural networks
+
+44:53.440 --> 45:00.440
+ is the universal approximation theorem that you can approximate any computable function
+
+45:00.440 --> 45:04.440
+ with just a finite number of neurons and a single hidden layer.
+
+45:04.440 --> 45:07.440
+ Do you find this theorem one surprising?
+
+45:07.440 --> 45:12.440
+ Do you find it useful, interesting, inspiring?
+
+45:12.440 --> 45:16.440
+ No, this one, I never found it very surprising.
+
+45:16.440 --> 45:22.440
+ It was known since the 80s, since I entered the field,
+
+45:22.440 --> 45:27.440
+ because it's basically the same as Viastras theorem,
+
+45:27.440 --> 45:34.440
+ which says that I can approximate any continuous function with a polynomial of sufficiently,
+
+45:34.440 --> 45:37.440
+ with a sufficient number of terms, monomials.
+
+45:37.440 --> 45:41.440
+ It's basically the same, and the proofs are very similar.
+
+45:41.440 --> 45:48.440
+ So your intuition was there was never any doubt that neural networks in theory could be very strong approximations.
+
+45:48.440 --> 45:58.440
+ The interesting question is that if this theorem says you can approximate fine,
+
+45:58.440 --> 46:06.440
+ but when you ask how many neurons, for instance, or in the case of how many monomials,
+
+46:06.440 --> 46:11.440
+ I need to get a good approximation.
+
+46:11.440 --> 46:20.440
+ Then it turns out that that depends on the dimensionality of your function, how many variables you have.
+
+46:20.440 --> 46:25.440
+ But it depends on the dimensionality of your function in a bad way.
+
+46:25.440 --> 46:35.440
+ For instance, suppose you want an error which is no worse than 10% in your approximation.
+
+46:35.440 --> 46:40.440
+ If you want to approximate your function within 10%,
+
+46:40.440 --> 46:48.440
+ then it turns out that the number of units you need are in the order of 10 to the dimensionality, d.
+
+46:48.440 --> 46:50.440
+ How many variables?
+
+46:50.440 --> 46:57.440
+ So if you have two variables, d is 2 and you have 100 units and OK.
+
+46:57.440 --> 47:02.440
+ But if you have, say, 200 by 200 pixel images,
+
+47:02.440 --> 47:06.440
+ now this is 40,000, whatever.
+
+47:06.440 --> 47:09.440
+ We again go to the size of the universe pretty quickly.
+
+47:09.440 --> 47:13.440
+ Exactly, 10 to the 40,000 or something.
+
+47:13.440 --> 47:21.440
+ And so this is called the curse of dimensionality, not quite appropriately.
+
+47:21.440 --> 47:27.440
+ And the hope is with the extra layers you can remove the curse.
+
+47:27.440 --> 47:34.440
+ What we proved is that if you have deep layers or hierarchical architecture
+
+47:34.440 --> 47:39.440
+ with the local connectivity of the type of convolutional deep learning,
+
+47:39.440 --> 47:46.440
+ and if you're dealing with a function that has this kind of hierarchical architecture,
+
+47:46.440 --> 47:50.440
+ then you avoid completely the curse.
+
+47:50.440 --> 47:53.440
+ You've spoken a lot about supervised deep learning.
+
+47:53.440 --> 47:58.440
+ What are your thoughts, hopes, views on the challenges of unsupervised learning
+
+47:58.440 --> 48:04.440
+ with GANs, with generative adversarial networks?
+
+48:04.440 --> 48:08.440
+ Do you see those as distinct, the power of GANs,
+
+48:08.440 --> 48:12.440
+ do you see those as distinct from supervised methods in neural networks,
+
+48:12.440 --> 48:16.440
+ or are they really all in the same representation ballpark?
+
+48:16.440 --> 48:24.440
+ GANs is one way to get estimation of probability densities,
+
+48:24.440 --> 48:29.440
+ which is a somewhat new way that people have not done before.
+
+48:29.440 --> 48:38.440
+ I don't know whether this will really play an important role in intelligence,
+
+48:38.440 --> 48:47.440
+ or it's interesting, I'm less enthusiastic about it than many people in the field.
+
+48:47.440 --> 48:53.440
+ I have the feeling that many people in the field are really impressed by the ability
+
+48:53.440 --> 49:00.440
+ of producing realistic looking images in this generative way.
+
+49:00.440 --> 49:02.440
+ Which describes the popularity of the methods,
+
+49:02.440 --> 49:10.440
+ but you're saying that while that's exciting and cool to look at, it may not be the tool that's useful for it.
+
+49:10.440 --> 49:12.440
+ So you describe it kind of beautifully.
+
+49:12.440 --> 49:17.440
+ Current supervised methods go N to infinity in terms of the number of labeled points,
+
+49:17.440 --> 49:20.440
+ and we really have to figure out how to go to N to 1.
+
+49:20.440 --> 49:24.440
+ And you're thinking GANs might help, but they might not be the right...
+
+49:24.440 --> 49:28.440
+ I don't think for that problem, which I really think is important.
+
+49:28.440 --> 49:35.440
+ I think they certainly have applications, for instance, in computer graphics.
+
+49:35.440 --> 49:43.440
+ I did work long ago, which was a little bit similar in terms of,
+
+49:43.440 --> 49:49.440
+ saying I have a network and I present images,
+
+49:49.440 --> 49:59.440
+ so the input is images and output is, for instance, the pose of the image, a face, how much is smiling,
+
+49:59.440 --> 50:02.440
+ is rotated 45 degrees or not.
+
+50:02.440 --> 50:08.440
+ What about having a network that I train with the same data set,
+
+50:08.440 --> 50:10.440
+ but now I invert input and output.
+
+50:10.440 --> 50:16.440
+ Now the input is the pose or the expression, a number, certain numbers,
+
+50:16.440 --> 50:19.440
+ and the output is the image and I train it.
+
+50:19.440 --> 50:27.440
+ And we did pretty good interesting results in terms of producing very realistic looking images.
+
+50:27.440 --> 50:35.440
+ It was less sophisticated mechanism, but the output was pretty less than GANs,
+
+50:35.440 --> 50:38.440
+ but the output was pretty much of the same quality.
+
+50:38.440 --> 50:43.440
+ So I think for computer graphics type application,
+
+50:43.440 --> 50:48.440
+ definitely GANs can be quite useful and not only for that,
+
+50:48.440 --> 51:01.440
+ but for helping, for instance, on this problem unsupervised example of reducing the number of labelled examples,
+
+51:01.440 --> 51:10.440
+ I think people, it's like they think they can get out more than they put in.
+
+51:10.440 --> 51:13.440
+ There's no free lunch, as you said.
+
+51:13.440 --> 51:16.440
+ What's your intuition?
+
+51:16.440 --> 51:24.440
+ How can we slow the growth of N to infinity in supervised learning?
+
+51:24.440 --> 51:29.440
+ So, for example, Mobileye has very successfully,
+
+51:29.440 --> 51:34.440
+ I mean essentially annotated large amounts of data to be able to drive a car.
+
+51:34.440 --> 51:40.440
+ Now, one thought is, so we're trying to teach machines, the school of AI,
+
+51:40.440 --> 51:45.440
+ and we're trying to, so how can we become better teachers, maybe?
+
+51:45.440 --> 51:47.440
+ That's one way.
+
+51:47.440 --> 51:58.440
+ I like that because, again, one caricature of the history of computer science,
+
+51:58.440 --> 52:09.440
+ it begins with programmers, expensive, continuous labellers, cheap,
+
+52:09.440 --> 52:16.440
+ and the future would be schools, like we have for kids.
+
+52:16.440 --> 52:26.440
+ Currently, the labelling methods, we're not selective about which examples we teach networks with.
+
+52:26.440 --> 52:33.440
+ I think the focus of making networks that learn much faster is often on the architecture side,
+
+52:33.440 --> 52:37.440
+ but how can we pick better examples with which to learn?
+
+52:37.440 --> 52:39.440
+ Do you have intuitions about that?
+
+52:39.440 --> 52:50.440
+ Well, that's part of the problem, but the other one is, if we look at biology,
+
+52:50.440 --> 52:58.440
+ the reasonable assumption, I think, is in the same spirit as I said,
+
+52:58.440 --> 53:03.440
+ evolution is opportunistic and has weak priors.
+
+53:03.440 --> 53:10.440
+ The way I think the intelligence of a child, a baby may develop,
+
+53:10.440 --> 53:17.440
+ is by bootstrapping weak priors from evolution.
+
+53:17.440 --> 53:26.440
+ For instance, you can assume that you have most organisms,
+
+53:26.440 --> 53:37.440
+ including human babies, built in some basic machinery to detect motion and relative motion.
+
+53:37.440 --> 53:46.440
+ In fact, we know all insects, from fruit flies to other animals, they have this.
+
+53:46.440 --> 53:55.440
+ Even in the retinas, in the very peripheral part, it's very conserved across species,
+
+53:55.440 --> 53:58.440
+ something that evolution discovered early.
+
+53:58.440 --> 54:05.440
+ It may be the reason why babies tend to look in the first few days to moving objects,
+
+54:05.440 --> 54:07.440
+ and not to not moving objects.
+
+54:07.440 --> 54:11.440
+ Now, moving objects means, okay, they're attracted by motion,
+
+54:11.440 --> 54:19.440
+ but motion also means that motion gives automatic segmentation from the background.
+
+54:19.440 --> 54:26.440
+ So because of motion boundaries, either the object is moving,
+
+54:26.440 --> 54:32.440
+ or the eye of the baby is tracking the moving object, and the background is moving.
+
+54:32.440 --> 54:37.440
+ Yeah, so just purely on the visual characteristics of the scene, that seems to be the most useful.
+
+54:37.440 --> 54:43.440
+ Right, so it's like looking at an object without background.
+
+54:43.440 --> 54:49.440
+ It's ideal for learning the object, otherwise it's really difficult, because you have so much stuff.
+
+54:49.440 --> 54:54.440
+ So suppose you do this at the beginning, first weeks,
+
+54:54.440 --> 55:01.440
+ then after that you can recognize the object, now they are imprinted, the number one,
+
+55:01.440 --> 55:05.440
+ even in the background, even without motion.
+
+55:05.440 --> 55:10.440
+ So that's the, by the way, I just want to ask on the object recognition problem,
+
+55:10.440 --> 55:16.440
+ so there is this being responsive to movement and doing edge detection, essentially.
+
+55:16.440 --> 55:20.440
+ What's the gap between being effectively,
+
+55:20.440 --> 55:27.440
+ effectively visually recognizing stuff, detecting where it is, and understanding the scene?
+
+55:27.440 --> 55:32.440
+ Is this a huge gap in many layers, or is it close?
+
+55:32.440 --> 55:35.440
+ No, I think that's a huge gap.
+
+55:35.440 --> 55:44.440
+ I think present algorithm with all the success that we have, and the fact that are a lot of very useful,
+
+55:44.440 --> 55:51.440
+ I think we are in a golden age for applications of low level vision,
+
+55:51.440 --> 55:56.440
+ and low level speech recognition, and so on, you know, Alexa, and so on.
+
+55:56.440 --> 56:01.440
+ There are many more things of similar level to be done, including medical diagnosis and so on,
+
+56:01.440 --> 56:11.440
+ but we are far from what we call understanding of a scene, of language, of actions, of people.
+
+56:11.440 --> 56:17.440
+ That is, despite the claims, that's, I think, very far.
+
+56:17.440 --> 56:19.440
+ We're a little bit off.
+
+56:19.440 --> 56:24.440
+ So in popular culture, and among many researchers, some of which I've spoken with,
+
+56:24.440 --> 56:34.440
+ the Sewell Russell and Elon Musk, in and out of the AI field, there's a concern about the existential threat of AI.
+
+56:34.440 --> 56:44.440
+ And how do you think about this concern, and is it valuable to think about large scale,
+
+56:44.440 --> 56:51.440
+ long term, unintended consequences of intelligent systems we try to build?
+
+56:51.440 --> 56:58.440
+ I always think it's better to worry first, you know, early rather than late.
+
+56:58.440 --> 56:59.440
+ So worry is good.
+
+56:59.440 --> 57:02.440
+ Yeah, I'm not against worrying at all.
+
+57:02.440 --> 57:15.440
+ Personally, I think that, you know, it will take a long time before there is real reason to be worried.
+
+57:15.440 --> 57:23.440
+ But as I said, I think it's good to put in place and think about possible safety against,
+
+57:23.440 --> 57:35.440
+ what I find a bit misleading are things like that have been said by people I know, like Elon Musk and what is Bostrom in particular,
+
+57:35.440 --> 57:39.440
+ and what is his first name, Nick Bostrom, right?
+
+57:39.440 --> 57:46.440
+ And, you know, and a couple of other people that, for instance, AI is more dangerous than nuclear weapons.
+
+57:46.440 --> 57:50.440
+ I think that's really wrong.
+
+57:50.440 --> 57:59.440
+ That can be misleading, because in terms of priority, we should still be more worried about nuclear weapons
+
+57:59.440 --> 58:05.440
+ and what people are doing about it and so on than AI.
+
+58:05.440 --> 58:15.440
+ And you've spoken about them as obvious and yourself saying that you think you'll be about 100 years out
+
+58:15.440 --> 58:20.440
+ before we have a general intelligence system that's on par with the human being.
+
+58:20.440 --> 58:22.440
+ Do you have any updates for those predictions?
+
+58:22.440 --> 58:23.440
+ Well, I think he said...
+
+58:23.440 --> 58:25.440
+ He said 20, I think.
+
+58:25.440 --> 58:26.440
+ He said 20, right.
+
+58:26.440 --> 58:27.440
+ This was a couple of years ago.
+
+58:27.440 --> 58:31.440
+ I have not asked him again, so I should have.
+
+58:31.440 --> 58:38.440
+ Your own prediction, what's your prediction about when you'll be truly surprised
+
+58:38.440 --> 58:42.440
+ and what's the confidence interval on that?
+
+58:42.440 --> 58:46.440
+ You know, it's so difficult to predict the future and even the present.
+
+58:46.440 --> 58:48.440
+ It's pretty hard to predict.
+
+58:48.440 --> 58:50.440
+ Right, but I would be...
+
+58:50.440 --> 58:52.440
+ As I said, this is completely...
+
+58:52.440 --> 58:56.440
+ I would be more like Rod Brooks.
+
+58:56.440 --> 58:59.440
+ I think he's about 200 years old.
+
+58:59.440 --> 59:01.440
+ 200 years.
+
+59:01.440 --> 59:06.440
+ When we have this kind of AGI system, artificial intelligence system,
+
+59:06.440 --> 59:12.440
+ you're sitting in a room with her, him, it,
+
+59:12.440 --> 59:17.440
+ do you think it will be the underlying design of such a system
+
+59:17.440 --> 59:19.440
+ and something we'll be able to understand?
+
+59:19.440 --> 59:20.440
+ It will be simple?
+
+59:20.440 --> 59:25.440
+ Do you think it will be explainable?
+
+59:25.440 --> 59:27.440
+ Understandable by us?
+
+59:27.440 --> 59:31.440
+ Your intuition, again, we're in the realm of philosophy a little bit.
+
+59:31.440 --> 59:35.440
+ Well, probably no.
+
+59:35.440 --> 59:42.440
+ But again, it depends what you really mean for understanding.
+
+59:42.440 --> 59:53.440
+ I think we don't understand how deep networks work.
+
+59:53.440 --> 59:56.440
+ I think we're beginning to have a theory now.
+
+59:56.440 --> 59:59.440
+ But in the case of deep networks,
+
+59:59.440 --> 1:00:06.440
+ or even in the case of the simpler kernel machines or linear classifier,
+
+1:00:06.440 --> 1:00:12.440
+ we really don't understand the individual units or so.
+
+1:00:12.440 --> 1:00:20.440
+ But we understand what the computation and the limitations and the properties of it are.
+
+1:00:20.440 --> 1:00:24.440
+ It's similar to many things.
+
+1:00:24.440 --> 1:00:29.440
+ Does it mean to understand how a fusion bomb works?
+
+1:00:29.440 --> 1:00:35.440
+ How many of us, you know, many of us understand the basic principle
+
+1:00:35.440 --> 1:00:40.440
+ and some of us may understand deeper details?
+
+1:00:40.440 --> 1:00:44.440
+ In that sense, understanding is, as a community, as a civilization,
+
+1:00:44.440 --> 1:00:46.440
+ can we build another copy of it?
+
+1:00:46.440 --> 1:00:47.440
+ Okay.
+
+1:00:47.440 --> 1:00:50.440
+ And in that sense, do you think there'll be,
+
+1:00:50.440 --> 1:00:56.440
+ there'll need to be some evolutionary component where it runs away from our understanding?
+
+1:00:56.440 --> 1:00:59.440
+ Or do you think it could be engineered from the ground up?
+
+1:00:59.440 --> 1:01:02.440
+ The same way you go from the transistor to PowerPoint?
+
+1:01:02.440 --> 1:01:03.440
+ Right.
+
+1:01:03.440 --> 1:01:09.440
+ So many years ago, this was actually 40, 41 years ago,
+
+1:01:09.440 --> 1:01:13.440
+ I wrote a paper with David Maher,
+
+1:01:13.440 --> 1:01:19.440
+ who was one of the founding fathers of computer vision, computational vision.
+
+1:01:19.440 --> 1:01:23.440
+ I wrote a paper about levels of understanding,
+
+1:01:23.440 --> 1:01:28.440
+ which is related to the question we discussed earlier about understanding PowerPoint,
+
+1:01:28.440 --> 1:01:31.440
+ understanding transistors and so on.
+
+1:01:31.440 --> 1:01:38.440
+ And, you know, in that kind of framework, we had a level of the hardware
+
+1:01:38.440 --> 1:01:41.440
+ and the top level of the algorithms.
+
+1:01:41.440 --> 1:01:44.440
+ We did not have learning.
+
+1:01:44.440 --> 1:01:54.440
+ Recently, I updated adding levels and one level I added to those three was learning.
+
+1:01:54.440 --> 1:01:59.440
+ So, and you can imagine, you could have a good understanding
+
+1:01:59.440 --> 1:02:04.440
+ of how you construct learning machine, like we do.
+
+1:02:04.440 --> 1:02:13.440
+ But being unable to describe in detail what the learning machines will discover, right?
+
+1:02:13.440 --> 1:02:19.440
+ Now, that would be still a powerful understanding if I can build a learning machine,
+
+1:02:19.440 --> 1:02:25.440
+ even if I don't understand in detail every time it learns something.
+
+1:02:25.440 --> 1:02:31.440
+ Just like our children, if they start listening to a certain type of music,
+
+1:02:31.440 --> 1:02:33.440
+ I don't know, Miley Cyrus or something,
+
+1:02:33.440 --> 1:02:37.440
+ you don't understand why they came to that particular preference,
+
+1:02:37.440 --> 1:02:39.440
+ but you understand the learning process.
+
+1:02:39.440 --> 1:02:41.440
+ That's very interesting.
+
+1:02:41.440 --> 1:02:50.440
+ So, on learning for systems to be part of our world,
+
+1:02:50.440 --> 1:02:56.440
+ it has a certain, one of the challenging things that you've spoken about is learning ethics,
+
+1:02:56.440 --> 1:02:59.440
+ learning morals.
+
+1:02:59.440 --> 1:03:06.440
+ And how hard do you think is the problem of, first of all, humans understanding our ethics?
+
+1:03:06.440 --> 1:03:10.440
+ What is the origin on the neural and low level of ethics?
+
+1:03:10.440 --> 1:03:12.440
+ What is it at the higher level?
+
+1:03:12.440 --> 1:03:17.440
+ Is it something that's learnable from machines in your intuition?
+
+1:03:17.440 --> 1:03:23.440
+ I think, yeah, ethics is learnable, very likely.
+
+1:03:23.440 --> 1:03:36.440
+ I think it's one of these problems where I think understanding the neuroscience of ethics,
+
+1:03:36.440 --> 1:03:42.440
+ people discuss there is an ethics of neuroscience.
+
+1:03:42.440 --> 1:03:46.440
+ How a neuroscientist should or should not behave,
+
+1:03:46.440 --> 1:03:53.440
+ can think of a neurosurgeon and the ethics that he or she has to be.
+
+1:03:53.440 --> 1:03:57.440
+ But I'm more interested in the neuroscience of ethics.
+
+1:03:57.440 --> 1:04:01.440
+ You're blowing my mind right now, the neuroscience of ethics, it's very meta.
+
+1:04:01.440 --> 1:04:09.440
+ And I think that would be important to understand also for being able to design machines
+
+1:04:09.440 --> 1:04:14.440
+ that are ethical machines in our sense of ethics.
+
+1:04:14.440 --> 1:04:20.440
+ And you think there is something in neuroscience, there's patterns,
+
+1:04:20.440 --> 1:04:25.440
+ tools in neuroscience that could help us shed some light on ethics
+
+1:04:25.440 --> 1:04:29.440
+ or is it more on the psychologist's sociology at a much higher level?
+
+1:04:29.440 --> 1:04:33.440
+ No, there is psychology, but there is also, in the meantime,
+
+1:04:33.440 --> 1:04:41.440
+ there is evidence, fMRI, of specific areas of the brain
+
+1:04:41.440 --> 1:04:44.440
+ that are involved in certain ethical judgment.
+
+1:04:44.440 --> 1:04:49.440
+ And not only this, you can stimulate those areas with magnetic fields
+
+1:04:49.440 --> 1:04:54.440
+ and change the ethical decisions.
+
+1:04:54.440 --> 1:05:00.440
+ So that's work by a colleague of mine, Rebecca Sacks,
+
+1:05:00.440 --> 1:05:04.440
+ and there are other researchers doing similar work.
+
+1:05:04.440 --> 1:05:11.440
+ And I think this is the beginning, but ideally at some point
+
+1:05:11.440 --> 1:05:17.440
+ we'll have an understanding of how this works and why it evolved, right?
+
+1:05:17.440 --> 1:05:21.440
+ The big why question, yeah, it must have some purpose.
+
+1:05:21.440 --> 1:05:29.440
+ Yeah, obviously it has some social purposes, probably.
+
+1:05:29.440 --> 1:05:34.440
+ If neuroscience holds the key to at least eliminate some aspect of ethics,
+
+1:05:34.440 --> 1:05:36.440
+ that means it could be a learnable problem.
+
+1:05:36.440 --> 1:05:38.440
+ Yeah, exactly.
+
+1:05:38.440 --> 1:05:41.440
+ And as we're getting into harder and harder questions,
+
+1:05:41.440 --> 1:05:44.440
+ let's go to the hard problem of consciousness.
+
+1:05:44.440 --> 1:05:51.440
+ Is this an important problem for us to think about and solve on the engineering
+
+1:05:51.440 --> 1:05:55.440
+ of intelligence side of your work, of our dream?
+
+1:05:55.440 --> 1:05:57.440
+ You know, it's unclear.
+
+1:05:57.440 --> 1:06:04.440
+ So, again, this is a deep problem, partly because it's very difficult
+
+1:06:04.440 --> 1:06:16.440
+ to define consciousness and there is a debate among neuroscientists
+
+1:06:16.440 --> 1:06:22.440
+ about whether consciousness and philosophers, of course,
+
+1:06:22.440 --> 1:06:30.440
+ whether consciousness is something that requires flesh and blood, so to speak,
+
+1:06:30.440 --> 1:06:40.440
+ or could be, you know, that we could have silicon devices that are conscious,
+
+1:06:40.440 --> 1:06:45.440
+ or up to a statement like everything has some degree of consciousness
+
+1:06:45.440 --> 1:06:48.440
+ and some more than others.
+
+1:06:48.440 --> 1:06:53.440
+ This is like Giulio Tonioni and Fee.
+
+1:06:53.440 --> 1:06:56.440
+ We just recently talked to Christof Ko.
+
+1:06:56.440 --> 1:07:00.440
+ Christof was my first graduate student.
+
+1:07:00.440 --> 1:07:06.440
+ Do you think it's important to illuminate aspects of consciousness
+
+1:07:06.440 --> 1:07:10.440
+ in order to engineer intelligence systems?
+
+1:07:10.440 --> 1:07:14.440
+ Do you think an intelligence system would ultimately have consciousness?
+
+1:07:14.440 --> 1:07:18.440
+ Are they intro linked?
+
+1:07:18.440 --> 1:07:23.440
+ You know, most of the people working in artificial intelligence, I think,
+
+1:07:23.440 --> 1:07:29.440
+ they answer, we don't strictly need consciousness to have an intelligence system.
+
+1:07:29.440 --> 1:07:35.440
+ That's sort of the easier question, because it's a very engineering answer to the question.
+
+1:07:35.440 --> 1:07:38.440
+ It has a touring test, we don't need consciousness.
+
+1:07:38.440 --> 1:07:47.440
+ But if you were to go, do you think it's possible that we need to have that kind of self awareness?
+
+1:07:47.440 --> 1:07:49.440
+ We may, yes.
+
+1:07:49.440 --> 1:08:00.440
+ So, for instance, I personally think that when test a machine or a person in a touring test,
+
+1:08:00.440 --> 1:08:10.440
+ in an extended touring testing, I think consciousness is part of what we require in that test,
+
+1:08:10.440 --> 1:08:14.440
+ you know, implicitly to say that this is intelligent.
+
+1:08:14.440 --> 1:08:17.440
+ Christof disagrees.
+
+1:08:17.440 --> 1:08:19.440
+ Yes, he does.
+
+1:08:19.440 --> 1:08:24.440
+ Despite many other romantic notions he holds, he disagrees with that one.
+
+1:08:24.440 --> 1:08:26.440
+ Yes, that's right.
+
+1:08:26.440 --> 1:08:29.440
+ So, you know, who would see?
+
+1:08:29.440 --> 1:08:37.440
+ Do you think, as a quick question, Ernest Becker's fear of death,
+
+1:08:37.440 --> 1:08:48.440
+ do you think mortality and those kinds of things are important for consciousness and for intelligence,
+
+1:08:48.440 --> 1:08:53.440
+ the finiteness of life, finiteness of existence,
+
+1:08:53.440 --> 1:09:00.440
+ or is that just a side effect of evolutionary side effect that's useful for natural selection?
+
+1:09:00.440 --> 1:09:05.440
+ Do you think this kind of thing that this interview is going to run out of time soon,
+
+1:09:05.440 --> 1:09:08.440
+ our life will run out of time soon?
+
+1:09:08.440 --> 1:09:12.440
+ Do you think that's needed to make this conversation good and life good?
+
+1:09:12.440 --> 1:09:14.440
+ You know, I never thought about it.
+
+1:09:14.440 --> 1:09:16.440
+ It's a very interesting question.
+
+1:09:16.440 --> 1:09:25.440
+ I think Steve Jobs in his commencement speech at Stanford argued that, you know,
+
+1:09:25.440 --> 1:09:30.440
+ having a finite life was important for stimulating achievements.
+
+1:09:30.440 --> 1:09:32.440
+ It was a different.
+
+1:09:32.440 --> 1:09:34.440
+ You live every day like it's your last, right?
+
+1:09:34.440 --> 1:09:35.440
+ Yeah.
+
+1:09:35.440 --> 1:09:45.440
+ So, rationally, I don't think strictly you need mortality for consciousness, but...
+
+1:09:45.440 --> 1:09:46.440
+ Who knows?
+
+1:09:46.440 --> 1:09:49.440
+ They seem to go together in our biological system, right?
+
+1:09:49.440 --> 1:09:51.440
+ Yeah.
+
+1:09:51.440 --> 1:09:57.440
+ You've mentioned before and the students are associated with...
+
+1:09:57.440 --> 1:10:01.440
+ AlphaGo immobilized the big recent success stories in AI.
+
+1:10:01.440 --> 1:10:05.440
+ I think it's captivated the entire world of what AI can do.
+
+1:10:05.440 --> 1:10:10.440
+ So, what do you think will be the next breakthrough?
+
+1:10:10.440 --> 1:10:13.440
+ What's your intuition about the next breakthrough?
+
+1:10:13.440 --> 1:10:16.440
+ Of course, I don't know where the next breakthrough is.
+
+1:10:16.440 --> 1:10:22.440
+ I think that there is a good chance, as I said before, that the next breakthrough
+
+1:10:22.440 --> 1:10:27.440
+ would also be inspired by, you know, neuroscience.
+
+1:10:27.440 --> 1:10:31.440
+ But which one?
+
+1:10:31.440 --> 1:10:32.440
+ I don't know.
+
+1:10:32.440 --> 1:10:33.440
+ And there's...
+
+1:10:33.440 --> 1:10:35.440
+ So, MIT has this quest for intelligence.
+
+1:10:35.440 --> 1:10:36.440
+ Yeah.
+
+1:10:36.440 --> 1:10:41.440
+ And there's a few moonshots which, in that spirit, which ones are you excited about?
+
+1:10:41.440 --> 1:10:42.440
+ What...
+
+1:10:42.440 --> 1:10:44.440
+ Which projects kind of...
+
+1:10:44.440 --> 1:10:48.440
+ Well, of course, I'm excited about one of the moonshots with...
+
+1:10:48.440 --> 1:10:52.440
+ Which is our center for brains, minds, and machines.
+
+1:10:52.440 --> 1:10:57.440
+ The one which is fully funded by NSF.
+
+1:10:57.440 --> 1:10:59.440
+ And it's a...
+
+1:10:59.440 --> 1:11:02.440
+ It is about visual intelligence.
+
+1:11:02.440 --> 1:11:05.440
+ And that one is particularly about understanding.
+
+1:11:05.440 --> 1:11:07.440
+ Visual intelligence.
+
+1:11:07.440 --> 1:11:16.440
+ Visual cortex and visual intelligence in the sense of how we look around ourselves
+
+1:11:16.440 --> 1:11:25.440
+ and understand the world around ourselves, you know, meaning what is going on,
+
+1:11:25.440 --> 1:11:31.440
+ how we could go from here to there without hitting obstacles.
+
+1:11:31.440 --> 1:11:36.440
+ You know, whether there are other agents, people in the environment.
+
+1:11:36.440 --> 1:11:41.440
+ These are all things that we perceive very quickly.
+
+1:11:41.440 --> 1:11:47.440
+ And it's something actually quite close to being conscious, not quite.
+
+1:11:47.440 --> 1:11:53.440
+ But there is this interesting experiment that was run at Google X,
+
+1:11:53.440 --> 1:11:58.440
+ which is, in a sense, is just a virtual reality experiment,
+
+1:11:58.440 --> 1:12:09.440
+ but in which they had subject sitting, say, in a chair with goggles, like Oculus and so on.
+
+1:12:09.440 --> 1:12:11.440
+ Earphones.
+
+1:12:11.440 --> 1:12:20.440
+ And they were seeing through the eyes of a robot nearby to cameras, microphones for receiving.
+
+1:12:20.440 --> 1:12:23.440
+ So their sensory system was there.
+
+1:12:23.440 --> 1:12:30.440
+ And the impression of all the subjects, very strong, they could not shake it off,
+
+1:12:30.440 --> 1:12:35.440
+ was that they were where the robot was.
+
+1:12:35.440 --> 1:12:42.440
+ They could look at themselves from the robot and still feel they were where the robot is.
+
+1:12:42.440 --> 1:12:45.440
+ They were looking at their body.
+
+1:12:45.440 --> 1:12:48.440
+ Their self had moved.
+
+1:12:48.440 --> 1:12:54.440
+ So some aspect of seeing understanding has to have ability to place yourself,
+
+1:12:54.440 --> 1:12:59.440
+ have a self awareness about your position in the world and what the world is.
+
+1:12:59.440 --> 1:13:04.440
+ So we may have to solve the heart problem of consciousness to solve it.
+
+1:13:04.440 --> 1:13:05.440
+ On their way, yes.
+
+1:13:05.440 --> 1:13:07.440
+ It's quite a moonshot.
+
+1:13:07.440 --> 1:13:14.440
+ So you've been an advisor to some incredible minds, including Demis Osabis, Christof Koch,
+
+1:13:14.440 --> 1:13:21.440
+ Amna Shashwar, like you said, all went on to become seminal figures in their respective fields.
+
+1:13:21.440 --> 1:13:28.440
+ From your own success as a researcher and from perspective as a mentor of these researchers,
+
+1:13:28.440 --> 1:13:33.440
+ having guided them in the way of advice,
+
+1:13:33.440 --> 1:13:39.440
+ what does it take to be successful in science and engineering careers?
+
+1:13:39.440 --> 1:13:47.440
+ Whether you're talking to somebody in their teens, 20s and 30s, what does that path look like?
+
+1:13:47.440 --> 1:13:52.440
+ It's curiosity and having fun.
+
+1:13:52.440 --> 1:14:01.440
+ And I think it's important also having fun with other curious minds.
+
+1:14:01.440 --> 1:14:06.440
+ It's the people you surround with to have fun and curiosity.
+
+1:14:06.440 --> 1:14:09.440
+ You mentioned Steve Jobs.
+
+1:14:09.440 --> 1:14:14.440
+ Is there also an underlying ambition that's unique that you saw,
+
+1:14:14.440 --> 1:14:18.440
+ or is it really does boil down to insatiable curiosity and fun?
+
+1:14:18.440 --> 1:14:20.440
+ Well, of course.
+
+1:14:20.440 --> 1:14:29.440
+ It's being curious in an active and ambitious way, yes, definitely.
+
+1:14:29.440 --> 1:14:38.440
+ But I think sometime in science, there are friends of mine who are like this.
+
+1:14:38.440 --> 1:14:44.440
+ You know, there are some of the scientists who like to work by themselves
+
+1:14:44.440 --> 1:14:54.440
+ and kind of communicate only when they complete their work or discover something.
+
+1:14:54.440 --> 1:15:02.440
+ I think I always found the actual process of discovering something
+
+1:15:02.440 --> 1:15:09.440
+ is more fun if it's together with other intelligent and curious and fun people.
+
+1:15:09.440 --> 1:15:13.440
+ So if you see the fun in that process, the side effect of that process
+
+1:15:13.440 --> 1:15:16.440
+ would be that you'll actually end up discovering something.
+
+1:15:16.440 --> 1:15:25.440
+ So as you've led many incredible efforts here, what's the secret to being a good advisor,
+
+1:15:25.440 --> 1:15:28.440
+ mentor, leader in a research setting?
+
+1:15:28.440 --> 1:15:35.440
+ Is it a similar spirit or what advice could you give to people, young faculty and so on?
+
+1:15:35.440 --> 1:15:42.440
+ It's partly repeating what I said about an environment that should be friendly and fun
+
+1:15:42.440 --> 1:15:52.440
+ and ambitious and, you know, I think I learned a lot from some of my advisors and friends
+
+1:15:52.440 --> 1:16:02.440
+ and some were physicists and there was, for instance, this behavior that was encouraged
+
+1:16:02.440 --> 1:16:08.440
+ of when somebody comes with a new idea in the group, unless it's really stupid
+
+1:16:08.440 --> 1:16:11.440
+ but you are always enthusiastic.
+
+1:16:11.440 --> 1:16:14.440
+ And then you're enthusiastic for a few minutes, for a few hours.
+
+1:16:14.440 --> 1:16:22.440
+ Then you start, you know, asking critically a few questions, testing this.
+
+1:16:22.440 --> 1:16:28.440
+ But, you know, this is a process that is, I think it's very good.
+
+1:16:28.440 --> 1:16:30.440
+ You have to be enthusiastic.
+
+1:16:30.440 --> 1:16:33.440
+ Sometimes people are very critical from the beginning.
+
+1:16:33.440 --> 1:16:35.440
+ That's not...
+
+1:16:35.440 --> 1:16:37.440
+ Yes, you have to give it a chance.
+
+1:16:37.440 --> 1:16:38.440
+ Yes.
+
+1:16:38.440 --> 1:16:39.440
+ That's seed to grow.
+
+1:16:39.440 --> 1:16:44.440
+ That said, with some of your ideas, which are quite revolutionary, so there's a witness,
+
+1:16:44.440 --> 1:16:49.440
+ especially in the human vision side and neuroscience side, there could be some pretty heated arguments.
+
+1:16:49.440 --> 1:16:51.440
+ Do you enjoy these?
+
+1:16:51.440 --> 1:16:55.440
+ Is that a part of science and academic pursuits that you enjoy?
+
+1:16:55.440 --> 1:16:56.440
+ Yeah.
+
+1:16:56.440 --> 1:17:00.440
+ Is that something that happens in your group as well?
+
+1:17:00.440 --> 1:17:02.440
+ Yeah, absolutely.
+
+1:17:02.440 --> 1:17:14.440
+ I also spent some time in Germany again, there is this tradition in which people are more forthright, less kind than here.
+
+1:17:14.440 --> 1:17:23.440
+ So, you know, in the US, when you write a bad letter, you still say, this guy is nice, you know.
+
+1:17:23.440 --> 1:17:25.440
+ Yes, yes.
+
+1:17:25.440 --> 1:17:26.440
+ So...
+
+1:17:26.440 --> 1:17:28.440
+ Yeah, here in America it's degrees of nice.
+
+1:17:28.440 --> 1:17:29.440
+ Yes.
+
+1:17:29.440 --> 1:17:31.440
+ It's all just degrees of nice, yeah.
+
+1:17:31.440 --> 1:17:44.440
+ Right, so as long as this does not become personal and it's really like, you know, a football game with its rules, that's great.
+
+1:17:44.440 --> 1:17:46.440
+ It's fun.
+
+1:17:46.440 --> 1:17:58.440
+ So, if you somehow find yourself in a position to ask one question of an oracle, like a genie, maybe a god, and you're guaranteed to get a clear answer,
+
+1:17:58.440 --> 1:18:00.440
+ what kind of question would you ask?
+
+1:18:00.440 --> 1:18:03.440
+ What would be the question you would ask?
+
+1:18:03.440 --> 1:18:09.440
+ In the spirit of our discussion, it could be, how could I become ten times more intelligent?
+
+1:18:09.440 --> 1:18:15.440
+ And so, but see, you only get a clear short answer.
+
+1:18:15.440 --> 1:18:18.440
+ So, do you think there's a clear short answer to that?
+
+1:18:18.440 --> 1:18:19.440
+ No.
+
+1:18:19.440 --> 1:18:22.440
+ And that's the answer you'll get.
+
+1:18:22.440 --> 1:18:23.440
+ Okay.
+
+1:18:23.440 --> 1:18:26.440
+ So, you've mentioned Flowers of Algernon.
+
+1:18:26.440 --> 1:18:27.440
+ Oh, yeah.
+
+1:18:27.440 --> 1:18:32.440
+ There's a story that inspired you in your childhood.
+
+1:18:32.440 --> 1:18:48.440
+ As this story of a mouse, a human achieving genius level intelligence, and then understanding what was happening while slowly becoming not intelligent again in this tragedy of gaining intelligence and losing intelligence.
+
+1:18:48.440 --> 1:18:59.440
+ Do you think in that spirit, in that story, do you think intelligence is a gift or a curse from the perspective of happiness and meaning of life?
+
+1:18:59.440 --> 1:19:10.440
+ You try to create an intelligent system that understands the universe, but on an individual level, the meaning of life, do you think intelligence is a gift?
+
+1:19:10.440 --> 1:19:16.440
+ It's a good question.
+
+1:19:16.440 --> 1:19:22.440
+ I don't know.
+
+1:19:22.440 --> 1:19:34.440
+ As one of the, as one people who consider the smartest people in the world, in some, in some dimension at the very least, what do you think?
+
+1:19:34.440 --> 1:19:35.440
+ I don't know.
+
+1:19:35.440 --> 1:19:39.440
+ It may be invariant to intelligence, let's agree of happiness.
+
+1:19:39.440 --> 1:19:43.440
+ It would be nice if it were.
+
+1:19:43.440 --> 1:19:44.440
+ That's the hope.
+
+1:19:44.440 --> 1:19:45.440
+ Yeah.
+
+1:19:45.440 --> 1:19:49.440
+ You could be smart and happy and clueless and happy.
+
+1:19:49.440 --> 1:19:51.440
+ Yeah.
+
+1:19:51.440 --> 1:19:56.440
+ As always on the discussion of the meaning of life is probably a good place to end.
+
+1:19:56.440 --> 1:19:58.440
+ Tomasso, thank you so much for talking today.
+
+1:19:58.440 --> 1:19:59.440
+ Thank you.
+
+1:19:59.440 --> 1:20:19.440
+ This was great.
+

vtt/episode_014_small.vtt

@@ -0,0 +1,4133 @@
+WEBVTT
+
+00:00.000 --> 00:02.240
+ The following is a conversation with Kyle Vogt.
+
+00:02.240 --> 00:05.120
+ He's the president and the CTO of Cruise Automation,
+
+00:05.120 --> 00:08.000
+ leading an effort to solve one of the biggest
+
+00:08.000 --> 00:10.880
+ robotics challenges of our time, vehicle automation.
+
+00:10.880 --> 00:13.120
+ He's a cofounder of two successful companies,
+
+00:13.120 --> 00:17.040
+ Twitch and Cruise, that have each sold for a billion dollars.
+
+00:17.040 --> 00:19.880
+ And he's a great example of the innovative spirit
+
+00:19.880 --> 00:22.160
+ that flourishes in Silicon Valley.
+
+00:22.160 --> 00:25.760
+ And now is facing an interesting and exciting challenge
+
+00:25.760 --> 00:30.040
+ of matching that spirit with the mass production
+
+00:30.040 --> 00:32.800
+ and the safety centered culture of a major automaker,
+
+00:32.800 --> 00:34.440
+ like General Motors.
+
+00:34.440 --> 00:36.520
+ This conversation is part of the MIT
+
+00:36.520 --> 00:38.560
+ Artificial General Intelligence series
+
+00:38.560 --> 00:41.040
+ and the Artificial Intelligence podcast.
+
+00:41.040 --> 00:44.840
+ If you enjoy it, please subscribe on YouTube, iTunes,
+
+00:44.840 --> 00:47.640
+ or simply connect with me on Twitter at Lex Friedman,
+
+00:47.640 --> 00:49.800
+ spelled F R I D.
+
+00:49.800 --> 00:53.480
+ And now here's my conversation with Kyle Vogt.
+
+00:53.480 --> 00:55.600
+ You grew up in Kansas, right?
+
+00:55.600 --> 00:58.080
+ Yeah, and I just saw that picture you had to hit know
+
+00:58.080 --> 01:00.400
+ there, so I'm a little bit worried about that now.
+
+01:00.400 --> 01:02.480
+ So in high school in Kansas City,
+
+01:02.480 --> 01:07.200
+ you joined Shawnee Mission North High School Robotics Team.
+
+01:07.200 --> 01:09.120
+ Now that wasn't your high school.
+
+01:09.120 --> 01:09.960
+ That's right.
+
+01:09.960 --> 01:13.920
+ That was the only high school in the area that had a teacher
+
+01:13.920 --> 01:16.080
+ who was willing to sponsor our first robotics team.
+
+01:16.080 --> 01:18.360
+ I was gonna troll you a little bit.
+
+01:18.360 --> 01:20.320
+ Jog your mouth a little bit with that kid.
+
+01:20.320 --> 01:22.880
+ I was trying to look super cool and intense.
+
+01:22.880 --> 01:23.720
+ You did?
+
+01:23.720 --> 01:25.680
+ Because this was BattleBots, this is serious business.
+
+01:25.680 --> 01:28.840
+ So we're standing there with a welded steel frame
+
+01:28.840 --> 01:30.240
+ and looking tough.
+
+01:30.240 --> 01:31.800
+ So go back there.
+
+01:31.800 --> 01:33.840
+ What does that drew you to robotics?
+
+01:33.840 --> 01:36.480
+ Well, I think, I've been trying to figure this out
+
+01:36.480 --> 01:37.920
+ for a while, but I've always liked building things
+
+01:37.920 --> 01:38.760
+ with Legos.
+
+01:38.760 --> 01:39.920
+ And when I was really, really young,
+
+01:39.920 --> 01:42.360
+ I wanted the Legos that had motors and other things.
+
+01:42.360 --> 01:44.840
+ And then, you know, Lego Mindstorms came out
+
+01:44.840 --> 01:48.280
+ and for the first time you could program Lego contraptions.
+
+01:48.280 --> 01:52.560
+ And I think things just sort of snowballed from that.
+
+01:52.560 --> 01:56.800
+ But I remember seeing, you know, the BattleBots TV show
+
+01:56.800 --> 01:59.320
+ on Comedy Central and thinking that is the coolest thing
+
+01:59.320 --> 02:01.200
+ in the world, I wanna be a part of that.
+
+02:01.200 --> 02:03.680
+ And not knowing a whole lot about how to build
+
+02:03.680 --> 02:06.880
+ these 200 pound fighting robots.
+
+02:06.880 --> 02:11.000
+ So I sort of obsessively poured over the internet forums
+
+02:11.000 --> 02:13.440
+ where all the creators for BattleBots would sort of hang out
+
+02:13.440 --> 02:16.120
+ and talk about, you know, document their build progress
+
+02:16.120 --> 02:17.120
+ and everything.
+
+02:17.120 --> 02:20.520
+ And I think I read, I must have read like, you know,
+
+02:20.520 --> 02:24.280
+ tens of thousands of forum posts from basically everything
+
+02:24.280 --> 02:26.400
+ that was out there on what these people were doing.
+
+02:26.400 --> 02:28.920
+ And eventually, like sort of triangulated how to put
+
+02:28.920 --> 02:33.040
+ some of these things together and ended up doing BattleBots,
+
+02:33.040 --> 02:34.800
+ which was, you know, it was like 13 or 14,
+
+02:34.800 --> 02:35.960
+ which was pretty awesome.
+
+02:35.960 --> 02:37.680
+ I'm not sure if the show's still running,
+
+02:37.680 --> 02:42.000
+ but so BattleBots is, there's not an artificial intelligence
+
+02:42.000 --> 02:44.200
+ component, it's remotely controlled.
+
+02:44.200 --> 02:46.720
+ And it's almost like a mechanical engineering challenge
+
+02:46.720 --> 02:49.560
+ of building things that can be broken.
+
+02:49.560 --> 02:50.680
+ They're radio controlled.
+
+02:50.680 --> 02:53.880
+ So, and I think that they allowed some limited form
+
+02:53.880 --> 02:56.600
+ of autonomy, but, you know, in a two minute match,
+
+02:56.600 --> 02:58.800
+ you're, in the way these things ran,
+
+02:58.800 --> 03:00.720
+ you're really doing yourself a disservice by trying
+
+03:00.720 --> 03:02.360
+ to automate it versus just, you know,
+
+03:02.360 --> 03:04.760
+ do the practical thing, which is drive it yourself.
+
+03:04.760 --> 03:06.960
+ And there's an entertainment aspect,
+
+03:06.960 --> 03:08.240
+ just going on YouTube.
+
+03:08.240 --> 03:11.200
+ There's like some of them wield an axe, some of them,
+
+03:11.200 --> 03:12.200
+ I mean, there's that fun.
+
+03:12.200 --> 03:13.760
+ So what drew you to that aspect?
+
+03:13.760 --> 03:15.400
+ Was it the mechanical engineering?
+
+03:15.400 --> 03:19.400
+ Was it the dream to create like Frankenstein
+
+03:19.400 --> 03:21.080
+ and sentient being?
+
+03:21.080 --> 03:23.960
+ Or was it just like the Lego, you like tinkering stuff?
+
+03:23.960 --> 03:26.000
+ I mean, that was just building something.
+
+03:26.000 --> 03:27.960
+ I think the idea of, you know,
+
+03:27.960 --> 03:30.920
+ this radio controlled machine that can do various things.
+
+03:30.920 --> 03:33.800
+ If it has like a weapon or something was pretty interesting.
+
+03:33.800 --> 03:36.440
+ I agree, it doesn't have the same appeal as, you know,
+
+03:36.440 --> 03:38.520
+ autonomous robots, which I, which I, you know,
+
+03:38.520 --> 03:40.320
+ sort of gravitated towards later on,
+
+03:40.320 --> 03:42.720
+ but it was definitely an engineering challenge
+
+03:42.720 --> 03:45.600
+ because everything you did in that competition
+
+03:45.600 --> 03:48.480
+ was pushing components to their limits.
+
+03:48.480 --> 03:52.960
+ So we would buy like these $40 DC motors
+
+03:52.960 --> 03:54.840
+ that came out of a winch,
+
+03:54.840 --> 03:57.280
+ like on the front of a pickup truck or something.
+
+03:57.280 --> 03:59.240
+ And we'd power the car with those
+
+03:59.240 --> 04:01.120
+ and we'd run them at like double or triple
+
+04:01.120 --> 04:02.440
+ their rated voltage.
+
+04:02.440 --> 04:04.160
+ So they immediately start overheating,
+
+04:04.160 --> 04:06.920
+ but for that two minute match, you can get, you know,
+
+04:06.920 --> 04:08.680
+ a significant increase in the power output
+
+04:08.680 --> 04:10.560
+ of those motors before they burn out.
+
+04:10.560 --> 04:12.760
+ And so you're doing the same thing for your battery packs,
+
+04:12.760 --> 04:14.360
+ all the materials in the system.
+
+04:14.360 --> 04:15.560
+ And I think there was something,
+
+04:15.560 --> 04:17.800
+ something intrinsically interesting
+
+04:17.800 --> 04:20.360
+ about just seeing like where things break.
+
+04:20.360 --> 04:23.360
+ And did you offline see where they break?
+
+04:23.360 --> 04:25.040
+ Did you take it to the testing point?
+
+04:25.040 --> 04:26.120
+ Like, how did you know two minutes?
+
+04:26.120 --> 04:29.680
+ Or was there a reckless, let's just go with it and see.
+
+04:29.680 --> 04:31.320
+ We weren't very good at battle bots.
+
+04:31.320 --> 04:34.200
+ We lost all of our matches the first round.
+
+04:34.200 --> 04:36.240
+ The one I built first,
+
+04:36.240 --> 04:38.120
+ both of them were these wedge shaped robots
+
+04:38.120 --> 04:39.800
+ because the wedge, even though it's sort of boring
+
+04:39.800 --> 04:41.240
+ to look at is extremely effective.
+
+04:41.240 --> 04:42.600
+ You drive towards another robot
+
+04:42.600 --> 04:44.720
+ and the front edge of it gets under them
+
+04:44.720 --> 04:46.760
+ and then they sort of flip over,
+
+04:46.760 --> 04:48.280
+ it's kind of like a door stopper.
+
+04:48.280 --> 04:51.920
+ And the first one had a pneumatic polished stainless steel
+
+04:51.920 --> 04:54.880
+ spike on the front that would shoot out about eight inches.
+
+04:54.880 --> 04:56.240
+ The purpose of which is what?
+
+04:56.240 --> 04:58.800
+ Pretty ineffective actually, but it looked cool.
+
+04:58.800 --> 05:00.880
+ And was it to help with the lift?
+
+05:00.880 --> 05:04.080
+ No, it was just to try to poke holes in the other robot.
+
+05:04.080 --> 05:05.960
+ And then the second time I did it,
+
+05:05.960 --> 05:09.560
+ which is the following, I think maybe 18 months later,
+
+05:09.560 --> 05:14.400
+ we had a titanium axe with a hardened steel tip on it
+
+05:14.400 --> 05:17.200
+ that was powered by a hydraulic cylinder,
+
+05:17.200 --> 05:20.400
+ which we were activating with liquid CO2,
+
+05:20.400 --> 05:23.880
+ which had its own set of problems.
+
+05:23.880 --> 05:26.320
+ So great, so that's kind of on the hardware side.
+
+05:26.320 --> 05:28.360
+ I mean, at a certain point,
+
+05:28.360 --> 05:31.240
+ there must have been born a fascination
+
+05:31.240 --> 05:32.440
+ on the software side.
+
+05:32.440 --> 05:35.520
+ So what was the first piece of code you've written?
+
+05:35.520 --> 05:38.600
+ If you didn't go back there, see what language was it?
+
+05:38.600 --> 05:40.600
+ What was it, was it EMAX, VAM?
+
+05:40.600 --> 05:44.640
+ Was it a more respectable, modern ID?
+
+05:44.640 --> 05:45.800
+ Do you remember any of this?
+
+05:45.800 --> 05:49.840
+ Yeah, well, I remember, I think maybe when I was in
+
+05:49.840 --> 05:52.440
+ third or fourth grade, I was at elementary school,
+
+05:52.440 --> 05:55.040
+ had a bunch of Apple II computers,
+
+05:55.040 --> 05:56.680
+ and we'd play games on those.
+
+05:56.680 --> 05:57.760
+ And I remember every once in a while,
+
+05:57.760 --> 06:01.320
+ something would crash or wouldn't start up correctly,
+
+06:01.320 --> 06:03.960
+ and it would dump you out to what I later learned
+
+06:03.960 --> 06:05.800
+ was like sort of a command prompt.
+
+06:05.800 --> 06:07.600
+ And my teacher would come over and type,
+
+06:07.600 --> 06:09.440
+ I actually remember this to this day for some reason,
+
+06:09.440 --> 06:12.160
+ like PR number six, or PR pound six,
+
+06:12.160 --> 06:13.840
+ which is peripheral six, which is the disk drive,
+
+06:13.840 --> 06:15.920
+ which would fire up the disk and load the program.
+
+06:15.920 --> 06:17.880
+ And I just remember thinking, wow, she's like a hacker,
+
+06:17.880 --> 06:20.760
+ like teach me these codes, these error codes,
+
+06:20.760 --> 06:22.720
+ that is what I called them at the time.
+
+06:22.720 --> 06:23.760
+ But she had no interest in that.
+
+06:23.760 --> 06:26.480
+ So it wasn't until I think about fifth grade
+
+06:26.480 --> 06:29.120
+ that I had a school where you could actually
+
+06:29.120 --> 06:30.600
+ go on these Apple II's and learn to program.
+
+06:30.600 --> 06:31.920
+ And so it was all in basic, you know,
+
+06:31.920 --> 06:34.240
+ where every line, you know, the line numbers are all,
+
+06:34.240 --> 06:35.640
+ or that every line is numbered,
+
+06:35.640 --> 06:38.000
+ and you have to like leave enough space
+
+06:38.000 --> 06:40.760
+ between the numbers so that if you want to tweak your code,
+
+06:40.760 --> 06:42.600
+ you go back and if the first line was 10
+
+06:42.600 --> 06:44.680
+ and the second line is 20, now you have to go back
+
+06:44.680 --> 06:45.640
+ and insert 15.
+
+06:45.640 --> 06:47.960
+ And if you need to add code in front of that,
+
+06:47.960 --> 06:49.720
+ you know, 11 or 12, and you hope you don't run out
+
+06:49.720 --> 06:51.880
+ of line numbers and have to redo the whole thing.
+
+06:51.880 --> 06:53.240
+ And there's go to statements?
+
+06:53.240 --> 06:56.920
+ Yeah, go to and is very basic, maybe hence the name,
+
+06:56.920 --> 06:58.200
+ but a lot of fun.
+
+06:58.200 --> 07:00.800
+ And that was like, that was, you know,
+
+07:00.800 --> 07:02.600
+ that's when, you know, when you first program,
+
+07:02.600 --> 07:03.560
+ you see the magic of it.
+
+07:03.560 --> 07:06.640
+ It's like, just like this world opens up with,
+
+07:06.640 --> 07:08.200
+ you know, endless possibilities for the things
+
+07:08.200 --> 07:10.600
+ you could build or accomplish with that computer.
+
+07:10.600 --> 07:13.400
+ So you got the bug then, so even starting with basic
+
+07:13.400 --> 07:16.720
+ and then what, C++ throughout, what did you,
+
+07:16.720 --> 07:18.200
+ was there a computer programming,
+
+07:18.200 --> 07:19.880
+ computer science classes in high school?
+
+07:19.880 --> 07:22.680
+ Not, not where I went, so it was self taught,
+
+07:22.680 --> 07:24.640
+ but I did a lot of programming.
+
+07:24.640 --> 07:28.560
+ The thing that, you know, sort of pushed me in the path
+
+07:28.560 --> 07:30.600
+ of eventually working on self driving cars
+
+07:30.600 --> 07:33.280
+ is actually one of these really long trips
+
+07:33.280 --> 07:38.000
+ driving from my house in Kansas to, I think, Las Vegas,
+
+07:38.000 --> 07:39.480
+ where we did the BattleBots competition.
+
+07:39.480 --> 07:42.760
+ And I had just gotten my, I think my learners permit
+
+07:42.760 --> 07:45.080
+ or early drivers permit.
+
+07:45.080 --> 07:48.280
+ And so I was driving this, you know, 10 hour stretch
+
+07:48.280 --> 07:50.600
+ across Western Kansas where it's just,
+
+07:50.600 --> 07:51.800
+ you're going straight on a highway
+
+07:51.800 --> 07:53.640
+ and it is mind numbingly boring.
+
+07:53.640 --> 07:54.960
+ And I remember thinking even then
+
+07:54.960 --> 07:58.080
+ with my sort of mediocre programming background
+
+07:58.080 --> 08:00.040
+ that this is something that a computer can do, right?
+
+08:00.040 --> 08:01.440
+ Let's take a picture of the road,
+
+08:01.440 --> 08:02.880
+ let's find the yellow lane markers
+
+08:02.880 --> 08:04.880
+ and, you know, steer the wheel.
+
+08:04.880 --> 08:06.600
+ And, you know, later I'd come to realize
+
+08:06.600 --> 08:09.800
+ this had been done, you know, since the 80s
+
+08:09.800 --> 08:12.760
+ or the 70s or even earlier, but I still wanted to do it.
+
+08:12.760 --> 08:14.840
+ And sort of immediately after that trip,
+
+08:14.840 --> 08:16.280
+ switched from sort of BattleBots,
+
+08:16.280 --> 08:18.640
+ which is more radio controlled machines
+
+08:18.640 --> 08:21.800
+ to thinking about building, you know,
+
+08:21.800 --> 08:23.600
+ autonomous vehicles of some scale,
+
+08:23.600 --> 08:25.080
+ start off with really small electric ones
+
+08:25.080 --> 08:28.280
+ and then, you know, progress to what we're doing now.
+
+08:28.280 --> 08:30.040
+ So what was your view of artificial intelligence
+
+08:30.040 --> 08:30.880
+ at that point?
+
+08:30.880 --> 08:31.880
+ What did you think?
+
+08:31.880 --> 08:35.040
+ So this is before there's been waves
+
+08:35.040 --> 08:36.680
+ in artificial intelligence, right?
+
+08:36.680 --> 08:39.480
+ The current wave with deep learning
+
+08:39.480 --> 08:41.760
+ makes people believe that you can solve
+
+08:41.760 --> 08:43.520
+ in a really rich, deep way,
+
+08:43.520 --> 08:46.200
+ the computer vision perception problem.
+
+08:46.200 --> 08:51.200
+ But like before the deep learning craze,
+
+08:51.320 --> 08:52.800
+ you know, how do you think about
+
+08:52.800 --> 08:55.320
+ how would you even go about building a thing
+
+08:55.320 --> 08:56.920
+ that perceives itself in the world,
+
+08:56.920 --> 08:59.160
+ localize itself in the world, moves around the world?
+
+08:59.160 --> 09:00.360
+ Like when you were younger, I mean,
+
+09:00.360 --> 09:02.120
+ as what was your thinking about it?
+
+09:02.120 --> 09:03.960
+ Well, prior to deep neural networks
+
+09:03.960 --> 09:05.360
+ or convolutional neural nets,
+
+09:05.360 --> 09:06.520
+ these modern techniques we have,
+
+09:06.520 --> 09:09.040
+ or at least ones that are in use today,
+
+09:09.040 --> 09:10.280
+ it was all heuristic space.
+
+09:10.280 --> 09:12.920
+ And so like old school image processing,
+
+09:12.920 --> 09:15.040
+ and I think extracting, you know,
+
+09:15.040 --> 09:18.000
+ yellow lane markers out of an image of a road
+
+09:18.000 --> 09:21.160
+ is one of the problems that lends itself
+
+09:21.160 --> 09:23.760
+ reasonably well to those heuristic base methods, you know,
+
+09:23.760 --> 09:26.760
+ like just do a threshold on the color yellow
+
+09:26.760 --> 09:28.520
+ and then try to fit some lines to that
+
+09:28.520 --> 09:30.320
+ using a huff transform or something
+
+09:30.320 --> 09:32.280
+ and then go from there.
+
+09:32.280 --> 09:34.800
+ Traffic light detection and stop sign detection,
+
+09:34.800 --> 09:35.920
+ red, yellow, green.
+
+09:35.920 --> 09:38.160
+ And I think you can, you could,
+
+09:38.160 --> 09:39.840
+ I mean, if you wanted to do a full,
+
+09:39.840 --> 09:41.960
+ I was just trying to make something that would stay
+
+09:41.960 --> 09:43.520
+ in between the lanes on a highway,
+
+09:43.520 --> 09:44.960
+ but if you wanted to do the full,
+
+09:46.920 --> 09:48.960
+ the full, you know, set of capabilities
+
+09:48.960 --> 09:50.520
+ needed for a driverless car,
+
+09:50.520 --> 09:53.360
+ I think you could, and we've done this at cruise,
+
+09:53.360 --> 09:54.440
+ you know, in the very first days,
+
+09:54.440 --> 09:56.320
+ you can start off with a really simple,
+
+09:56.320 --> 09:58.000
+ you know, human written heuristic
+
+09:58.000 --> 09:59.800
+ just to get the scaffolding in place
+
+09:59.800 --> 10:01.720
+ for your system, traffic light detection,
+
+10:01.720 --> 10:02.960
+ probably a really simple, you know,
+
+10:02.960 --> 10:04.760
+ color thresholding on day one
+
+10:04.760 --> 10:06.520
+ just to get the system up and running
+
+10:06.520 --> 10:08.640
+ before you migrate to, you know,
+
+10:08.640 --> 10:11.080
+ a deep learning based technique or something else.
+
+10:11.080 --> 10:12.800
+ And, you know, back in, when I was doing this,
+
+10:12.800 --> 10:15.120
+ my first one, it was on a Pentium 203,
+
+10:15.120 --> 10:17.840
+ 233 megahertz computer in it.
+
+10:17.840 --> 10:19.920
+ And I think I wrote the first version in basic,
+
+10:19.920 --> 10:21.600
+ which is like an interpreted language.
+
+10:21.600 --> 10:23.760
+ It's extremely slow because that's the thing
+
+10:23.760 --> 10:24.800
+ I knew at the time.
+
+10:24.800 --> 10:27.840
+ And so there was no, no chance at all of using,
+
+10:27.840 --> 10:30.440
+ there's no computational power to do
+
+10:30.440 --> 10:33.480
+ any sort of reasonable deep nets like you have today.
+
+10:33.480 --> 10:35.360
+ So I don't know what kids these days are doing.
+
+10:35.360 --> 10:37.920
+ Are kids these days, you know, at age 13
+
+10:37.920 --> 10:39.360
+ using neural networks in their garage?
+
+10:39.360 --> 10:40.200
+ I mean, that would be awesome.
+
+10:40.200 --> 10:43.040
+ I get emails all the time from, you know,
+
+10:43.040 --> 10:46.160
+ like 11, 12 year olds saying, I'm having, you know,
+
+10:46.160 --> 10:48.760
+ I'm trying to follow this TensorFlow tutorial
+
+10:48.760 --> 10:50.800
+ and I'm having this problem.
+
+10:50.800 --> 10:55.800
+ And the general approach in the deep learning community
+
+10:55.800 --> 11:00.200
+ is of extreme optimism of, as opposed to,
+
+11:00.200 --> 11:02.000
+ you mentioned like heuristics, you can,
+
+11:02.000 --> 11:04.800
+ you can, you can separate the autonomous driving problem
+
+11:04.800 --> 11:07.520
+ into modules and try to solve it sort of rigorously,
+
+11:07.520 --> 11:09.040
+ where you can just do it end to end.
+
+11:09.040 --> 11:11.840
+ And most people just kind of love the idea that,
+
+11:11.840 --> 11:13.360
+ you know, us humans do it end to end,
+
+11:13.360 --> 11:15.360
+ we just perceive and act.
+
+11:15.360 --> 11:17.040
+ We should be able to use that,
+
+11:17.040 --> 11:18.720
+ do the same kind of thing with your own nets.
+
+11:18.720 --> 11:20.920
+ And that, that kind of thinking,
+
+11:20.920 --> 11:22.840
+ you don't want to criticize that kind of thinking
+
+11:22.840 --> 11:24.640
+ because eventually they will be right.
+
+11:24.640 --> 11:26.360
+ Yeah. And so it's exciting.
+
+11:26.360 --> 11:28.720
+ And especially when they're younger to explore that
+
+11:28.720 --> 11:30.640
+ is a really exciting approach.
+
+11:30.640 --> 11:35.480
+ But yeah, it's, it's changed the, the language,
+
+11:35.480 --> 11:37.240
+ the kind of stuff you're tinkering with.
+
+11:37.240 --> 11:40.920
+ It's kind of exciting to see when these teenagers grow up.
+
+11:40.920 --> 11:43.760
+ Yeah, I can only imagine if you, if your starting point
+
+11:43.760 --> 11:46.720
+ is, you know, Python and TensorFlow at age 13,
+
+11:46.720 --> 11:47.800
+ where you end up, you know,
+
+11:47.800 --> 11:51.040
+ after 10 or 15 years of that, that's, that's pretty cool.
+
+11:51.040 --> 11:53.760
+ Because of GitHub, because the state tools
+
+11:53.760 --> 11:55.440
+ for solving most of the major problems
+
+11:55.440 --> 11:56.920
+ that are artificial intelligence
+
+11:56.920 --> 12:00.240
+ are within a few lines of code for most kids.
+
+12:00.240 --> 12:02.280
+ And that's incredible to think about,
+
+12:02.280 --> 12:04.280
+ also on the entrepreneurial side.
+
+12:04.280 --> 12:08.520
+ And, and, and at that point, was there any thought
+
+12:08.520 --> 12:11.960
+ about entrepreneurship before you came to college
+
+12:11.960 --> 12:15.160
+ is sort of doing your building this into a thing
+
+12:15.160 --> 12:17.800
+ that impacts the world on a large scale?
+
+12:17.800 --> 12:19.840
+ Yeah, I've always wanted to start a company.
+
+12:19.840 --> 12:22.600
+ I think that's, you know, just a cool concept
+
+12:22.600 --> 12:25.240
+ of creating something and exchanging it
+
+12:25.240 --> 12:28.360
+ for value or creating value, I guess.
+
+12:28.360 --> 12:31.120
+ So in high school, I was, I was trying to build like,
+
+12:31.120 --> 12:33.600
+ you know, servo motor drivers, little circuit boards
+
+12:33.600 --> 12:36.920
+ and sell them online or other, other things like that.
+
+12:36.920 --> 12:40.320
+ And certainly knew at some point I wanted to do a startup,
+
+12:40.320 --> 12:42.840
+ but it wasn't really, I'd say until college until I felt
+
+12:42.840 --> 12:46.720
+ like I had the, I guess the right combination
+
+12:46.720 --> 12:48.960
+ of the environment, the smart people around you
+
+12:48.960 --> 12:52.360
+ and some free time and a lot of free time at MIT.
+
+12:52.360 --> 12:55.800
+ So you came to MIT as an undergrad 2004.
+
+12:55.800 --> 12:57.080
+ That's right.
+
+12:57.080 --> 12:59.040
+ And that's when the first DARPA Grand Challenge
+
+12:59.040 --> 12:59.880
+ was happening.
+
+12:59.880 --> 13:00.720
+ Yeah.
+
+13:00.720 --> 13:03.360
+ The timing of that is beautifully poetic.
+
+13:03.360 --> 13:05.680
+ So how'd you get yourself involved in that one?
+
+13:05.680 --> 13:07.080
+ Originally there wasn't a
+
+13:07.080 --> 13:07.920
+ Official entry?
+
+13:07.920 --> 13:09.520
+ Yeah, faculty sponsored thing.
+
+13:09.520 --> 13:12.760
+ And so a bunch of undergrads, myself included,
+
+13:12.760 --> 13:14.160
+ started meeting and got together
+
+13:14.160 --> 13:17.800
+ and tried to, to haggle together some sponsorships.
+
+13:17.800 --> 13:20.120
+ We got a vehicle donated, a bunch of sensors
+
+13:20.120 --> 13:21.600
+ and tried to put something together.
+
+13:21.600 --> 13:24.640
+ And so we had, our team was probably mostly freshmen
+
+13:24.640 --> 13:26.800
+ and sophomores, you know, which, which was not really
+
+13:26.800 --> 13:30.960
+ a fair, fair fight against maybe the, you know, postdoc
+
+13:30.960 --> 13:32.840
+ and faculty led teams from other schools.
+
+13:32.840 --> 13:35.000
+ But we, we got something up and running.
+
+13:35.000 --> 13:37.400
+ We had our vehicle drive by wire and, you know,
+
+13:37.400 --> 13:42.440
+ very, very basic control and things, but on the day
+
+13:42.440 --> 13:46.800
+ of the qualifying, sort of pre qualifying round,
+
+13:46.800 --> 13:50.840
+ the one and only steering motor that we had purchased,
+
+13:50.840 --> 13:52.600
+ the thing that we had, you know, retrofitted to turn
+
+13:52.600 --> 13:55.760
+ the steering wheel on the truck died.
+
+13:55.760 --> 13:58.440
+ And so our vehicle was just dead in the water, couldn't steer.
+
+13:58.440 --> 13:59.880
+ So we didn't make it very far.
+
+13:59.880 --> 14:00.920
+ On the hardware side.
+
+14:00.920 --> 14:03.000
+ So was there a software component?
+
+14:03.000 --> 14:06.200
+ Was there, like, how did your view of autonomous vehicles
+
+14:06.200 --> 14:08.040
+ in terms of artificial intelligence
+
+14:09.520 --> 14:10.720
+ evolve in this moment?
+
+14:10.720 --> 14:12.400
+ I mean, you know, like you said,
+
+14:12.400 --> 14:14.080
+ from the 80s has been autonomous vehicles,
+
+14:14.080 --> 14:16.720
+ but really that was the birth of the modern wave.
+
+14:16.720 --> 14:20.080
+ The, the thing that captivated everyone's imagination
+
+14:20.080 --> 14:21.520
+ that we can actually do this.
+
+14:21.520 --> 14:26.000
+ So how, were you captivated in that way?
+
+14:26.000 --> 14:27.600
+ So how did your view of autonomous vehicles
+
+14:27.600 --> 14:29.000
+ change at that point?
+
+14:29.000 --> 14:33.760
+ I'd say at that point in time, it was, it was a curiosity
+
+14:33.760 --> 14:35.840
+ as in like, is this really possible?
+
+14:35.840 --> 14:38.440
+ And I think that was generally the spirit
+
+14:38.440 --> 14:43.280
+ and the purpose of that original DARPA Grand Challenge,
+
+14:43.280 --> 14:45.520
+ which was to just get a whole bunch
+
+14:45.520 --> 14:48.680
+ of really brilliant people exploring the space
+
+14:48.680 --> 14:49.880
+ and pushing the limits.
+
+14:49.880 --> 14:51.960
+ And, and I think like to this day,
+
+14:51.960 --> 14:54.160
+ that DARPA challenge with its, you know,
+
+14:54.160 --> 14:57.120
+ million dollar prize pool was probably one
+
+14:57.120 --> 15:00.840
+ of the most effective, you know, uses of taxpayer money,
+
+15:00.840 --> 15:03.320
+ dollar for dollar that I've seen, you know,
+
+15:03.320 --> 15:06.720
+ because that, that small sort of initiative
+
+15:06.720 --> 15:10.440
+ that DARPA put put out sort of, in my view,
+
+15:10.440 --> 15:12.560
+ was the catalyst or the tipping point
+
+15:12.560 --> 15:14.800
+ for this, this whole next wave
+
+15:14.800 --> 15:16.120
+ of autonomous vehicle development.
+
+15:16.120 --> 15:17.160
+ So that was pretty cool.
+
+15:17.160 --> 15:20.240
+ So let me jump around a little bit on that point.
+
+15:20.240 --> 15:23.240
+ They also did the urban challenge where it was in the city,
+
+15:23.240 --> 15:25.920
+ but it was very artificial and there's no pedestrians
+
+15:25.920 --> 15:27.640
+ and there's very little human involvement
+
+15:27.640 --> 15:30.480
+ except a few professional drivers.
+
+15:30.480 --> 15:31.640
+ Yeah.
+
+15:31.640 --> 15:33.560
+ Do you think there's room, and then there was
+
+15:33.560 --> 15:35.360
+ the robotics challenge with human robots?
+
+15:35.360 --> 15:36.200
+ Right.
+
+15:36.200 --> 15:38.720
+ So in your now role as looking at this,
+
+15:38.720 --> 15:41.640
+ you're trying to solve one of the, you know,
+
+15:41.640 --> 15:43.120
+ autonomous driving, one of the harder,
+
+15:43.120 --> 15:45.480
+ more difficult places in San Francisco.
+
+15:45.480 --> 15:47.320
+ Is there a role for DARPA to step in
+
+15:47.320 --> 15:49.680
+ to also kind of help out, like,
+
+15:49.680 --> 15:54.000
+ challenge with new ideas, specifically pedestrians
+
+15:54.000 --> 15:55.880
+ and so on, all these kinds of interesting things?
+
+15:55.880 --> 15:57.680
+ Well, I haven't thought about it from that perspective.
+
+15:57.680 --> 15:59.280
+ Is there anything DARPA could do today
+
+15:59.280 --> 16:00.680
+ to further accelerate things?
+
+16:00.680 --> 16:04.880
+ And I would say my instinct is that that's maybe not
+
+16:04.880 --> 16:07.040
+ the highest and best use of their resources in time
+
+16:07.040 --> 16:10.640
+ because, like, kick starting and spinning up the flywheel
+
+16:10.640 --> 16:12.720
+ is I think what they did in this case
+
+16:12.720 --> 16:14.200
+ for very, very little money.
+
+16:14.200 --> 16:15.800
+ But today this has become,
+
+16:16.880 --> 16:19.000
+ this has become, like, commercially interesting
+
+16:19.000 --> 16:20.680
+ to very large companies and the amount of money
+
+16:20.680 --> 16:23.040
+ going into it and the amount of people,
+
+16:23.040 --> 16:24.840
+ like, going through your class and learning
+
+16:24.840 --> 16:27.200
+ about these things and developing these skills
+
+16:27.200 --> 16:29.120
+ is just, you know, orders of magnitude
+
+16:29.120 --> 16:30.840
+ more than it was back then.
+
+16:30.840 --> 16:33.080
+ And so there's enough momentum and inertia
+
+16:33.080 --> 16:36.520
+ and energy and investment dollars into this space right now
+
+16:36.520 --> 16:39.960
+ that I don't, I don't, I think they're,
+
+16:39.960 --> 16:42.200
+ I think they're, they can just say mission accomplished
+
+16:42.200 --> 16:44.320
+ and move on to the next area of technology
+
+16:44.320 --> 16:45.360
+ that needs help.
+
+16:46.280 --> 16:49.120
+ So then stepping back to MIT,
+
+16:49.120 --> 16:50.880
+ you left MIT Junior Junior year,
+
+16:50.880 --> 16:53.080
+ what was that decision like?
+
+16:53.080 --> 16:55.680
+ As I said, I always wanted to do a company
+
+16:55.680 --> 16:59.080
+ or start a company and this opportunity landed in my lap
+
+16:59.080 --> 17:01.960
+ which was a couple of guys from Yale
+
+17:01.960 --> 17:04.240
+ were starting a new company and I Googled them
+
+17:04.240 --> 17:06.720
+ and found that they had started a company previously
+
+17:06.720 --> 17:10.640
+ and sold it actually on eBay for about a quarter million bucks
+
+17:10.640 --> 17:12.880
+ which was a pretty interesting story.
+
+17:12.880 --> 17:15.760
+ But so I thought to myself, these guys are, you know,
+
+17:15.760 --> 17:19.080
+ rock star entrepreneurs, they've done this before,
+
+17:19.080 --> 17:20.720
+ they must be driving around in Ferraris
+
+17:20.720 --> 17:22.320
+ because they sold their company.
+
+17:23.320 --> 17:26.000
+ And, you know, I thought I could learn a lot from them.
+
+17:26.000 --> 17:28.320
+ So I teamed up with those guys and, you know,
+
+17:28.320 --> 17:32.000
+ went out during, went out to California during IAP
+
+17:32.000 --> 17:36.440
+ which is MIT's month off on one way ticket
+
+17:36.440 --> 17:38.040
+ and basically never went back.
+
+17:38.040 --> 17:39.280
+ We were having so much fun,
+
+17:39.280 --> 17:42.040
+ we felt like we were building something and creating something
+
+17:42.040 --> 17:44.440
+ and it was gonna be interesting that, you know,
+
+17:44.440 --> 17:46.800
+ I was just all in and got completely hooked
+
+17:46.800 --> 17:49.640
+ and that business was Justin TV
+
+17:49.640 --> 17:52.400
+ which is originally a reality show about a guy named Justin
+
+17:53.720 --> 17:57.120
+ which morphed into a live video streaming platform
+
+17:57.120 --> 18:00.320
+ which then morphed into what is Twitch today.
+
+18:00.320 --> 18:03.720
+ So that was quite an unexpected journey.
+
+18:04.720 --> 18:07.000
+ So no regrets?
+
+18:07.000 --> 18:07.840
+ No.
+
+18:07.840 --> 18:09.120
+ Looking back, it was just an obvious,
+
+18:09.120 --> 18:10.720
+ I mean, one way ticket.
+
+18:10.720 --> 18:12.760
+ I mean, if we just pause on that for a second,
+
+18:12.760 --> 18:17.680
+ there was no, how did you know these were the right guys?
+
+18:17.680 --> 18:19.520
+ This is the right decision.
+
+18:19.520 --> 18:22.640
+ You didn't think it was just follow the heart kind of thing?
+
+18:22.640 --> 18:24.520
+ Well, I didn't know, but, you know,
+
+18:24.520 --> 18:26.520
+ just trying something for a month during IAP
+
+18:26.520 --> 18:28.240
+ seems pretty low risk, right?
+
+18:28.240 --> 18:30.760
+ And then, you know, well, maybe I'll take a semester off.
+
+18:30.760 --> 18:32.280
+ MIT's pretty flexible about that.
+
+18:32.280 --> 18:33.840
+ You can always go back, right?
+
+18:33.840 --> 18:35.680
+ And then after two or three cycles of that,
+
+18:35.680 --> 18:36.960
+ I eventually threw in the towel.
+
+18:36.960 --> 18:41.920
+ But, you know, I think it's, I guess in that case,
+
+18:41.920 --> 18:44.880
+ I felt like I could always hit the undo button if I had to.
+
+18:44.880 --> 18:45.720
+ Right.
+
+18:45.720 --> 18:49.600
+ But nevertheless, from when you look in retrospect,
+
+18:49.600 --> 18:51.680
+ I mean, it seems like a brave decision.
+
+18:51.680 --> 18:53.200
+ You know, it would be difficult
+
+18:53.200 --> 18:54.320
+ for a lot of people to make.
+
+18:54.320 --> 18:55.440
+ It wasn't as popular.
+
+18:55.440 --> 18:58.120
+ I'd say that the general, you know,
+
+18:58.120 --> 19:01.480
+ flux of people out of MIT at the time was mostly
+
+19:01.480 --> 19:04.120
+ into, you know, finance or consulting jobs
+
+19:04.120 --> 19:05.720
+ in Boston or New York.
+
+19:05.720 --> 19:07.840
+ And very few people were going to California
+
+19:07.840 --> 19:09.080
+ to start companies.
+
+19:09.080 --> 19:12.240
+ But today, I'd say that's probably inverted,
+
+19:12.240 --> 19:15.400
+ which is just a sign of a sign of the times, I guess.
+
+19:15.400 --> 19:16.080
+ Yeah.
+
+19:16.080 --> 19:21.560
+ So there's a story about midnight of March 18, 2007,
+
+19:21.560 --> 19:25.720
+ where TechCrunch, I guess, announced Justin TV earlier
+
+19:25.720 --> 19:29.080
+ than it was supposed to a few hours.
+
+19:29.080 --> 19:30.360
+ The site didn't work.
+
+19:30.360 --> 19:32.520
+ I don't know if any of this is true, you can tell me.
+
+19:32.520 --> 19:36.200
+ And you and one of the folks at Justin TV,
+
+19:36.200 --> 19:39.240
+ Emma Shear, coded through the night.
+
+19:39.240 --> 19:41.440
+ Can you take me through that experience?
+
+19:41.440 --> 19:47.160
+ So let me say a few nice things that the article I read quoted
+
+19:47.160 --> 19:49.600
+ Justin Khan said that you were known for bureau coding
+
+19:49.600 --> 19:53.520
+ through problems and being a creative genius.
+
+19:53.520 --> 19:59.440
+ So on that night, what was going through your head?
+
+19:59.440 --> 20:01.400
+ Or maybe I put another way, how do you
+
+20:01.400 --> 20:02.520
+ solve these problems?
+
+20:02.520 --> 20:05.480
+ What's your approach to solving these kind of problems
+
+20:05.480 --> 20:07.080
+ where the line between success and failure
+
+20:07.080 --> 20:09.680
+ seems to be pretty thin?
+
+20:09.680 --> 20:10.680
+ That's a good question.
+
+20:10.680 --> 20:13.400
+ Well, first of all, that's nice of Justin to say that.
+
+20:13.400 --> 20:16.880
+ I think I would have been maybe 21 years old then
+
+20:16.880 --> 20:18.800
+ and not very experienced at programming.
+
+20:18.800 --> 20:22.680
+ But as with everything in a startup,
+
+20:22.680 --> 20:24.720
+ you're sort of racing against the clock.
+
+20:24.720 --> 20:27.320
+ And so our plan was the second we
+
+20:27.320 --> 20:32.600
+ had this live streaming camera backpack up and running
+
+20:32.600 --> 20:33.600
+ where Justin could wear it.
+
+20:33.600 --> 20:35.320
+ And no matter where he went in the city,
+
+20:35.320 --> 20:36.400
+ it would be streaming live video.
+
+20:36.400 --> 20:37.960
+ And this is even before the iPhones,
+
+20:37.960 --> 20:40.880
+ this is like hard to do back then.
+
+20:40.880 --> 20:41.800
+ We would launch.
+
+20:41.800 --> 20:45.160
+ And so we thought we were there and the backpack was working.
+
+20:45.160 --> 20:47.080
+ And then we sent out all the emails
+
+20:47.080 --> 20:49.920
+ to launch the company and do the press thing.
+
+20:49.920 --> 20:53.000
+ And then we weren't quite actually there.
+
+20:53.000 --> 20:55.880
+ And then we thought, oh, well, they're
+
+20:55.880 --> 21:00.160
+ not going to announce it until maybe 10 AM the next morning.
+
+21:00.160 --> 21:01.880
+ And it's, I don't know, it's 5 PM now.
+
+21:01.880 --> 21:03.640
+ So how many hours do we have left?
+
+21:03.640 --> 21:08.000
+ What is that, like 17 hours to go?
+
+21:08.000 --> 21:10.440
+ And that was going to be fine.
+
+21:10.440 --> 21:11.440
+ Was the problem obvious?
+
+21:11.440 --> 21:13.280
+ Did you understand what could possibly be?
+
+21:13.280 --> 21:16.520
+ Like how complicated was the system at that point?
+
+21:16.520 --> 21:18.840
+ It was pretty messy.
+
+21:18.840 --> 21:22.760
+ So to get a live video feed that looked decent working
+
+21:22.760 --> 21:25.680
+ from anywhere in San Francisco, I
+
+21:25.680 --> 21:28.600
+ put together this system where we had like three or four
+
+21:28.600 --> 21:29.880
+ cell phone data modems.
+
+21:29.880 --> 21:32.200
+ And they were like, we take the video stream
+
+21:32.200 --> 21:35.600
+ and sort of spray it across these three or four modems
+
+21:35.600 --> 21:38.080
+ and then try to catch all the packets on the other side
+
+21:38.080 --> 21:39.480
+ with unreliable cell phone networks.
+
+21:39.480 --> 21:41.080
+ Pretty low level networking.
+
+21:41.080 --> 21:41.720
+ Yeah.
+
+21:41.720 --> 21:44.760
+ And putting these sort of protocols
+
+21:44.760 --> 21:47.560
+ on top of all that to reassemble and reorder the packets
+
+21:47.560 --> 21:49.720
+ and have time buffers and error correction
+
+21:49.720 --> 21:50.960
+ and all that kind of stuff.
+
+21:50.960 --> 21:53.960
+ And the night before, it was just
+
+21:53.960 --> 21:56.280
+ staticky. Every once in a while, the image would go
+
+21:56.280 --> 21:59.640
+ staticky and there would be this horrible like screeching
+
+21:59.640 --> 22:02.080
+ audio noise because the audio was also corrupted.
+
+22:02.080 --> 22:04.600
+ And this would happen like every five to 10 minutes or so.
+
+22:04.600 --> 22:08.080
+ And it was a really, you know, off of putting to the viewers.
+
+22:08.080 --> 22:08.880
+ Yeah.
+
+22:08.880 --> 22:10.200
+ How do you tackle that problem?
+
+22:10.200 --> 22:13.280
+ What was the, you're just freaking out behind a computer.
+
+22:13.280 --> 22:16.880
+ There's the word, are there other folks working on this problem?
+
+22:16.880 --> 22:18.120
+ Like were you behind a whiteboard?
+
+22:18.120 --> 22:22.000
+ Were you doing a hair coding?
+
+22:22.000 --> 22:23.840
+ Yeah, it's a little lonely because there's four of us
+
+22:23.840 --> 22:26.880
+ working on the company and only two people really wrote code.
+
+22:26.880 --> 22:29.200
+ And Emmett wrote the website in the chat system
+
+22:29.200 --> 22:32.400
+ and I wrote the software for this video streaming device
+
+22:32.400 --> 22:34.280
+ and video server.
+
+22:34.280 --> 22:36.240
+ And so, you know, it was my sole responsibility
+
+22:36.240 --> 22:37.320
+ to figure that out.
+
+22:37.320 --> 22:39.440
+ And I think it's those, you know,
+
+22:39.440 --> 22:42.200
+ setting deadlines, trying to move quickly and everything
+
+22:42.200 --> 22:44.200
+ where you're in that moment of intense pressure
+
+22:44.200 --> 22:46.960
+ that sometimes people do their best and most interesting work.
+
+22:46.960 --> 22:48.800
+ And so even though that was a terrible moment,
+
+22:48.800 --> 22:50.760
+ I look back on it fondly because that's like, you know,
+
+22:50.760 --> 22:54.720
+ that's one of those character defining moments, I think.
+
+22:54.720 --> 22:59.480
+ So in 2013, October, you founded Cruise Automation.
+
+22:59.480 --> 23:00.200
+ Yeah.
+
+23:00.200 --> 23:04.200
+ So progressing forward, another exceptionally successful
+
+23:04.200 --> 23:09.920
+ company was acquired by GM in 2016 for $1 billion.
+
+23:09.920 --> 23:14.120
+ But in October 2013, what was on your mind?
+
+23:14.120 --> 23:16.360
+ What was the plan?
+
+23:16.360 --> 23:19.840
+ How does one seriously start to tackle
+
+23:19.840 --> 23:22.800
+ one of the hardest robotics, most important impact
+
+23:22.800 --> 23:24.960
+ for robotics problems of our age?
+
+23:24.960 --> 23:28.760
+ After going through Twitch, Twitch was,
+
+23:28.760 --> 23:31.480
+ and is today pretty successful.
+
+23:31.480 --> 23:36.880
+ But the work was, the result was entertainment mostly.
+
+23:36.880 --> 23:39.840
+ Like the better the product was, the more we would entertain
+
+23:39.840 --> 23:42.760
+ people and then, you know, make money on the ad revenues
+
+23:42.760 --> 23:43.760
+ and other things.
+
+23:43.760 --> 23:45.000
+ And that was a good thing.
+
+23:45.000 --> 23:46.320
+ It felt good to entertain people.
+
+23:46.320 --> 23:49.120
+ But I figured like, you know, what is really the point
+
+23:49.120 --> 23:51.120
+ of becoming a really good engineer
+
+23:51.120 --> 23:53.160
+ and developing these skills other than, you know,
+
+23:53.160 --> 23:53.960
+ my own enjoyment.
+
+23:53.960 --> 23:55.760
+ And I realized I wanted something that scratched
+
+23:55.760 --> 23:57.680
+ more of an existential itch, like something
+
+23:57.680 --> 23:59.440
+ that truly matters.
+
+23:59.440 --> 24:03.680
+ And so I basically made this list of requirements
+
+24:03.680 --> 24:06.160
+ for a new, if I was going to do another company.
+
+24:06.160 --> 24:08.000
+ And the one thing I knew in the back of my head
+
+24:08.000 --> 24:12.320
+ that Twitch took like eight years to become successful.
+
+24:12.320 --> 24:14.880
+ And so whatever I do, I better be willing to commit,
+
+24:14.880 --> 24:17.000
+ you know, at least 10 years to something.
+
+24:17.000 --> 24:20.400
+ And when you think about things from that perspective,
+
+24:20.400 --> 24:21.760
+ you certainly, I think, raise the bar
+
+24:21.760 --> 24:23.200
+ on what you choose to work on.
+
+24:23.200 --> 24:24.320
+ So for me, the three things where
+
+24:24.320 --> 24:27.120
+ it had to be something where the technology itself
+
+24:27.120 --> 24:28.960
+ determines the success of the product,
+
+24:28.960 --> 24:31.840
+ like hard, really juicy technology problems,
+
+24:31.840 --> 24:33.600
+ because that's what motivates me.
+
+24:33.600 --> 24:36.280
+ And then it had to have a direct and positive impact
+
+24:36.280 --> 24:37.640
+ on society in some way.
+
+24:37.640 --> 24:39.200
+ So an example would be like, you know,
+
+24:39.200 --> 24:41.560
+ health care, self driving cars because they save lives,
+
+24:41.560 --> 24:43.600
+ other things where there's a clear connection to somehow
+
+24:43.600 --> 24:45.200
+ improving other people's lives.
+
+24:45.200 --> 24:47.160
+ And the last one is it had to be a big business
+
+24:47.160 --> 24:50.200
+ because for the positive impact to matter,
+
+24:50.200 --> 24:51.240
+ it's got to be a large scale.
+
+24:51.240 --> 24:52.080
+ Scale, yeah.
+
+24:52.080 --> 24:53.840
+ And I was thinking about that for a while
+
+24:53.840 --> 24:55.960
+ and I made like a, I tried writing a Gmail clone
+
+24:55.960 --> 24:57.640
+ and looked at some other ideas.
+
+24:57.640 --> 24:59.480
+ And then it just sort of light bulb went off
+
+24:59.480 --> 25:00.440
+ like self driving cars.
+
+25:00.440 --> 25:02.360
+ Like that was the most fun I had ever had
+
+25:02.360 --> 25:04.040
+ in college working on that.
+
+25:04.040 --> 25:05.960
+ And like, well, what's the state of the technology
+
+25:05.960 --> 25:08.440
+ has been 10 years, maybe times have changed
+
+25:08.440 --> 25:10.800
+ and maybe now is the time to make this work.
+
+25:10.800 --> 25:13.320
+ And I poked around and looked at the only other thing
+
+25:13.320 --> 25:15.480
+ out there really at the time was the Google self driving
+
+25:15.480 --> 25:16.680
+ car project.
+
+25:16.680 --> 25:19.600
+ And I thought surely there's a way to, you know,
+
+25:19.600 --> 25:21.600
+ have an entrepreneur mindset and sort of solve
+
+25:21.600 --> 25:23.520
+ the minimum viable product here.
+
+25:23.520 --> 25:25.200
+ And so I just took the plunge right then and there
+
+25:25.200 --> 25:26.680
+ and said, this, this is something I know
+
+25:26.680 --> 25:27.840
+ I can commit 10 years to.
+
+25:27.840 --> 25:30.760
+ It's probably the greatest applied AI problem
+
+25:30.760 --> 25:32.000
+ of our generation.
+
+25:32.000 --> 25:34.240
+ And if it works, it's going to be both a huge business
+
+25:34.240 --> 25:37.040
+ and therefore like probably the most positive impact
+
+25:37.040 --> 25:38.280
+ I can possibly have on the world.
+
+25:38.280 --> 25:40.920
+ So after that light bulb went off,
+
+25:40.920 --> 25:43.000
+ I went all in on cruise immediately
+
+25:43.000 --> 25:45.560
+ and got to work.
+
+25:45.560 --> 25:47.360
+ Did you have an idea how to solve this problem?
+
+25:47.360 --> 25:49.640
+ Which aspect of the problem to solve?
+
+25:49.640 --> 25:53.720
+ You know, slow, like we just had Oliver from voyage here
+
+25:53.720 --> 25:56.560
+ slow moving retirement communities,
+
+25:56.560 --> 25:58.080
+ urban driving, highway driving.
+
+25:58.080 --> 26:00.400
+ Did you have like, did you have a vision
+
+26:00.400 --> 26:03.560
+ of the city of the future or, you know,
+
+26:03.560 --> 26:06.400
+ the transportation is largely automated,
+
+26:06.400 --> 26:07.240
+ that kind of thing.
+
+26:07.240 --> 26:12.240
+ Or was it sort of more fuzzy and gray area than that?
+
+26:12.240 --> 26:16.640
+ My analysis of the situation is that Google's putting a lot,
+
+26:16.640 --> 26:19.200
+ had been putting a lot of money into that project.
+
+26:19.200 --> 26:20.760
+ They had a lot more resources.
+
+26:20.760 --> 26:23.720
+ And so, and they still hadn't cracked
+
+26:23.720 --> 26:26.200
+ the fully driverless car.
+
+26:26.200 --> 26:28.520
+ You know, this is 2013, I guess.
+
+26:29.480 --> 26:33.360
+ So I thought, what can I do to sort of go from zero
+
+26:33.360 --> 26:35.600
+ to, you know, significant scale
+
+26:35.600 --> 26:37.280
+ so I can actually solve the real problem,
+
+26:37.280 --> 26:38.640
+ which is the driverless cars.
+
+26:38.640 --> 26:40.480
+ And I thought, here's the strategy.
+
+26:40.480 --> 26:44.080
+ We'll start by doing a really simple problem
+
+26:44.080 --> 26:45.560
+ or solving a really simple problem
+
+26:45.560 --> 26:48.080
+ that creates value for people.
+
+26:48.080 --> 26:50.040
+ So it eventually ended up deciding
+
+26:50.040 --> 26:51.800
+ on automating highway driving,
+
+26:51.800 --> 26:54.240
+ which is relatively more straightforward
+
+26:54.240 --> 26:56.440
+ as long as there's a backup driver there.
+
+26:56.440 --> 26:58.480
+ And, you know, the go to market
+
+26:58.480 --> 27:00.240
+ will be able to retrofit people's cars
+
+27:00.240 --> 27:02.240
+ and just sell these products directly.
+
+27:02.240 --> 27:04.520
+ And the idea was, we'll take all the revenue
+
+27:04.520 --> 27:08.320
+ and profits from that and use it to do the,
+
+27:08.320 --> 27:10.920
+ to sort of reinvest that in research for doing
+
+27:10.920 --> 27:12.600
+ fully driverless cars.
+
+27:12.600 --> 27:13.960
+ And that was the plan.
+
+27:13.960 --> 27:15.720
+ The only thing that really changed along the way
+
+27:15.720 --> 27:17.360
+ between then and now is,
+
+27:17.360 --> 27:19.000
+ we never really launched the first product.
+
+27:19.000 --> 27:21.680
+ We had enough interest from investors
+
+27:21.680 --> 27:24.120
+ and enough of a signal that this was something
+
+27:24.120 --> 27:25.000
+ that we should be working on,
+
+27:25.000 --> 27:28.400
+ that after about a year of working on the highway autopilot,
+
+27:28.400 --> 27:31.040
+ we had it working, you know, at a prototype stage,
+
+27:31.040 --> 27:33.120
+ but we just completely abandoned that
+
+27:33.120 --> 27:34.960
+ and said, we're gonna go all in on driverless cars
+
+27:34.960 --> 27:36.480
+ now is the time.
+
+27:36.480 --> 27:38.120
+ Can't think of anything that's more exciting.
+
+27:38.120 --> 27:39.720
+ And if it works more impactful,
+
+27:39.720 --> 27:41.360
+ so we're just gonna go for it.
+
+27:41.360 --> 27:43.440
+ The idea of retrofit is kind of interesting.
+
+27:43.440 --> 27:44.280
+ Yeah.
+
+27:44.280 --> 27:46.880
+ Being able to, it's how you achieve scale.
+
+27:46.880 --> 27:47.880
+ It's a really interesting idea,
+
+27:47.880 --> 27:51.120
+ is it's something that's still in the back of your mind
+
+27:51.120 --> 27:52.800
+ as a possibility?
+
+27:52.800 --> 27:53.640
+ Not at all.
+
+27:53.640 --> 27:57.080
+ I've come full circle on that one after trying
+
+27:57.080 --> 27:58.880
+ to build a retrofit product.
+
+27:58.880 --> 28:01.240
+ And I'll touch on some of the complexities of that.
+
+28:01.240 --> 28:04.240
+ And then also having been inside an OEM
+
+28:04.240 --> 28:05.400
+ and seeing how things work
+
+28:05.400 --> 28:08.320
+ and how a vehicle is developed and validated.
+
+28:08.320 --> 28:09.360
+ When it comes to something
+
+28:09.360 --> 28:11.280
+ that has safety critical implications,
+
+28:11.280 --> 28:12.520
+ like controlling the steering
+
+28:12.520 --> 28:15.280
+ and other control inputs on your car,
+
+28:15.280 --> 28:17.720
+ it's pretty hard to get there with a retrofit.
+
+28:17.720 --> 28:20.520
+ Or if you did, even if you did,
+
+28:20.520 --> 28:23.280
+ it creates a whole bunch of new complications around
+
+28:23.280 --> 28:25.400
+ liability or how did you truly validate that?
+
+28:25.400 --> 28:27.480
+ Or, you know, something in the base vehicle fails
+
+28:27.480 --> 28:29.880
+ and causes your system to fail, whose fault is it?
+
+28:31.560 --> 28:34.080
+ Or if the car's anti lock brake systems
+
+28:34.080 --> 28:36.680
+ or other things kick in or the software has been,
+
+28:36.680 --> 28:38.240
+ it's different in one version of the car.
+
+28:38.240 --> 28:40.080
+ You retrofit versus another and you don't know
+
+28:40.080 --> 28:43.000
+ because the manufacturer has updated it behind the scenes.
+
+28:43.000 --> 28:45.400
+ There's basically an infinite list of long tail issues
+
+28:45.400 --> 28:46.240
+ that can get you.
+
+28:46.240 --> 28:47.760
+ And if you're dealing with a safety critical product,
+
+28:47.760 --> 28:48.960
+ that's not really acceptable.
+
+28:48.960 --> 28:52.160
+ That's a really convincing summary of why
+
+28:52.160 --> 28:53.160
+ it's really challenging.
+
+28:53.160 --> 28:54.360
+ But I didn't know all that at the time.
+
+28:54.360 --> 28:55.480
+ So we tried it anyway.
+
+28:55.480 --> 28:57.160
+ But as a pitch also at the time,
+
+28:57.160 --> 28:58.400
+ it's a really strong one.
+
+28:58.400 --> 29:00.720
+ That's how you achieve scale and that's how you beat
+
+29:00.720 --> 29:03.360
+ the current, the leader at the time of Google
+
+29:03.360 --> 29:04.720
+ or the only one in the market.
+
+29:04.720 --> 29:06.840
+ The other big problem we ran into,
+
+29:06.840 --> 29:08.240
+ which is perhaps the biggest problem
+
+29:08.240 --> 29:10.280
+ from a business model perspective,
+
+29:10.280 --> 29:15.280
+ is we had kind of assumed that we started with an Audi S4
+
+29:15.440 --> 29:16.880
+ as the vehicle we retrofitted
+
+29:16.880 --> 29:18.760
+ with this highway driving capability.
+
+29:18.760 --> 29:21.040
+ And we had kind of assumed that if we just knock out
+
+29:21.040 --> 29:23.360
+ like three make and models of vehicle,
+
+29:23.360 --> 29:25.880
+ that'll cover like 80% of the San Francisco market.
+
+29:25.880 --> 29:27.400
+ Doesn't everyone there drive, I don't know,
+
+29:27.400 --> 29:30.240
+ a BMW or a Honda Civic or one of these three cars?
+
+29:30.240 --> 29:32.040
+ And then we surveyed our users and we found out
+
+29:32.040 --> 29:33.480
+ that it's all over the place.
+
+29:33.480 --> 29:36.680
+ We would, to get even a decent number of units sold,
+
+29:36.680 --> 29:39.880
+ we'd have to support like 20 or 50 different models.
+
+29:39.880 --> 29:42.200
+ And each one is a little butterfly that takes time
+
+29:42.200 --> 29:44.800
+ and effort to maintain that retrofit integration
+
+29:44.800 --> 29:47.120
+ and custom hardware and all this.
+
+29:47.120 --> 29:49.240
+ So it was a tough business.
+
+29:49.240 --> 29:54.240
+ So GM manufactures and sells over nine million cars a year.
+
+29:54.280 --> 29:58.560
+ And what you with crews are trying to do
+
+29:58.560 --> 30:01.160
+ some of the most cutting edge innovation
+
+30:01.160 --> 30:03.000
+ in terms of applying AI.
+
+30:03.000 --> 30:06.040
+ And so how do those, you've talked about it a little bit
+
+30:06.040 --> 30:07.760
+ before, but it's also just fascinating to me,
+
+30:07.760 --> 30:09.360
+ we work a lot of automakers.
+
+30:10.560 --> 30:12.880
+ The difference between the gap between Detroit
+
+30:12.880 --> 30:14.680
+ and Silicon Valley, let's say,
+
+30:14.680 --> 30:17.320
+ just to be sort of poetic about it, I guess.
+
+30:17.320 --> 30:18.680
+ How do you close that gap?
+
+30:18.680 --> 30:21.480
+ How do you take GM into the future
+
+30:21.480 --> 30:24.840
+ where a large part of the fleet would be autonomous perhaps?
+
+30:24.840 --> 30:28.520
+ I wanna start by acknowledging that GM is made up of
+
+30:28.520 --> 30:30.240
+ tens of thousands of really brilliant,
+
+30:30.240 --> 30:32.720
+ motivated people who wanna be a part of the future.
+
+30:32.720 --> 30:35.240
+ And so it's pretty fun to work with them.
+
+30:35.240 --> 30:37.480
+ The attitude inside a car company like that
+
+30:37.480 --> 30:41.240
+ is embracing this transformation and change
+
+30:41.240 --> 30:42.360
+ rather than fearing it.
+
+30:42.360 --> 30:45.440
+ And I think that's a testament to the leadership at GM
+
+30:45.440 --> 30:47.680
+ and that's flown all the way through to everyone
+
+30:47.680 --> 30:49.280
+ you talk to, even the people in the assembly plants
+
+30:49.280 --> 30:51.200
+ working on these cars.
+
+30:51.200 --> 30:52.040
+ So that's really great.
+
+30:52.040 --> 30:55.160
+ So starting from that position makes it a lot easier.
+
+30:55.160 --> 30:59.160
+ So then when the people in San Francisco
+
+30:59.160 --> 31:01.400
+ but cruise interact with the people at GM,
+
+31:01.400 --> 31:02.960
+ at least we have this common set of values,
+
+31:02.960 --> 31:05.000
+ which is that we really want this stuff to work
+
+31:05.000 --> 31:06.040
+ because we think it's important
+
+31:06.040 --> 31:07.440
+ and we think it's the future.
+
+31:08.360 --> 31:11.520
+ That's not to say those two cultures don't clash.
+
+31:11.520 --> 31:12.440
+ They absolutely do.
+
+31:12.440 --> 31:14.760
+ There's different sort of value systems.
+
+31:14.760 --> 31:17.960
+ Like in a car company, the thing that gets you promoted
+
+31:17.960 --> 31:22.600
+ and sort of the reward system is following the processes,
+
+31:22.600 --> 31:26.080
+ delivering the program on time and on budget.
+
+31:26.080 --> 31:30.440
+ So any sort of risk taking is discouraged in many ways
+
+31:30.440 --> 31:34.000
+ because if a program is late
+
+31:34.000 --> 31:36.200
+ or if you shut down the plant for a day,
+
+31:36.200 --> 31:37.560
+ you can count the millions of dollars
+
+31:37.560 --> 31:39.600
+ that burn by pretty quickly.
+
+31:39.600 --> 31:43.800
+ Whereas I think most Silicon Valley companies
+
+31:43.800 --> 31:48.280
+ and in cruise and the methodology we were employing,
+
+31:48.280 --> 31:50.080
+ especially around the time of the acquisition,
+
+31:50.080 --> 31:53.800
+ the reward structure is about trying to solve
+
+31:53.800 --> 31:56.120
+ these complex problems in any way, shape or form
+
+31:56.120 --> 31:59.640
+ or coming up with crazy ideas that 90% of them won't work.
+
+31:59.640 --> 32:02.920
+ And so meshing that culture
+
+32:02.920 --> 32:05.480
+ of sort of continuous improvement and experimentation
+
+32:05.480 --> 32:07.400
+ with one where everything needs to be
+
+32:07.400 --> 32:08.480
+ rigorously defined up front
+
+32:08.480 --> 32:12.760
+ so that you never slip a deadline or miss a budget
+
+32:12.760 --> 32:13.600
+ was a pretty big challenge
+
+32:13.600 --> 32:16.960
+ and that we're over three years in now
+
+32:16.960 --> 32:18.360
+ after the acquisition.
+
+32:18.360 --> 32:20.480
+ And I'd say like the investment we made
+
+32:20.480 --> 32:23.600
+ in figuring out how to work together successfully
+
+32:23.600 --> 32:24.440
+ and who should do what
+
+32:24.440 --> 32:26.360
+ and how we bridge the gaps
+
+32:26.360 --> 32:27.680
+ between these very different systems
+
+32:27.680 --> 32:29.520
+ and way of doing engineering work
+
+32:29.520 --> 32:30.920
+ is now one of our greatest assets
+
+32:30.920 --> 32:32.320
+ because I think we have this really powerful thing
+
+32:32.320 --> 32:35.560
+ but for a while it was both GM and cruise
+
+32:35.560 --> 32:37.440
+ were very steep on the learning curve.
+
+32:37.440 --> 32:38.920
+ Yeah, so I'm sure it was very stressful.
+
+32:38.920 --> 32:39.960
+ It's really important work
+
+32:39.960 --> 32:43.680
+ because that's how to revolutionize the transportation.
+
+32:43.680 --> 32:46.640
+ Really to revolutionize any system,
+
+32:46.640 --> 32:48.200
+ you look at the healthcare system
+
+32:48.200 --> 32:49.680
+ or you look at the legal system.
+
+32:49.680 --> 32:52.040
+ I have people like Laura's come up to me all the time
+
+32:52.040 --> 32:53.920
+ like everything they're working on
+
+32:53.920 --> 32:55.960
+ can easily be automated.
+
+32:55.960 --> 32:57.480
+ But then that's not a good feeling.
+
+32:57.480 --> 32:58.320
+ Yeah.
+
+32:58.320 --> 32:59.160
+ Well, it's not a good feeling,
+
+32:59.160 --> 33:01.200
+ but also there's no way to automate
+
+33:01.200 --> 33:06.200
+ because the entire infrastructure is really based
+
+33:06.360 --> 33:08.360
+ is older and it moves very slowly.
+
+33:08.360 --> 33:11.560
+ And so how do you close the gap between?
+
+33:11.560 --> 33:13.880
+ I haven't, how can I replace?
+
+33:13.880 --> 33:15.720
+ Of course, Laura's the one be replaced with an app
+
+33:15.720 --> 33:17.920
+ but you could replace a lot of aspect
+
+33:17.920 --> 33:20.160
+ when most of the data is still on paper.
+
+33:20.160 --> 33:23.400
+ And so the same thing with automotive.
+
+33:23.400 --> 33:26.080
+ I mean, it's fundamentally software.
+
+33:26.080 --> 33:28.560
+ So it's basically hiring software engineers.
+
+33:28.560 --> 33:30.320
+ It's thinking of software world.
+
+33:30.320 --> 33:32.560
+ I mean, I'm pretty sure nobody in Silicon Valley
+
+33:32.560 --> 33:34.640
+ has ever hit a deadline.
+
+33:34.640 --> 33:36.000
+ So and then on GM.
+
+33:36.000 --> 33:37.400
+ That's probably true, yeah.
+
+33:37.400 --> 33:39.920
+ And GM side is probably the opposite.
+
+33:39.920 --> 33:42.720
+ So that's that culture gap is really fascinating.
+
+33:42.720 --> 33:45.160
+ So you're optimistic about the future of that.
+
+33:45.160 --> 33:47.440
+ Yeah, I mean, from what I've seen, it's impressive.
+
+33:47.440 --> 33:49.400
+ And I think like, especially in Silicon Valley,
+
+33:49.400 --> 33:51.440
+ it's easy to write off building cars
+
+33:51.440 --> 33:53.120
+ because people have been doing that
+
+33:53.120 --> 33:54.960
+ for over a hundred years now in this country.
+
+33:54.960 --> 33:57.080
+ And so it seems like that's a solved problem,
+
+33:57.080 --> 33:58.840
+ but that doesn't mean it's an easy problem.
+
+33:58.840 --> 34:02.280
+ And I think it would be easy to sort of overlook that
+
+34:02.280 --> 34:06.080
+ and think that we're Silicon Valley engineers,
+
+34:06.080 --> 34:08.960
+ we can solve any problem, building a car,
+
+34:08.960 --> 34:13.200
+ it's been done, therefore it's not a real engineering
+
+34:13.200 --> 34:14.600
+ challenge.
+
+34:14.600 --> 34:17.480
+ But after having seen just the sheer scale
+
+34:17.480 --> 34:21.360
+ and magnitude and industrialization that occurs
+
+34:21.360 --> 34:23.280
+ inside of an automotive assembly plant,
+
+34:23.280 --> 34:25.840
+ that is a lot of work that I am very glad
+
+34:25.840 --> 34:28.200
+ that we don't have to reinvent
+
+34:28.200 --> 34:29.480
+ to make self driving cars work.
+
+34:29.480 --> 34:31.680
+ And so to have partners who have done that for a hundred
+
+34:31.680 --> 34:32.960
+ years and have these great processes
+
+34:32.960 --> 34:35.720
+ and this huge infrastructure and supply base
+
+34:35.720 --> 34:38.760
+ that we can tap into is just remarkable
+
+34:38.760 --> 34:43.760
+ because the scope and surface area of the problem
+
+34:44.560 --> 34:47.400
+ of deploying fleets of self driving cars is so large
+
+34:47.400 --> 34:50.320
+ that we're constantly looking for ways to do less
+
+34:50.320 --> 34:52.920
+ so we can focus on the things that really matter more.
+
+34:52.920 --> 34:55.360
+ And if we had to figure out how to build and assemble
+
+34:55.360 --> 35:00.120
+ and test and build the cars themselves,
+
+35:00.120 --> 35:01.640
+ I mean, we work closely with GM on that,
+
+35:01.640 --> 35:03.240
+ but if we had to develop all that capability
+
+35:03.240 --> 35:08.240
+ in house as well, that would just make the problem
+
+35:08.320 --> 35:10.200
+ really intractable, I think.
+
+35:10.200 --> 35:14.880
+ So yeah, just like your first entry at the MIT DARPA
+
+35:14.880 --> 35:17.680
+ challenge when it was what the motor that failed
+
+35:17.680 --> 35:19.000
+ and somebody that knows what they're doing
+
+35:19.000 --> 35:20.040
+ with the motor did it.
+
+35:20.040 --> 35:22.080
+ It would have been nice if we could focus on the software
+
+35:22.080 --> 35:23.880
+ and not the hardware platform.
+
+35:23.880 --> 35:24.800
+ Yeah, right.
+
+35:24.800 --> 35:27.080
+ So from your perspective now,
+
+35:28.080 --> 35:29.960
+ there's so many ways that autonomous vehicles
+
+35:29.960 --> 35:34.280
+ can impact society in the next year, five years, 10 years.
+
+35:34.280 --> 35:37.080
+ What do you think is the biggest opportunity
+
+35:37.080 --> 35:39.360
+ to make money in autonomous driving,
+
+35:40.560 --> 35:44.720
+ sort of make it a financially viable thing in the near term?
+
+35:44.720 --> 35:49.120
+ What do you think would be the biggest impact there?
+
+35:49.120 --> 35:52.160
+ Well, the things that drive the economics
+
+35:52.160 --> 35:53.600
+ for fleets of self driving cars
+
+35:53.600 --> 35:56.440
+ are there's sort of a handful of variables.
+
+35:56.440 --> 36:00.400
+ One is the cost to build the vehicle itself.
+
+36:00.400 --> 36:03.720
+ So the material cost, what's the cost of all your sensors,
+
+36:03.720 --> 36:05.200
+ plus the cost of the vehicle
+
+36:05.200 --> 36:07.560
+ and all the other components on it.
+
+36:07.560 --> 36:09.520
+ Another one is the lifetime of the vehicle.
+
+36:09.520 --> 36:12.480
+ It's very different if your vehicle drives 100,000 miles
+
+36:12.480 --> 36:14.800
+ and then it falls apart versus 2 million.
+
+36:16.720 --> 36:18.840
+ And then if you have a fleet,
+
+36:18.840 --> 36:22.920
+ it's kind of like an airplane or an airline
+
+36:22.920 --> 36:26.120
+ where once you produce the vehicle,
+
+36:26.120 --> 36:27.880
+ you want it to be in operation
+
+36:27.880 --> 36:30.760
+ as many hours a day as possible producing revenue.
+
+36:30.760 --> 36:32.480
+ And then the other piece of that
+
+36:32.480 --> 36:35.280
+ is how are you generating revenue?
+
+36:35.280 --> 36:36.880
+ I think that's kind of what you're asking in.
+
+36:36.880 --> 36:38.400
+ I think the obvious things today
+
+36:38.400 --> 36:40.080
+ are the ride sharing business
+
+36:40.080 --> 36:42.760
+ because that's pretty clear that there's demand for that.
+
+36:42.760 --> 36:46.240
+ There's existing markets you can tap into and...
+
+36:46.240 --> 36:47.960
+ Large urban areas, that kind of thing.
+
+36:47.960 --> 36:48.800
+ Yeah, yeah.
+
+36:48.800 --> 36:51.200
+ And I think that there are some real benefits
+
+36:51.200 --> 36:54.520
+ to having cars without drivers
+
+36:54.520 --> 36:56.040
+ compared to sort of the status quo
+
+36:56.040 --> 36:58.520
+ for people who use ride share services today.
+
+36:58.520 --> 37:01.040
+ You know, your privacy, consistency,
+
+37:01.040 --> 37:02.440
+ hopefully significantly improve safety,
+
+37:02.440 --> 37:05.120
+ all these benefits versus the current product.
+
+37:05.120 --> 37:06.520
+ But it's a crowded market.
+
+37:06.520 --> 37:08.000
+ And then other opportunities
+
+37:08.000 --> 37:09.600
+ which you've seen a lot of activity in the last,
+
+37:09.600 --> 37:12.560
+ really in the last six or 12 months is delivery,
+
+37:12.560 --> 37:17.560
+ whether that's parcels and packages, food or groceries.
+
+37:17.800 --> 37:20.320
+ Those are all sort of, I think, opportunities
+
+37:20.320 --> 37:23.640
+ that are pretty ripe for these.
+
+37:23.640 --> 37:26.000
+ Once you have this core technology,
+
+37:26.000 --> 37:28.080
+ which is the fleet of autonomous vehicles,
+
+37:28.080 --> 37:30.920
+ there's all sorts of different business opportunities
+
+37:30.920 --> 37:32.080
+ you can build on top of that.
+
+37:32.080 --> 37:34.520
+ But I think the important thing, of course,
+
+37:34.520 --> 37:36.440
+ is that there's zero monetization opportunity
+
+37:36.440 --> 37:37.520
+ until you actually have that fleet
+
+37:37.520 --> 37:39.160
+ of very capable driverless cars
+
+37:39.160 --> 37:41.040
+ that are as good or better than humans.
+
+37:41.040 --> 37:44.120
+ And that's sort of where the entire industry
+
+37:44.120 --> 37:45.920
+ is sort of in this holding pattern right now.
+
+37:45.920 --> 37:47.960
+ Yeah, they're trying to achieve that baseline.
+
+37:47.960 --> 37:51.520
+ But you said sort of not reliability consistency.
+
+37:51.520 --> 37:52.360
+ It's kind of interesting.
+
+37:52.360 --> 37:54.200
+ I think I heard you say somewhere,
+
+37:54.200 --> 37:55.440
+ not sure if that's what you meant,
+
+37:55.440 --> 37:58.240
+ but I can imagine a situation
+
+37:58.240 --> 38:01.200
+ where you would get an autonomous vehicle.
+
+38:01.200 --> 38:04.560
+ And when you get into an Uber or Lyft,
+
+38:04.560 --> 38:05.960
+ you don't get to choose the driver
+
+38:05.960 --> 38:07.320
+ in a sense that you don't get to choose
+
+38:07.320 --> 38:09.080
+ the personality of the driving.
+
+38:09.080 --> 38:12.040
+ Do you think there's room
+
+38:12.040 --> 38:14.120
+ to define the personality of the car
+
+38:14.120 --> 38:15.040
+ the way it drives you,
+
+38:15.040 --> 38:17.600
+ in terms of aggressiveness, for example,
+
+38:17.600 --> 38:21.120
+ in terms of sort of pushing the boundaries.
+
+38:21.120 --> 38:22.760
+ One of the biggest challenges in autonomous driving
+
+38:22.760 --> 38:27.760
+ is the trade off between sort of safety and assertiveness.
+
+38:28.600 --> 38:30.920
+ And do you think there's any room
+
+38:30.920 --> 38:35.920
+ for the human to take a role in that decision?
+
+38:36.040 --> 38:38.080
+ Sort of accept some of the liability, I guess.
+
+38:38.080 --> 38:41.000
+ I wouldn't say, no, I'd say within reasonable bounds,
+
+38:41.000 --> 38:42.280
+ as in we're not gonna,
+
+38:43.200 --> 38:44.360
+ I think it'd be higher than likely
+
+38:44.360 --> 38:46.600
+ we'd expose any knob that would let you
+
+38:46.600 --> 38:50.240
+ significantly increase safety risk.
+
+38:50.240 --> 38:53.080
+ I think that's just not something we'd be willing to do.
+
+38:53.080 --> 38:56.760
+ But I think driving style or like,
+
+38:56.760 --> 38:59.120
+ are you gonna relax the comfort constraints slightly
+
+38:59.120 --> 39:00.160
+ or things like that?
+
+39:00.160 --> 39:02.400
+ All of those things make sense and are plausible.
+
+39:02.400 --> 39:04.480
+ I see all those as nice optimizations.
+
+39:04.480 --> 39:06.760
+ Once again, we get the core problem solved
+
+39:06.760 --> 39:08.120
+ in these fleets out there.
+
+39:08.120 --> 39:10.440
+ But the other thing we've sort of observed
+
+39:10.440 --> 39:12.560
+ is that you have this intuition
+
+39:12.560 --> 39:15.400
+ that if you sort of slam your foot on the gas
+
+39:15.400 --> 39:16.680
+ right after the light turns green
+
+39:16.680 --> 39:18.160
+ and aggressively accelerate,
+
+39:18.160 --> 39:19.720
+ you're gonna get there faster.
+
+39:19.720 --> 39:22.080
+ But the actual impact of doing that is pretty small.
+
+39:22.080 --> 39:23.680
+ You feel like you're getting there faster,
+
+39:23.680 --> 39:26.680
+ but so the same would be true for AVs.
+
+39:26.680 --> 39:29.640
+ Even if they don't slam the pedal to the floor
+
+39:29.640 --> 39:31.000
+ when the light turns green,
+
+39:31.000 --> 39:32.520
+ they're gonna get you there within,
+
+39:32.520 --> 39:33.600
+ if it's a 15 minute trip,
+
+39:33.600 --> 39:36.400
+ within 30 seconds of what you would have done otherwise
+
+39:36.400 --> 39:37.800
+ if you were going really aggressively.
+
+39:37.800 --> 39:40.760
+ So I think there's this sort of self deception
+
+39:40.760 --> 39:44.440
+ that my aggressive driving style is getting me there faster.
+
+39:44.440 --> 39:46.640
+ Well, so that's, you know, some of the things I study,
+
+39:46.640 --> 39:48.760
+ some of the things I'm fascinated by the psychology of that.
+
+39:48.760 --> 39:50.640
+ And I don't think it matters
+
+39:50.640 --> 39:52.240
+ that it doesn't get you there faster.
+
+39:52.240 --> 39:55.520
+ It's the emotional release.
+
+39:55.520 --> 39:59.080
+ Driving is a place, being inside our car,
+
+39:59.080 --> 40:00.880
+ somebody said it's like the real world version
+
+40:00.880 --> 40:02.920
+ of being a troll.
+
+40:02.920 --> 40:04.960
+ So you have this protection, this mental protection,
+
+40:04.960 --> 40:06.640
+ and you're able to sort of yell at the world,
+
+40:06.640 --> 40:08.200
+ like release your anger, whatever it is.
+
+40:08.200 --> 40:10.040
+ But so there's an element of that
+
+40:10.040 --> 40:12.000
+ that I think autonomous vehicles
+
+40:12.000 --> 40:15.400
+ would also have to, you know, giving an outlet to people,
+
+40:15.400 --> 40:19.120
+ but it doesn't have to be through driving or honking
+
+40:19.120 --> 40:21.200
+ or so on, there might be other outlets.
+
+40:21.200 --> 40:24.040
+ But I think to just sort of even just put that aside,
+
+40:24.040 --> 40:26.880
+ the baseline is really, you know, that's the focus,
+
+40:26.880 --> 40:28.200
+ that's the thing you need to solve,
+
+40:28.200 --> 40:31.000
+ and then the fun human things can be solved after.
+
+40:31.000 --> 40:34.680
+ But so from the baseline of just solving autonomous driving,
+
+40:34.680 --> 40:36.000
+ you're working in San Francisco,
+
+40:36.000 --> 40:38.960
+ one of the more difficult cities to operate in,
+
+40:38.960 --> 40:42.080
+ what is the, in your view currently,
+
+40:42.080 --> 40:45.040
+ the hardest aspect of autonomous driving?
+
+40:46.880 --> 40:49.200
+ Negotiating with pedestrians,
+
+40:49.200 --> 40:51.400
+ is it edge cases of perception?
+
+40:51.400 --> 40:52.760
+ Is it planning?
+
+40:52.760 --> 40:54.520
+ Is there a mechanical engineering?
+
+40:54.520 --> 40:57.040
+ Is it data, fleet stuff?
+
+40:57.040 --> 41:01.200
+ What are your thoughts on the more challenging aspects there?
+
+41:01.200 --> 41:02.240
+ That's a good question.
+
+41:02.240 --> 41:03.520
+ I think before we go to that though,
+
+41:03.520 --> 41:05.080
+ I just want to, I like what you said
+
+41:05.080 --> 41:07.600
+ about the psychology aspect of this,
+
+41:07.600 --> 41:09.680
+ because I think one observation I've made is,
+
+41:09.680 --> 41:11.760
+ I think I read somewhere that I think it's,
+
+41:11.760 --> 41:13.880
+ maybe Americans on average spend, you know,
+
+41:13.880 --> 41:16.520
+ over an hour a day on social media,
+
+41:16.520 --> 41:18.280
+ like staring at Facebook.
+
+41:18.280 --> 41:20.080
+ And so that's just, you know,
+
+41:20.080 --> 41:21.600
+ 60 minutes of your life, you're not getting back.
+
+41:21.600 --> 41:23.120
+ It's probably not super productive.
+
+41:23.120 --> 41:26.200
+ And so that's 3,600 seconds, right?
+
+41:26.200 --> 41:29.160
+ And that's, that's time, you know,
+
+41:29.160 --> 41:30.600
+ it's a lot of time you're giving up.
+
+41:30.600 --> 41:34.080
+ And if you compare that to people being on the road,
+
+41:34.080 --> 41:35.360
+ if another vehicle,
+
+41:35.360 --> 41:37.600
+ whether it's a human driver or autonomous vehicle,
+
+41:37.600 --> 41:39.840
+ delays them by even three seconds,
+
+41:39.840 --> 41:41.920
+ they're laying in on the horn, you know,
+
+41:41.920 --> 41:43.280
+ even though that's, that's, you know,
+
+41:43.280 --> 41:45.280
+ one 1,000th of the time they waste
+
+41:45.280 --> 41:46.360
+ looking at Facebook every day.
+
+41:46.360 --> 41:48.640
+ So there's, there's definitely some,
+
+41:48.640 --> 41:50.040
+ you know, psychology aspects of this,
+
+41:50.040 --> 41:50.880
+ I think that are pretty interesting.
+
+41:50.880 --> 41:51.720
+ Road rage in general.
+
+41:51.720 --> 41:52.960
+ And then the question, of course,
+
+41:52.960 --> 41:54.960
+ is if everyone is in self driving cars,
+
+41:54.960 --> 41:57.560
+ do they even notice these three second delays anymore?
+
+41:57.560 --> 41:58.920
+ Because they're doing other things
+
+41:58.920 --> 42:01.720
+ or reading or working or just talking to each other.
+
+42:01.720 --> 42:03.200
+ So it'll be interesting to see where that goes.
+
+42:03.200 --> 42:05.120
+ In a certain aspect, people,
+
+42:05.120 --> 42:06.360
+ people need to be distracted
+
+42:06.360 --> 42:07.360
+ by something entertaining,
+
+42:07.360 --> 42:09.160
+ something useful inside the car
+
+42:09.160 --> 42:10.960
+ so they don't pay attention to the external world.
+
+42:10.960 --> 42:14.240
+ And then, and then they can take whatever psychology
+
+42:14.240 --> 42:17.400
+ and bring it back to Twitter and then focus on that
+
+42:17.400 --> 42:19.640
+ as opposed to sort of interacting,
+
+42:20.920 --> 42:23.200
+ sort of putting the emotion out there into the world.
+
+42:23.200 --> 42:24.560
+ So it's an interesting problem,
+
+42:24.560 --> 42:26.960
+ but baseline autonomy.
+
+42:26.960 --> 42:28.760
+ I guess you could say self driving cars,
+
+42:28.760 --> 42:31.680
+ you know, at scale will lower the collective blood pressure
+
+42:31.680 --> 42:33.920
+ of society probably by a couple of points
+
+42:33.920 --> 42:35.760
+ without all that road rage and stress.
+
+42:35.760 --> 42:37.480
+ So that's a good, good externality.
+
+42:38.560 --> 42:41.760
+ So back to your question about the technology
+
+42:41.760 --> 42:43.760
+ and the, I guess the biggest problems.
+
+42:43.760 --> 42:45.560
+ And I have a hard time answering that question
+
+42:45.560 --> 42:48.680
+ because, you know, we've been at this,
+
+42:48.680 --> 42:51.440
+ like specifically focusing on driverless cars
+
+42:51.440 --> 42:53.520
+ and all the technology needed to enable that
+
+42:53.520 --> 42:55.160
+ for a little over four and a half years now.
+
+42:55.160 --> 42:58.080
+ And even a year or two in,
+
+42:58.080 --> 43:02.960
+ I felt like we had completed the functionality needed
+
+43:02.960 --> 43:04.800
+ to get someone from point A to point B.
+
+43:04.800 --> 43:07.280
+ As in, if we need to do a left turn maneuver
+
+43:07.280 --> 43:08.960
+ or if we need to drive around a, you know,
+
+43:08.960 --> 43:11.800
+ a double parked vehicle into oncoming traffic
+
+43:11.800 --> 43:13.840
+ or navigate through construction zones,
+
+43:13.840 --> 43:15.960
+ the scaffolding and the building blocks
+
+43:15.960 --> 43:17.800
+ was there pretty early on.
+
+43:17.800 --> 43:22.360
+ And so the challenge is not any one scenario or situation
+
+43:22.360 --> 43:25.520
+ for which, you know, we fail at 100% of those.
+
+43:25.520 --> 43:28.960
+ It's more, you know, we're benchmarking against a pretty good
+
+43:28.960 --> 43:31.320
+ or pretty high standard, which is human driving.
+
+43:31.320 --> 43:33.320
+ All things considered, humans are excellent
+
+43:33.320 --> 43:36.240
+ at handling edge cases and unexpected scenarios
+
+43:36.240 --> 43:38.400
+ where it's computers are the opposite.
+
+43:38.400 --> 43:43.080
+ And so beating that baseline set by humans is the challenge.
+
+43:43.080 --> 43:46.520
+ And so what we've been doing for quite some time now
+
+43:46.520 --> 43:50.760
+ is basically it's this continuous improvement process
+
+43:50.760 --> 43:55.000
+ where we find sort of the most, you know, uncomfortable
+
+43:55.000 --> 43:59.840
+ or the things that could lead to a safety issue
+
+43:59.840 --> 44:00.960
+ or other things, all these events.
+
+44:00.960 --> 44:02.520
+ And then we sort of categorize them
+
+44:02.520 --> 44:04.560
+ and rework parts of our system
+
+44:04.560 --> 44:06.200
+ to make incremental improvements
+
+44:06.200 --> 44:08.040
+ and do that over and over and over again.
+
+44:08.040 --> 44:10.160
+ And we just see sort of the overall performance
+
+44:10.160 --> 44:12.120
+ of the system, you know,
+
+44:12.120 --> 44:13.960
+ actually increasing in a pretty steady clip.
+
+44:13.960 --> 44:15.360
+ But there's no one thing.
+
+44:15.360 --> 44:17.360
+ There's actually like thousands of little things
+
+44:17.360 --> 44:19.880
+ and just like polishing functionality
+
+44:19.880 --> 44:21.640
+ and making sure that it handles, you know,
+
+44:21.640 --> 44:26.120
+ every version and possible permutation of a situation
+
+44:26.120 --> 44:29.200
+ by either applying more deep learning systems
+
+44:30.120 --> 44:32.960
+ or just by, you know, adding more test coverage
+
+44:32.960 --> 44:35.760
+ or new scenarios that we develop against
+
+44:35.760 --> 44:37.160
+ and just grinding on that.
+
+44:37.160 --> 44:40.120
+ We're sort of in the unsexy phase of development right now
+
+44:40.120 --> 44:41.800
+ which is doing the real engineering work
+
+44:41.800 --> 44:44.120
+ that it takes to go from prototype to production.
+
+44:44.120 --> 44:46.960
+ You're basically scaling the grinding.
+
+44:46.960 --> 44:50.560
+ So sort of taking seriously the process
+
+44:50.560 --> 44:54.040
+ of all those edge cases, both with human experts
+
+44:54.040 --> 44:57.520
+ and machine learning methods to cover,
+
+44:57.520 --> 44:59.320
+ to cover all those situations.
+
+44:59.320 --> 45:00.760
+ Yeah, and the exciting thing for me is
+
+45:00.760 --> 45:03.000
+ I don't think that grinding ever stops
+
+45:03.000 --> 45:04.840
+ because there's a moment in time
+
+45:04.840 --> 45:08.760
+ where you've crossed that threshold of human performance
+
+45:08.760 --> 45:10.000
+ and become superhuman.
+
+45:11.200 --> 45:13.560
+ But there's no reason, there's no first principles reason
+
+45:13.560 --> 45:17.560
+ that AV capability will tap out anywhere near humans.
+
+45:17.560 --> 45:20.280
+ Like there's no reason it couldn't be 20 times better
+
+45:20.280 --> 45:22.120
+ whether that's, you know, just better driving
+
+45:22.120 --> 45:24.240
+ or safer driving or more comfortable driving
+
+45:24.240 --> 45:26.800
+ or even a thousand times better given enough time.
+
+45:26.800 --> 45:31.480
+ And we intend to basically chase that, you know, forever
+
+45:31.480 --> 45:32.840
+ to build the best possible product.
+
+45:32.840 --> 45:33.960
+ Better and better and better
+
+45:33.960 --> 45:36.400
+ and always new edge cases come up and new experiences.
+
+45:36.400 --> 45:39.520
+ So, and you want to automate that process
+
+45:39.520 --> 45:40.720
+ as much as possible.
+
+45:42.680 --> 45:45.160
+ So what do you think in general in society
+
+45:45.160 --> 45:48.200
+ when do you think we may have hundreds of thousands
+
+45:48.200 --> 45:50.200
+ of fully autonomous vehicles driving around?
+
+45:50.200 --> 45:53.560
+ So first of all, predictions, nobody knows the future.
+
+45:53.560 --> 45:55.360
+ You're a part of the leading people
+
+45:55.360 --> 45:56.560
+ trying to define that future,
+
+45:56.560 --> 45:58.560
+ but even then you still don't know.
+
+45:58.560 --> 46:02.240
+ But if you think about hundreds of thousands of vehicles,
+
+46:02.240 --> 46:05.840
+ so a significant fraction of vehicles
+
+46:05.840 --> 46:07.600
+ in major cities are autonomous.
+
+46:07.600 --> 46:10.800
+ Do you think, are you with Rodney Brooks
+
+46:10.800 --> 46:13.960
+ who is 2050 and beyond?
+
+46:13.960 --> 46:17.200
+ Or are you more with Elon Musk
+
+46:17.200 --> 46:20.600
+ who is, we should have had that two years ago?
+
+46:20.600 --> 46:23.840
+ Well, I mean, I'd love to have it two years ago,
+
+46:23.840 --> 46:26.120
+ but we're not there yet.
+
+46:26.120 --> 46:28.480
+ So I guess the way I would think about that
+
+46:28.480 --> 46:31.240
+ is let's flip that question around.
+
+46:31.240 --> 46:34.200
+ So what would prevent you to reach hundreds
+
+46:34.200 --> 46:36.320
+ of thousands of vehicles and...
+
+46:36.320 --> 46:38.200
+ That's a good rephrasing.
+
+46:38.200 --> 46:43.200
+ Yeah, so the, I'd say that it seems the consensus
+
+46:43.200 --> 46:45.200
+ among the people developing self driving cars today
+
+46:45.200 --> 46:49.200
+ is to sort of start with some form of an easier environment,
+
+46:49.200 --> 46:52.200
+ whether it means lacking, inclement weather,
+
+46:52.200 --> 46:55.200
+ or mostly sunny or whatever it is.
+
+46:55.200 --> 46:59.200
+ And then add capability for more complex situations
+
+46:59.200 --> 47:00.200
+ over time.
+
+47:00.200 --> 47:05.200
+ And so if you're only able to deploy in areas
+
+47:05.200 --> 47:07.200
+ that meet sort of your criteria
+
+47:07.200 --> 47:09.200
+ or that the current don't meet,
+
+47:09.200 --> 47:13.200
+ operating domain of the software you developed,
+
+47:13.200 --> 47:16.200
+ that may put a cap on how many cities you could deploy in.
+
+47:16.200 --> 47:19.200
+ But then as those restrictions start to fall away,
+
+47:19.200 --> 47:22.200
+ like maybe you add capability to drive really well
+
+47:22.200 --> 47:25.200
+ and safely and have you rain or snow,
+
+47:25.200 --> 47:28.200
+ that probably opens up the market by two or three fold
+
+47:28.200 --> 47:31.200
+ in terms of the cities you can expand into and so on.
+
+47:31.200 --> 47:33.200
+ And so the real question is,
+
+47:33.200 --> 47:35.200
+ I know today if we wanted to,
+
+47:35.200 --> 47:39.200
+ we could produce that many autonomous vehicles,
+
+47:39.200 --> 47:41.200
+ but we wouldn't be able to make use of all of them yet
+
+47:41.200 --> 47:44.200
+ because we would sort of saturate the demand in the cities
+
+47:44.200 --> 47:47.200
+ in which we would want to operate initially.
+
+47:47.200 --> 47:49.200
+ So if I were to guess what the timeline is
+
+47:49.200 --> 47:51.200
+ for those things falling away
+
+47:51.200 --> 47:54.200
+ and reaching hundreds, thousands of vehicles.
+
+47:54.200 --> 47:55.200
+ Maybe a range is better.
+
+47:55.200 --> 47:57.200
+ I would say less than five years.
+
+47:57.200 --> 47:58.200
+ Less than five years.
+
+47:58.200 --> 47:59.200
+ Yeah.
+
+47:59.200 --> 48:02.200
+ And of course you're working hard to make that happen.
+
+48:02.200 --> 48:05.200
+ So you started two companies that were eventually acquired
+
+48:05.200 --> 48:08.200
+ for each $4 billion.
+
+48:08.200 --> 48:10.200
+ So you're a pretty good person to ask,
+
+48:10.200 --> 48:13.200
+ what does it take to build a successful startup?
+
+48:13.200 --> 48:18.200
+ I think there's sort of survivor bias here a little bit,
+
+48:18.200 --> 48:20.200
+ but I can try to find some common threads
+
+48:20.200 --> 48:22.200
+ for the things that worked for me, which is...
+
+48:24.200 --> 48:26.200
+ In both of these companies,
+
+48:26.200 --> 48:28.200
+ I was really passionate about the core technology.
+
+48:28.200 --> 48:31.200
+ I actually lay awake at night thinking about these problems
+
+48:31.200 --> 48:33.200
+ and how to solve them.
+
+48:33.200 --> 48:35.200
+ And I think that's helpful because when you start a business,
+
+48:35.200 --> 48:37.200
+ there are...
+
+48:37.200 --> 48:40.200
+ To this day, there are these crazy ups and downs.
+
+48:40.200 --> 48:43.200
+ One day you think the business is just on top of the world
+
+48:43.200 --> 48:45.200
+ and unstoppable and the next day you think,
+
+48:45.200 --> 48:47.200
+ okay, this is all going to end.
+
+48:47.200 --> 48:50.200
+ It's just going south and it's going to be over tomorrow.
+
+48:52.200 --> 48:55.200
+ And so I think having a true passion that you can fall back on
+
+48:55.200 --> 48:57.200
+ and knowing that you would be doing it
+
+48:57.200 --> 48:58.200
+ even if you weren't getting paid for it
+
+48:58.200 --> 49:00.200
+ helps you weather those tough times.
+
+49:00.200 --> 49:02.200
+ So that's one thing.
+
+49:02.200 --> 49:05.200
+ I think the other one is really good people.
+
+49:05.200 --> 49:07.200
+ So I've always been surrounded by really good cofounders
+
+49:07.200 --> 49:09.200
+ that are logical thinkers,
+
+49:09.200 --> 49:11.200
+ are always pushing their limits
+
+49:11.200 --> 49:13.200
+ and have very high levels of integrity.
+
+49:13.200 --> 49:15.200
+ So that's Dan Kahn in my current company
+
+49:15.200 --> 49:17.200
+ and actually his brother and a couple other guys
+
+49:17.200 --> 49:19.200
+ for Justin TV and Twitch.
+
+49:19.200 --> 49:23.200
+ And then I think the last thing is just,
+
+49:23.200 --> 49:26.200
+ I guess, persistence or perseverance.
+
+49:26.200 --> 49:29.200
+ And that can apply to sticking to
+
+49:29.200 --> 49:33.200
+ having conviction around the original premise of your idea
+
+49:33.200 --> 49:36.200
+ and sticking around to do all the unsexy work
+
+49:36.200 --> 49:38.200
+ to actually make it come to fruition,
+
+49:38.200 --> 49:41.200
+ including dealing with whatever it is
+
+49:41.200 --> 49:43.200
+ that you're not passionate about,
+
+49:43.200 --> 49:47.200
+ whether that's finance or HR or operations or those things.
+
+49:47.200 --> 49:49.200
+ As long as you are grinding away
+
+49:49.200 --> 49:52.200
+ and working towards that North Star for your business,
+
+49:52.200 --> 49:54.200
+ whatever it is and you don't give up
+
+49:54.200 --> 49:56.200
+ and you're making progress every day,
+
+49:56.200 --> 49:58.200
+ it seems like eventually you'll end up in a good place.
+
+49:58.200 --> 50:00.200
+ And the only things that can slow you down
+
+50:00.200 --> 50:01.200
+ are running out of money
+
+50:01.200 --> 50:03.200
+ or I suppose your competitor is destroying you,
+
+50:03.200 --> 50:06.200
+ but I think most of the time it's people giving up
+
+50:06.200 --> 50:08.200
+ or somehow destroying things themselves
+
+50:08.200 --> 50:10.200
+ rather than being beaten by their competition
+
+50:10.200 --> 50:11.200
+ or running out of money.
+
+50:11.200 --> 50:14.200
+ Yeah, if you never quit, eventually you'll arrive.
+
+50:14.200 --> 50:16.200
+ It's a much more concise version
+
+50:16.200 --> 50:18.200
+ of what I was trying to say.
+
+50:18.200 --> 50:21.200
+ So you went the Y Combinator out twice.
+
+50:21.200 --> 50:23.200
+ What do you think, in a quick question,
+
+50:23.200 --> 50:25.200
+ do you think is the best way to raise funds
+
+50:25.200 --> 50:27.200
+ in the early days?
+
+50:27.200 --> 50:30.200
+ Or not just funds, but just community,
+
+50:30.200 --> 50:32.200
+ develop your idea and so on.
+
+50:32.200 --> 50:37.200
+ Can you do it solo or maybe with a cofounder
+
+50:37.200 --> 50:39.200
+ like self funded?
+
+50:39.200 --> 50:40.200
+ Do you think Y Combinator is good?
+
+50:40.200 --> 50:41.200
+ Is it good to do VC route?
+
+50:41.200 --> 50:43.200
+ Is there no right answer or is there,
+
+50:43.200 --> 50:45.200
+ from the Y Combinator experience,
+
+50:45.200 --> 50:47.200
+ something that you could take away
+
+50:47.200 --> 50:49.200
+ that that was the right path to take?
+
+50:49.200 --> 50:50.200
+ There's no one size fits all answer,
+
+50:50.200 --> 50:54.200
+ but if your ambition I think is to see how big
+
+50:54.200 --> 50:57.200
+ you can make something or rapidly expand
+
+50:57.200 --> 50:59.200
+ and capture a market or solve a problem
+
+50:59.200 --> 51:02.200
+ or whatever it is, then going the venture
+
+51:02.200 --> 51:04.200
+ back route is probably a good approach
+
+51:04.200 --> 51:07.200
+ so that capital doesn't become your primary constraint.
+
+51:07.200 --> 51:10.200
+ Y Combinator, I love because it puts you
+
+51:10.200 --> 51:13.200
+ in this sort of competitive environment
+
+51:13.200 --> 51:16.200
+ where you're surrounded by the top,
+
+51:16.200 --> 51:19.200
+ maybe 1% of other really highly motivated
+
+51:19.200 --> 51:22.200
+ peers who are in the same place.
+
+51:22.200 --> 51:26.200
+ In that environment I think just breeds success.
+
+51:26.200 --> 51:28.200
+ If you're surrounded by really brilliant
+
+51:28.200 --> 51:30.200
+ hardworking people, you're going to feel
+
+51:30.200 --> 51:32.200
+ sort of compelled or inspired to try
+
+51:32.200 --> 51:35.200
+ to emulate them or beat them.
+
+51:35.200 --> 51:37.200
+ So even though I had done it once before
+
+51:37.200 --> 51:41.200
+ and I felt like I'm pretty self motivated,
+
+51:41.200 --> 51:43.200
+ I thought this is going to be a hard problem,
+
+51:43.200 --> 51:45.200
+ I can use all the help I can get.
+
+51:45.200 --> 51:46.200
+ So surrounding myself with other entrepreneurs
+
+51:46.200 --> 51:48.200
+ is going to make me work a little bit harder
+
+51:48.200 --> 51:51.200
+ or push a little harder then it's worth it.
+
+51:51.200 --> 51:54.200
+ That's why I did it, for example, the second time.
+
+51:54.200 --> 51:57.200
+ Let's go full soft, go existential.
+
+51:57.200 --> 52:00.200
+ If you go back and do something differently in your life,
+
+52:00.200 --> 52:06.200
+ starting in high school and MIT, leaving MIT,
+
+52:06.200 --> 52:08.200
+ you could have gone to the PhD route,
+
+52:08.200 --> 52:13.200
+ doing startup, going to see about a startup in California
+
+52:13.200 --> 52:15.200
+ or maybe some aspects of fundraising.
+
+52:15.200 --> 52:17.200
+ Is there something you regret,
+
+52:17.200 --> 52:20.200
+ not necessarily regret, but if you go back,
+
+52:20.200 --> 52:22.200
+ you could do differently?
+
+52:22.200 --> 52:24.200
+ I think I've made a lot of mistakes,
+
+52:24.200 --> 52:26.200
+ pretty much everything you can screw up,
+
+52:26.200 --> 52:28.200
+ I think I've screwed up at least once.
+
+52:28.200 --> 52:30.200
+ But I don't regret those things.
+
+52:30.200 --> 52:32.200
+ I think it's hard to look back on things,
+
+52:32.200 --> 52:34.200
+ even if they didn't go well and call it a regret,
+
+52:34.200 --> 52:37.200
+ because hopefully it took away some new knowledge
+
+52:37.200 --> 52:39.200
+ or learning from that.
+
+52:42.200 --> 52:45.200
+ I would say there's a period,
+
+52:45.200 --> 52:47.200
+ the closest I can come to this,
+
+52:47.200 --> 52:49.200
+ there's a period in just in TV,
+
+52:49.200 --> 52:54.200
+ I think after seven years where the company was going
+
+52:54.200 --> 52:57.200
+ one direction, which is towards Twitch and video gaming.
+
+52:57.200 --> 52:58.200
+ I'm not a video gamer.
+
+52:58.200 --> 53:01.200
+ I don't really even use Twitch at all.
+
+53:01.200 --> 53:04.200
+ I was still working on the core technology there,
+
+53:04.200 --> 53:06.200
+ but my heart was no longer in it,
+
+53:06.200 --> 53:08.200
+ because the business that we were creating
+
+53:08.200 --> 53:10.200
+ was not something that I was personally passionate about.
+
+53:10.200 --> 53:12.200
+ It didn't meet your bar of existential impact.
+
+53:12.200 --> 53:16.200
+ Yeah, and I'd say I probably spent an extra year or two
+
+53:16.200 --> 53:20.200
+ working on that, and I'd say I would have just tried
+
+53:20.200 --> 53:22.200
+ to do something different sooner.
+
+53:22.200 --> 53:26.200
+ Because those were two years where I felt like,
+
+53:26.200 --> 53:29.200
+ from this philosophical or existential thing,
+
+53:29.200 --> 53:31.200
+ I just felt that something was missing.
+
+53:31.200 --> 53:34.200
+ If I could look back now and tell myself,
+
+53:34.200 --> 53:35.200
+ I would have said exactly that.
+
+53:35.200 --> 53:38.200
+ You're not getting any meaning out of your work personally
+
+53:38.200 --> 53:39.200
+ right now.
+
+53:39.200 --> 53:41.200
+ You should find a way to change that.
+
+53:41.200 --> 53:44.200
+ And that's part of the pitch I used
+
+53:44.200 --> 53:46.200
+ to basically everyone who joins Cruise today.
+
+53:46.200 --> 53:48.200
+ It's like, hey, you've got that now by coming here.
+
+53:48.200 --> 53:51.200
+ Well, maybe you needed the two years of that existential dread
+
+53:51.200 --> 53:53.200
+ to develop the feeling that ultimately
+
+53:53.200 --> 53:55.200
+ it was the fire that created Cruise.
+
+53:55.200 --> 53:56.200
+ So you never know.
+
+53:56.200 --> 53:57.200
+ You can't repair.
+
+53:57.200 --> 53:58.200
+ Good theory, yeah.
+
+53:58.200 --> 53:59.200
+ So last question.
+
+53:59.200 --> 54:02.200
+ What does 2019 hold for Cruise?
+
+54:02.200 --> 54:05.200
+ After this, I guess we're going to go and talk to your class.
+
+54:05.200 --> 54:08.200
+ But one of the big things is going from prototype to production
+
+54:08.200 --> 54:09.200
+ for autonomous cars.
+
+54:09.200 --> 54:10.200
+ And what does that mean?
+
+54:10.200 --> 54:11.200
+ What does that look like?
+
+54:11.200 --> 54:14.200
+ 2019 for us is the year that we try to cross over
+
+54:14.200 --> 54:17.200
+ that threshold and reach superhuman level of performance
+
+54:17.200 --> 54:20.200
+ to some degree with the software and have all the other
+
+54:20.200 --> 54:23.200
+ of the thousands of little building blocks in place
+
+54:23.200 --> 54:27.200
+ to launch our first commercial product.
+
+54:27.200 --> 54:30.200
+ So that's what's in store for us.
+
+54:30.200 --> 54:32.200
+ And we've got a lot of work to do.
+
+54:32.200 --> 54:35.200
+ We've got a lot of brilliant people working on it.
+
+54:35.200 --> 54:37.200
+ So it's all up to us now.
+
+54:37.200 --> 54:38.200
+ Yeah.
+
+54:38.200 --> 54:41.200
+ So Charlie Miller and Chris Vell is like the people I've
+
+54:41.200 --> 54:42.200
+ crossed paths with.
+
+54:42.200 --> 54:43.200
+ Oh, great, yeah.
+
+54:43.200 --> 54:46.200
+ It sounds like you have an amazing team.
+
+54:46.200 --> 54:49.200
+ So like I said, it's one of the most, I think, one of the most
+
+54:49.200 --> 54:52.200
+ important problems in artificial intelligence of this century.
+
+54:52.200 --> 54:53.200
+ It'll be one of the most defining.
+
+54:53.200 --> 54:55.200
+ It's super exciting that you work on it.
+
+54:55.200 --> 54:59.200
+ And the best of luck in 2019.
+
+54:59.200 --> 55:01.200
+ I'm really excited to see what Cruise comes up with.
+
+55:01.200 --> 55:02.200
+ Thank you.
+
+55:02.200 --> 55:03.200
+ Thanks for having me today.
+
+55:03.200 --> 55:08.200
+ Thank you.
+

vtt/episode_015_small.vtt

@@ -0,0 +1,1826 @@
+WEBVTT
+
+00:00.000 --> 00:05.440
+ The following is a conversation with Leslie Kailbling. She is a roboticist and professor at
+
+00:05.440 --> 00:12.080
+ MIT. She is recognized for her work in reinforcement learning, planning, robot navigation, and several
+
+00:12.080 --> 00:18.560
+ other topics in AI. She won the Ijkai Computers and Thought Award and was the editor in chief
+
+00:18.560 --> 00:24.320
+ of the prestigious journal machine learning research. This conversation is part of the
+
+00:24.320 --> 00:30.400
+ artificial intelligence podcast at MIT and beyond. If you enjoy it, subscribe on YouTube,
+
+00:30.400 --> 00:37.760
+ iTunes, or simply connect with me on Twitter at Lex Friedman, spelled F R I D. And now,
+
+00:37.760 --> 00:45.360
+ here's my conversation with Leslie Kailbling. What made me get excited about AI, I can say
+
+00:45.360 --> 00:49.920
+ that, is I read Girdle Escherbach when I was in high school. That was pretty formative for me
+
+00:49.920 --> 00:59.280
+ because it exposed the interestingness of primitives and combination and how you can
+
+00:59.280 --> 01:06.000
+ make complex things out of simple parts and ideas of AI and what kinds of programs might
+
+01:06.000 --> 01:12.800
+ generate intelligent behavior. So you first fell in love with AI reasoning logic versus robots?
+
+01:12.800 --> 01:18.240
+ Yeah, the robots came because my first job, so I finished an undergraduate degree in philosophy
+
+01:18.240 --> 01:24.160
+ at Stanford and was about to finish masters in computer science and I got hired at SRI
+
+01:25.440 --> 01:30.960
+ in their AI lab and they were building a robot. It was a kind of a follow on to shaky,
+
+01:30.960 --> 01:35.840
+ but all the shaky people were not there anymore. And so my job was to try to get this robot to
+
+01:35.840 --> 01:41.200
+ do stuff and that's really kind of what got me interested in robots. So maybe taking a small
+
+01:41.200 --> 01:46.160
+ step back to your bachelor's in Stanford philosophy, did master's in PhD in computer science,
+
+01:46.160 --> 01:51.440
+ but the bachelor's in philosophy. So what was that journey like? What elements of philosophy
+
+01:52.320 --> 01:55.200
+ do you think you bring to your work in computer science?
+
+01:55.200 --> 02:00.080
+ So it's surprisingly relevant. So part of the reason that I didn't do a computer science
+
+02:00.080 --> 02:04.560
+ undergraduate degree was that there wasn't one at Stanford at the time, but that there's part of
+
+02:04.560 --> 02:09.200
+ philosophy and in fact Stanford has a special sub major in something called now Symbolic Systems,
+
+02:09.200 --> 02:15.520
+ which is logic, model, theory, formal semantics of natural language. And so that's actually
+
+02:15.520 --> 02:19.680
+ a perfect preparation for work in AI and computer science.
+
+02:19.680 --> 02:23.760
+ That's kind of interesting. So if you were interested in artificial intelligence,
+
+02:26.000 --> 02:32.560
+ what kind of majors were people even thinking about taking? What is in your science? So besides
+
+02:32.560 --> 02:37.120
+ philosophies, what were you supposed to do if you were fascinated by the idea of creating
+
+02:37.120 --> 02:41.440
+ intelligence? There weren't enough people who did that for that even to be a conversation.
+
+02:41.440 --> 02:49.920
+ I mean, I think probably philosophy. I mean, it's interesting in my graduating class of
+
+02:49.920 --> 02:57.120
+ undergraduate philosophers, probably maybe slightly less than half went on in computer
+
+02:57.120 --> 03:02.240
+ science, slightly less than half went on in law, and like one or two went on in philosophy.
+
+03:03.360 --> 03:07.920
+ So it was a common kind of connection. Do you think AI researchers have a role,
+
+03:07.920 --> 03:12.480
+ be part time philosophers, or should they stick to the solid science and engineering
+
+03:12.480 --> 03:17.200
+ without sort of taking the philosophizing tangents? I mean, you work with robots,
+
+03:17.200 --> 03:22.960
+ you think about what it takes to create intelligent beings. Aren't you the perfect person to think
+
+03:22.960 --> 03:27.440
+ about the big picture philosophy at all? The parts of philosophy that are closest to AI,
+
+03:27.440 --> 03:30.400
+ I think, or at least the closest to AI that I think about are stuff like
+
+03:30.400 --> 03:38.400
+ belief and knowledge and denotation and that kind of stuff. It's quite formal, and it's
+
+03:38.400 --> 03:44.000
+ like just one step away from the kinds of computer science work that we do kind of routinely.
+
+03:45.680 --> 03:53.040
+ I think that there are important questions still about what you can do with a machine and what
+
+03:53.040 --> 03:57.680
+ you can't and so on. Although at least my personal view is that I'm completely a materialist,
+
+03:57.680 --> 04:01.920
+ and I don't think that there's any reason why we can't make a robot be
+
+04:02.800 --> 04:06.800
+ behaviorally indistinguishable from a human. And the question of whether it's
+
+04:08.480 --> 04:13.600
+ distinguishable internally, whether it's a zombie or not in philosophy terms, I actually don't,
+
+04:14.720 --> 04:16.960
+ I don't know, and I don't know if I care too much about that.
+
+04:16.960 --> 04:22.080
+ Right, but there is a philosophical notions there, mathematical and philosophical,
+
+04:22.080 --> 04:27.520
+ because we don't know so much of how difficult that is, how difficult is a perception problem.
+
+04:27.520 --> 04:32.640
+ How difficult is the planning problem? How difficult is it to operate in this world successfully?
+
+04:32.640 --> 04:37.920
+ Because our robots are not currently as successful as human beings in many tasks.
+
+04:37.920 --> 04:44.320
+ The question about the gap between current robots and human beings borders a little bit
+
+04:44.320 --> 04:52.400
+ on philosophy. The expanse of knowledge that's required to operate in this world and the ability
+
+04:52.400 --> 04:57.280
+ to form common sense knowledge, the ability to reason about uncertainty, much of the work
+
+04:57.280 --> 05:05.040
+ you've been doing, there's open questions there that, I don't know, require to activate a certain
+
+05:06.320 --> 05:09.840
+ big picture view. To me, that doesn't seem like a philosophical gap at all.
+
+05:10.640 --> 05:14.240
+ To me, there is a big technical gap. There's a huge technical gap,
+
+05:15.040 --> 05:19.360
+ but I don't see any reason why it's more than a technical gap.
+
+05:19.360 --> 05:28.400
+ Perfect. When you mentioned AI, you mentioned SRI, and maybe can you describe to me when you
+
+05:28.400 --> 05:37.680
+ first fell in love with robotics, with robots, or inspired, so you mentioned flaky or shaky flaky,
+
+05:38.400 --> 05:42.720
+ and what was the robot that first captured your imagination of what's possible?
+
+05:42.720 --> 05:47.920
+ Right. The first robot I worked with was flaky. Shaky was a robot that the SRI people had built,
+
+05:47.920 --> 05:53.360
+ but by the time, I think when I arrived, it was sitting in a corner of somebody's office
+
+05:53.360 --> 06:00.640
+ dripping hydraulic fluid into a pan, but it's iconic. Really, everybody should read the Shaky
+
+06:00.640 --> 06:07.840
+ Tech Report because it has so many good ideas in it. They invented ASTAR Search and symbolic
+
+06:07.840 --> 06:15.520
+ planning and learning macro operators. They had low level kind of configuration space planning for
+
+06:15.520 --> 06:20.160
+ the robot. They had vision. That's the basic ideas of a ton of things.
+
+06:20.160 --> 06:27.920
+ Can you take a step by it? Shaky was a mobile robot, but it could push objects,
+
+06:27.920 --> 06:31.680
+ and so it would move things around. With which actuator?
+
+06:31.680 --> 06:40.080
+ With its self, with its base. They had painted the baseboards black,
+
+06:40.080 --> 06:48.320
+ so it used vision to localize itself in a map. It detected objects. It could detect objects that
+
+06:48.320 --> 06:54.800
+ were surprising to it. It would plan and replan based on what it saw. It reasoned about whether
+
+06:54.800 --> 07:02.240
+ to look and take pictures. It really had the basics of so many of the things that we think about now.
+
+07:03.280 --> 07:05.360
+ How did it represent the space around it?
+
+07:05.360 --> 07:09.680
+ It had representations at a bunch of different levels of abstraction,
+
+07:09.680 --> 07:13.920
+ so it had, I think, a kind of an occupancy grid of some sort at the lowest level.
+
+07:14.880 --> 07:20.000
+ At the high level, it was abstract, symbolic kind of rooms and connectivity.
+
+07:20.000 --> 07:22.160
+ So where does Flakey come in?
+
+07:22.160 --> 07:29.600
+ Yeah, okay. I showed up at SRI and we were building a brand new robot. As I said, none of the people
+
+07:29.600 --> 07:34.240
+ from the previous project were there or involved anymore, so we were starting from scratch.
+
+07:34.240 --> 07:43.920
+ My advisor was Stan Rosenstein. He ended up being my thesis advisor. He was motivated by this idea
+
+07:43.920 --> 07:52.400
+ of situated computation or situated automata. The idea was that the tools of logical reasoning were
+
+07:52.400 --> 08:01.200
+ important, but possibly only for the engineers or designers to use in the analysis of a system,
+
+08:01.200 --> 08:05.600
+ but not necessarily to be manipulated in the head of the system itself.
+
+08:06.400 --> 08:09.920
+ So I might use logic to prove a theorem about the behavior of my robot,
+
+08:10.480 --> 08:14.400
+ even if the robot's not using logic, and it's had to prove theorems. So that was kind of the
+
+08:14.400 --> 08:22.160
+ distinction. And so the idea was to kind of use those principles to make a robot do stuff.
+
+08:22.800 --> 08:28.960
+ But a lot of the basic things we had to kind of learn for ourselves, because I had zero
+
+08:28.960 --> 08:32.160
+ background in robotics. I didn't know anything about control. I didn't know anything about
+
+08:32.160 --> 08:36.640
+ sensors. So we reinvented a lot of wheels on the way to getting that robot to do stuff.
+
+08:36.640 --> 08:39.120
+ Do you think that was an advantage or hindrance?
+
+08:39.120 --> 08:45.600
+ Oh, no. I'm big in favor of wheel reinvention, actually. I mean, I think you learned a lot
+
+08:45.600 --> 08:51.920
+ by doing it. It's important though to eventually have the pointers so that you can see what's
+
+08:51.920 --> 08:58.080
+ really going on. But I think you can appreciate much better the good solutions once you've
+
+08:58.080 --> 09:00.400
+ messed around a little bit on your own and found a bad one.
+
+09:00.400 --> 09:04.880
+ Yeah, I think you mentioned reinventing reinforcement learning and referring to
+
+09:04.880 --> 09:10.960
+ rewards as pleasures, a pleasure, I think, which I think is a nice name for it.
+
+09:12.800 --> 09:18.960
+ It's more fun, almost. Do you think you could tell the history of AI, machine learning,
+
+09:18.960 --> 09:23.520
+ reinforcement learning, how you think about it from the 50s to now?
+
+09:23.520 --> 09:29.440
+ One thing is that it oscillates. So things become fashionable and then they go out and
+
+09:29.440 --> 09:34.480
+ then something else becomes cool and then it goes out and so on. So there's some interesting
+
+09:34.480 --> 09:41.600
+ sociological process that actually drives a lot of what's going on. Early days was cybernetics and
+
+09:41.600 --> 09:48.320
+ control and the idea that of homeostasis, people who made these robots that could,
+
+09:48.320 --> 09:54.400
+ I don't know, try to plug into the wall when they needed power and then come loose and roll
+
+09:54.400 --> 10:00.960
+ around and do stuff. And then I think over time, they thought, well, that was inspiring, but people
+
+10:00.960 --> 10:04.880
+ said, no, no, no, we want to get maybe closer to what feels like real intelligence or human
+
+10:04.880 --> 10:15.040
+ intelligence. And then maybe the expert systems people tried to do that, but maybe a little
+
+10:15.040 --> 10:21.760
+ too superficially. So we get this surface understanding of what intelligence is like,
+
+10:21.760 --> 10:25.840
+ because I understand how a steel mill works and I can try to explain it to you and you can write
+
+10:25.840 --> 10:31.520
+ it down in logic and then we can make a computer infer that. And then that didn't work out.
+
+10:32.400 --> 10:37.520
+ But what's interesting, I think, is when a thing starts to not be working very well,
+
+10:38.720 --> 10:44.480
+ it's not only do we change methods, we change problems. So it's not like we have better ways
+
+10:44.480 --> 10:48.160
+ of doing the problem of the expert systems people are trying to do. We have no ways of
+
+10:48.160 --> 10:56.800
+ trying to do that problem. Oh, yeah, no, I think maybe a few. But we kind of give up on that problem
+
+10:56.800 --> 11:01.520
+ and we switch to a different problem. And we work that for a while and we make progress.
+
+11:01.520 --> 11:04.960
+ As a broad community. As a community. And there's a lot of people who would argue,
+
+11:04.960 --> 11:09.760
+ you don't give up on the problem. It's just the decrease in the number of people working on it.
+
+11:09.760 --> 11:13.920
+ You almost kind of like put it on the shelf. So we'll come back to this 20 years later.
+
+11:13.920 --> 11:19.360
+ Yeah, I think that's right. Or you might decide that it's malformed. Like you might say,
+
+11:21.600 --> 11:26.800
+ it's wrong to just try to make something that does superficial symbolic reasoning behave like a
+
+11:26.800 --> 11:34.000
+ doctor. You can't do that until you've had the sensory motor experience of being a doctor or
+
+11:34.000 --> 11:38.560
+ something. So there's arguments that say that that problem was not well formed. Or it could be
+
+11:38.560 --> 11:44.160
+ that it is well formed, but we just weren't approaching it well. So you mentioned that your
+
+11:44.160 --> 11:49.120
+ favorite part of logic and symbolic systems is that they give short names for large sets.
+
+11:49.840 --> 11:56.320
+ So there is some use to this. They use symbolic reasoning. So looking at expert systems
+
+11:56.960 --> 12:01.760
+ and symbolic computing, what do you think are the roadblocks that were hit in the 80s and 90s?
+
+12:02.640 --> 12:08.320
+ Okay, so right. So the fact that I'm not a fan of expert systems doesn't mean that I'm not a fan
+
+12:08.320 --> 12:16.640
+ of some kind of symbolic reasoning. So let's see roadblocks. Well, the main roadblock, I think,
+
+12:16.640 --> 12:25.040
+ was that the idea that humans could articulate their knowledge effectively into some kind of
+
+12:25.040 --> 12:30.560
+ logical statements. So it's not just the cost, the effort, but really just the capability of
+
+12:30.560 --> 12:37.120
+ doing it. Right. Because we're all experts in vision, but totally don't have introspective access
+
+12:37.120 --> 12:45.440
+ into how we do that. Right. And it's true that, I mean, I think the idea was, well, of course,
+
+12:45.440 --> 12:49.040
+ even people then would know, of course, I wouldn't ask you to please write down the rules that you
+
+12:49.040 --> 12:54.000
+ use for recognizing a water bottle. That's crazy. And everyone understood that. But we might ask
+
+12:54.000 --> 13:00.800
+ you to please write down the rules you use for deciding, I don't know what tie to put on or
+
+13:00.800 --> 13:08.240
+ or how to set up a microphone or something like that. But even those things, I think people maybe,
+
+13:08.880 --> 13:12.720
+ I think what they found, I'm not sure about this, but I think what they found was that the
+
+13:12.720 --> 13:19.120
+ so called experts could give explanations that sort of post hoc explanations for how and why
+
+13:19.120 --> 13:27.680
+ they did things, but they weren't necessarily very good. And then they depended on maybe some
+
+13:27.680 --> 13:33.280
+ kinds of perceptual things, which again, they couldn't really define very well. So I think,
+
+13:33.280 --> 13:38.800
+ I think fundamentally, I think that the underlying problem with that was the assumption that people
+
+13:38.800 --> 13:45.280
+ could articulate how and why they make their decisions. Right. So it's almost encoding the
+
+13:45.280 --> 13:51.440
+ knowledge from converting from expert to something that a machine can understand and reason with.
+
+13:51.440 --> 13:58.880
+ No, no, no, not even just encoding, but getting it out of you. Not not not writing it. I mean,
+
+13:58.880 --> 14:03.680
+ yes, hard also to write it down for the computer. But I don't think that people can
+
+14:04.240 --> 14:10.080
+ produce it. You can tell me a story about why you do stuff. But I'm not so sure that's the why.
+
+14:11.440 --> 14:16.960
+ Great. So there are still on the hierarchical planning side,
+
+14:16.960 --> 14:25.120
+ places where symbolic reasoning is very useful. So as you've talked about, so
+
+14:27.840 --> 14:34.400
+ where so don't where's the gap? Yeah, okay, good. So saying that humans can't provide a
+
+14:34.400 --> 14:40.560
+ description of their reasoning processes. That's okay, fine. But that doesn't mean that it's not
+
+14:40.560 --> 14:44.880
+ good to do reasoning of various styles inside a computer. Those are just two orthogonal points.
+
+14:44.880 --> 14:50.560
+ So then the question is, what kind of reasoning should you do inside a computer?
+
+14:50.560 --> 14:55.680
+ Right. And the answer is, I think you need to do all different kinds of reasoning inside
+
+14:55.680 --> 15:01.680
+ a computer, depending on what kinds of problems you face. I guess the question is, what kind of
+
+15:01.680 --> 15:12.880
+ things can you encode symbolically so you can reason about? I think the idea about and and
+
+15:12.880 --> 15:18.080
+ even symbolic, I don't even like that terminology because I don't know what it means technically
+
+15:18.080 --> 15:24.240
+ and formally. I do believe in abstractions. So abstractions are critical, right? You cannot
+
+15:24.240 --> 15:30.240
+ reason at completely fine grain about everything in your life, right? You can't make a plan at the
+
+15:30.240 --> 15:37.680
+ level of images and torques for getting a PhD. So you have to reduce the size of the state space
+
+15:37.680 --> 15:43.040
+ and you have to reduce the horizon if you're going to reason about getting a PhD or even buying
+
+15:43.040 --> 15:50.080
+ the ingredients to make dinner. And so how can you reduce the spaces and the horizon of the
+
+15:50.080 --> 15:53.200
+ reasoning you have to do? And the answer is abstraction, spatial abstraction, temporal
+
+15:53.200 --> 15:58.080
+ abstraction. I think abstraction along the lines of goals is also interesting, like you might
+
+15:58.800 --> 16:03.840
+ or well, abstraction and decomposition. Goals is maybe more of a decomposition thing.
+
+16:03.840 --> 16:08.880
+ So I think that's where these kinds of, if you want to call it symbolic or discrete
+
+16:08.880 --> 16:15.440
+ models come in. You talk about a room of your house instead of your pose. You talk about
+
+16:16.800 --> 16:22.560
+ doing something during the afternoon instead of at 2.54. And you do that because it makes
+
+16:22.560 --> 16:30.000
+ your reasoning problem easier and also because you don't have enough information
+
+16:30.000 --> 16:37.120
+ to reason in high fidelity about your pose of your elbow at 2.35 this afternoon anyway.
+
+16:37.120 --> 16:39.440
+ Right. When you're trying to get a PhD.
+
+16:39.440 --> 16:41.600
+ Right. Or when you're doing anything really.
+
+16:41.600 --> 16:44.400
+ Yeah, okay. Except for at that moment. At that moment,
+
+16:44.400 --> 16:48.160
+ you do have to reason about the pose of your elbow, maybe. But then maybe you do that in some
+
+16:48.160 --> 16:55.680
+ continuous joint space kind of model. And so again, my biggest point about all of this is that
+
+16:55.680 --> 17:01.440
+ there should be, the dogma is not the thing, right? It shouldn't be that I am in favor
+
+17:01.440 --> 17:06.320
+ against symbolic reasoning and you're in favor against neural networks. It should be that just
+
+17:07.600 --> 17:12.240
+ computer science tells us what the right answer to all these questions is if we were smart enough
+
+17:12.240 --> 17:16.960
+ to figure it out. Yeah. When you try to actually solve the problem with computers, the right answer
+
+17:16.960 --> 17:22.880
+ comes out. You mentioned abstractions. I mean, neural networks form abstractions or rather,
+
+17:22.880 --> 17:30.320
+ there's automated ways to form abstractions and there's expert driven ways to form abstractions
+
+17:30.320 --> 17:35.920
+ and expert human driven ways. And humans just seems to be way better at forming abstractions
+
+17:35.920 --> 17:44.080
+ currently and certain problems. So when you're referring to 2.45 a.m. versus afternoon,
+
+17:44.960 --> 17:49.920
+ how do we construct that taxonomy? Is there any room for automated construction of such
+
+17:49.920 --> 17:55.200
+ abstractions? Oh, I think eventually, yeah. I mean, I think when we get to be better
+
+17:56.160 --> 18:02.240
+ and machine learning engineers, we'll build algorithms that build awesome abstractions.
+
+18:02.240 --> 18:06.720
+ That are useful in this kind of way that you're describing. Yeah. So let's then step from
+
+18:07.840 --> 18:14.400
+ the abstraction discussion and let's talk about BOMMDP's
+
+18:14.400 --> 18:21.440
+ Partially Observable Markov Decision Processes. So uncertainty. So first, what are Markov Decision
+
+18:21.440 --> 18:27.520
+ Processes? What are Markov Decision Processes? Maybe how much of our world can be models and
+
+18:27.520 --> 18:32.080
+ MDPs? How much when you wake up in the morning and you're making breakfast, how do you think
+
+18:32.080 --> 18:38.080
+ of yourself as an MDP? So how do you think about MDPs and how they relate to our world?
+
+18:38.080 --> 18:43.040
+ Well, so there's a stance question, right? So a stance is a position that I take with
+
+18:43.040 --> 18:52.160
+ respect to a problem. So I as a researcher or a person who designed systems can decide to make
+
+18:52.160 --> 18:58.960
+ a model of the world around me in some terms. So I take this messy world and I say, I'm going to
+
+18:58.960 --> 19:04.640
+ treat it as if it were a problem of this formal kind, and then I can apply solution concepts
+
+19:04.640 --> 19:09.120
+ or algorithms or whatever to solve that formal thing, right? So of course, the world is not
+
+19:09.120 --> 19:14.080
+ anything. It's not an MDP or a POMDP. I don't know what it is, but I can model aspects of it
+
+19:14.080 --> 19:19.280
+ in some way or some other way. And when I model some aspect of it in a certain way, that gives me
+
+19:19.280 --> 19:25.600
+ some set of algorithms I can use. You can model the world in all kinds of ways. Some have some
+
+19:26.400 --> 19:32.880
+ are more accepting of uncertainty, more easily modeling uncertainty of the world. Some really
+
+19:32.880 --> 19:40.720
+ force the world to be deterministic. And so certainly MDPs model the uncertainty of the world.
+
+19:40.720 --> 19:47.200
+ Yes. Model some uncertainty. They model not present state uncertainty, but they model uncertainty
+
+19:47.200 --> 19:53.840
+ in the way the future will unfold. Right. So what are Markov decision processes?
+
+19:53.840 --> 19:57.680
+ So Markov decision process is a model. It's a kind of a model that you can make that says,
+
+19:57.680 --> 20:05.600
+ I know completely the current state of my system. And what it means to be a state is that I have
+
+20:05.600 --> 20:10.720
+ all the information right now that will let me make predictions about the future as well as I
+
+20:10.720 --> 20:14.640
+ can. So that remembering anything about my history wouldn't make my predictions any better.
+
+20:18.720 --> 20:23.680
+ But then it also says that then I can take some actions that might change the state of the world
+
+20:23.680 --> 20:28.800
+ and that I don't have a deterministic model of those changes. I have a probabilistic model
+
+20:28.800 --> 20:35.600
+ of how the world might change. It's a useful model for some kinds of systems. I mean, it's
+
+20:35.600 --> 20:43.280
+ certainly not a good model for most problems. I think because for most problems, you don't
+
+20:43.280 --> 20:49.680
+ actually know the state. For most problems, it's partially observed. So that's now a different
+
+20:49.680 --> 20:56.480
+ problem class. So okay, that's where the problem depies, the partially observed Markov decision
+
+20:56.480 --> 21:03.600
+ process step in. So how do they address the fact that you can't observe most the incomplete
+
+21:03.600 --> 21:09.360
+ information about most of the world around you? Right. So now the idea is we still kind of postulate
+
+21:09.360 --> 21:14.080
+ that there exists a state. We think that there is some information about the world out there
+
+21:14.640 --> 21:18.800
+ such that if we knew that we could make good predictions, but we don't know the state.
+
+21:18.800 --> 21:23.840
+ And so then we have to think about how, but we do get observations. Maybe I get images or I hear
+
+21:23.840 --> 21:29.520
+ things or I feel things and those might be local or noisy. And so therefore they don't tell me
+
+21:29.520 --> 21:35.440
+ everything about what's going on. And then I have to reason about given the history of actions
+
+21:35.440 --> 21:40.000
+ I've taken and observations I've gotten, what do I think is going on in the world? And then
+
+21:40.000 --> 21:43.920
+ given my own kind of uncertainty about what's going on in the world, I can decide what actions to
+
+21:43.920 --> 21:51.120
+ take. And so difficult is this problem of planning under uncertainty in your view and your
+
+21:51.120 --> 21:57.840
+ long experience of modeling the world, trying to deal with this uncertainty in
+
+21:57.840 --> 22:04.240
+ especially in real world systems. Optimal planning for even discrete POMDPs can be
+
+22:04.240 --> 22:12.000
+ undecidable depending on how you set it up. And so lots of people say I don't use POMDPs
+
+22:12.000 --> 22:17.600
+ because they are intractable. And I think that that's a kind of a very funny thing to say because
+
+22:18.880 --> 22:23.120
+ the problem you have to solve is the problem you have to solve. So if the problem you have to
+
+22:23.120 --> 22:28.160
+ solve is intractable, that's what makes us AI people, right? So we solve, we understand that
+
+22:28.160 --> 22:34.320
+ the problem we're solving is wildly intractable that we will never be able to solve it optimally,
+
+22:34.320 --> 22:41.360
+ at least I don't. Yeah, right. So later we can come back to an idea about bounded optimality
+
+22:41.360 --> 22:44.960
+ and something. But anyway, we can't come up with optimal solutions to these problems.
+
+22:45.520 --> 22:51.200
+ So we have to make approximations. Approximations in modeling approximations in solution algorithms
+
+22:51.200 --> 22:58.160
+ and so on. And so I don't have a problem with saying, yeah, my problem actually it is POMDP in
+
+22:58.160 --> 23:02.880
+ continuous space with continuous observations. And it's so computationally complex. I can't
+
+23:02.880 --> 23:10.320
+ even think about it's, you know, big O whatever. But that doesn't prevent me from it helps me
+
+23:10.320 --> 23:17.360
+ gives me some clarity to think about it that way. And to then take steps to make approximation
+
+23:17.360 --> 23:22.080
+ after approximation to get down to something that's like computable in some reasonable time.
+
+23:22.080 --> 23:27.920
+ When you think about optimality, you know, the community broadly has shifted on that, I think,
+
+23:27.920 --> 23:35.600
+ a little bit in how much they value the idea of optimality of chasing an optimal solution.
+
+23:35.600 --> 23:42.240
+ How is your views of chasing an optimal solution changed over the years when you work with robots?
+
+23:42.240 --> 23:49.920
+ That's interesting. I think we have a little bit of a methodological crisis, actually,
+
+23:49.920 --> 23:54.000
+ from the theoretical side. I mean, I do think that theory is important and that right now we're not
+
+23:54.000 --> 24:00.640
+ doing much of it. So there's lots of empirical hacking around and training this and doing that
+
+24:00.640 --> 24:05.440
+ and reporting numbers. But is it good? Is it bad? We don't know. It's very hard to say things.
+
+24:08.240 --> 24:15.920
+ And if you look at like computer science theory, so people talked for a while,
+
+24:15.920 --> 24:21.280
+ everyone was about solving problems optimally or completely. And then there were interesting
+
+24:21.280 --> 24:27.520
+ relaxations. So people look at, oh, can I, are there regret bounds? Or can I do some kind of,
+
+24:27.520 --> 24:33.280
+ you know, approximation? Can I prove something that I can approximately solve this problem or
+
+24:33.280 --> 24:38.160
+ that I get closer to the solution as I spend more time and so on? What's interesting, I think,
+
+24:38.160 --> 24:47.680
+ is that we don't have good approximate solution concepts for very difficult problems. Right?
+
+24:47.680 --> 24:52.640
+ I like to, you know, I like to say that I'm interested in doing a very bad job of very big
+
+24:52.640 --> 25:02.960
+ problems. Right. So very bad job, very big problems. I like to do that. But I wish I could say
+
+25:02.960 --> 25:09.600
+ something. I wish I had a, I don't know, some kind of a formal solution concept
+
+25:10.320 --> 25:16.640
+ that I could use to say, oh, this algorithm actually, it gives me something. Like, I know
+
+25:16.640 --> 25:21.760
+ what I'm going to get. I can do something other than just run it and get out. So that notion
+
+25:21.760 --> 25:28.640
+ is still somewhere deeply compelling to you. The notion that you can say, you can drop
+
+25:28.640 --> 25:33.440
+ thing on the table says this, you can expect this, this algorithm will give me some good results.
+
+25:33.440 --> 25:38.960
+ I hope there's, I hope science will, I mean, there's engineering and there's science,
+
+25:38.960 --> 25:44.720
+ I think that they're not exactly the same. And I think right now we're making huge engineering
+
+25:45.600 --> 25:49.840
+ like leaps and bounds. So the engineering is running away ahead of the science, which is cool.
+
+25:49.840 --> 25:54.800
+ And often how it goes, right? So we're making things and nobody knows how and why they work,
+
+25:54.800 --> 26:03.200
+ roughly. But we need to turn that into science. There's some form. It's, yeah,
+
+26:03.200 --> 26:07.200
+ there's some room for formalizing. We need to know what the principles are. Why does this work?
+
+26:07.200 --> 26:12.480
+ Why does that not work? I mean, for while people build bridges by trying, but now we can often
+
+26:12.480 --> 26:17.520
+ predict whether it's going to work or not without building it. Can we do that for learning systems
+
+26:17.520 --> 26:23.600
+ or for robots? See, your hope is from a materialistic perspective that intelligence,
+
+26:23.600 --> 26:28.000
+ artificial intelligence systems, robots are kind of just fancier bridges.
+
+26:29.200 --> 26:33.600
+ Belief space. What's the difference between belief space and state space? So we mentioned
+
+26:33.600 --> 26:42.000
+ MDPs, FOMDPs, you reasoning about, you sense the world, there's a state. What's this belief
+
+26:42.000 --> 26:49.040
+ space idea? Yeah. Okay, that sounds good. It sounds good. So belief space, that is, instead of
+
+26:49.040 --> 26:54.880
+ thinking about what's the state of the world and trying to control that as a robot, I think about
+
+26:55.760 --> 27:01.120
+ what is the space of beliefs that I could have about the world? What's, if I think of a belief
+
+27:01.120 --> 27:06.640
+ as a probability distribution of the ways the world could be, a belief state is a distribution,
+
+27:06.640 --> 27:13.040
+ and then my control problem, if I'm reasoning about how to move through a world I'm uncertain about,
+
+27:14.160 --> 27:18.880
+ my control problem is actually the problem of controlling my beliefs. So I think about taking
+
+27:18.880 --> 27:23.120
+ actions, not just what effect they'll have on the world outside, but what effect they'll have on my
+
+27:23.120 --> 27:29.920
+ own understanding of the world outside. And so that might compel me to ask a question or look
+
+27:29.920 --> 27:35.280
+ somewhere to gather information, which may not really change the world state, but it changes
+
+27:35.280 --> 27:43.440
+ my own belief about the world. That's a powerful way to empower the agent to reason about the
+
+27:43.440 --> 27:47.840
+ world, to explore the world. What kind of problems does it allow you to solve to
+
+27:49.040 --> 27:54.560
+ consider belief space versus just state space? Well, any problem that requires deliberate
+
+27:54.560 --> 28:02.800
+ information gathering. So if in some problems, like chess, there's no uncertainty, or maybe
+
+28:02.800 --> 28:06.320
+ there's uncertainty about the opponent. There's no uncertainty about the state.
+
+28:08.400 --> 28:14.000
+ And some problems, there's uncertainty, but you gather information as you go. You might say,
+
+28:14.000 --> 28:18.240
+ oh, I'm driving my autonomous car down the road, and it doesn't know perfectly where it is, but
+
+28:18.240 --> 28:23.280
+ the LiDARs are all going all the time. So I don't have to think about whether to gather information.
+
+28:24.160 --> 28:28.800
+ But if you're a human driving down the road, you sometimes look over your shoulder to see what's
+
+28:28.800 --> 28:36.320
+ going on behind you in the lane. And you have to decide whether you should do that now. And you
+
+28:36.320 --> 28:40.400
+ have to trade off the fact that you're not seeing in front of you, and you're looking behind you,
+
+28:40.400 --> 28:45.440
+ and how valuable is that information, and so on. And so to make choices about information
+
+28:45.440 --> 28:56.080
+ gathering, you have to reason in belief space. Also to just take into account your own uncertainty
+
+28:56.080 --> 29:03.280
+ before trying to do things. So you might say, if I understand where I'm standing relative to the
+
+29:03.280 --> 29:08.880
+ door jam, pretty accurately, then it's okay for me to go through the door. But if I'm really not
+
+29:08.880 --> 29:14.240
+ sure where the door is, then it might be better to not do that right now. The degree of your
+
+29:14.240 --> 29:18.800
+ uncertainty about the world is actually part of the thing you're trying to optimize in forming the
+
+29:18.800 --> 29:26.560
+ plan, right? So this idea of a long horizon of planning for a PhD or just even how to get out
+
+29:26.560 --> 29:32.720
+ of the house or how to make breakfast, you show this presentation of the WTF, where's the fork
+
+29:33.360 --> 29:42.000
+ of a robot looking to sink. And can you describe how we plan in this world is this idea of hierarchical
+
+29:42.000 --> 29:52.000
+ planning we've mentioned? Yeah, how can a robot hope to plan about something with such a long
+
+29:52.000 --> 29:58.400
+ horizon where the goal is quite far away? People since probably reasoning began have thought about
+
+29:58.400 --> 30:02.560
+ hierarchical reasoning, the temporal hierarchy in particular. Well, there's spatial hierarchy,
+
+30:02.560 --> 30:06.240
+ but let's talk about temporal hierarchy. So you might say, oh, I have this long
+
+30:06.240 --> 30:13.680
+ execution I have to do, but I can divide it into some segments abstractly, right? So maybe
+
+30:14.400 --> 30:19.360
+ have to get out of the house, I have to get in the car, I have to drive, and so on. And so
+
+30:20.800 --> 30:25.920
+ you can plan if you can build abstractions. So this we started out by talking about abstractions,
+
+30:25.920 --> 30:30.080
+ and we're back to that now. If you can build abstractions in your state space,
+
+30:30.080 --> 30:37.760
+ and abstractions, sort of temporal abstractions, then you can make plans at a high level. And you
+
+30:37.760 --> 30:42.320
+ can say, I'm going to go to town, and then I'll have to get gas, and I can go here, and I can do
+
+30:42.320 --> 30:47.360
+ this other thing. And you can reason about the dependencies and constraints among these actions,
+
+30:47.920 --> 30:55.600
+ again, without thinking about the complete details. What we do in our hierarchical planning work is
+
+30:55.600 --> 31:00.960
+ then say, all right, I make a plan at a high level of abstraction. I have to have some
+
+31:00.960 --> 31:06.640
+ reason to think that it's feasible without working it out in complete detail. And that's
+
+31:06.640 --> 31:10.800
+ actually the interesting step. I always like to talk about walking through an airport, like
+
+31:12.160 --> 31:16.720
+ you can plan to go to New York and arrive at the airport, and then find yourself in an office
+
+31:16.720 --> 31:21.520
+ building later. You can't even tell me in advance what your plan is for walking through the airport,
+
+31:21.520 --> 31:26.320
+ partly because you're too lazy to think about it maybe, but partly also because you just don't
+
+31:26.320 --> 31:30.960
+ have the information. You don't know what gate you're landing in or what people are going to be
+
+31:30.960 --> 31:37.040
+ in front of you or anything. So there's no point in planning in detail. But you have to have,
+
+31:38.000 --> 31:43.760
+ you have to make a leap of faith that you can figure it out once you get there. And it's really
+
+31:43.760 --> 31:52.000
+ interesting to me how you arrive at that. How do you, so you have learned over your lifetime to be
+
+31:52.000 --> 31:56.800
+ able to make some kinds of predictions about how hard it is to achieve some kinds of sub goals.
+
+31:57.440 --> 32:01.440
+ And that's critical. Like you would never plan to fly somewhere if you couldn't,
+
+32:02.000 --> 32:05.200
+ didn't have a model of how hard it was to do some of the intermediate steps.
+
+32:05.200 --> 32:09.440
+ So one of the things we're thinking about now is how do you do this kind of very aggressive
+
+32:09.440 --> 32:16.400
+ generalization to situations that you haven't been in and so on to predict how long will it
+
+32:16.400 --> 32:20.400
+ take to walk through the Kuala Lumpur airport? Like you could give me an estimate and it wouldn't
+
+32:20.400 --> 32:26.800
+ be crazy. And you have to have an estimate of that in order to make plans that involve
+
+32:26.800 --> 32:30.080
+ walking through the Kuala Lumpur airport, even if you don't need to know it in detail.
+
+32:31.040 --> 32:35.520
+ So I'm really interested in these kinds of abstract models and how do we acquire them.
+
+32:35.520 --> 32:39.760
+ But once we have them, we can use them to do hierarchical reasoning, which I think is very
+
+32:39.760 --> 32:46.400
+ important. Yeah, there's this notion of goal regression and preimage backchaining.
+
+32:46.400 --> 32:53.760
+ This idea of starting at the goal and just forming these big clouds of states. I mean,
+
+32:54.560 --> 33:01.840
+ it's almost like saying to the airport, you know, you know, once you show up to the airport,
+
+33:01.840 --> 33:08.560
+ you're like a few steps away from the goal. So thinking of it this way is kind of interesting.
+
+33:08.560 --> 33:15.600
+ I don't know if you have further comments on that of starting at the goal. Yeah, I mean,
+
+33:15.600 --> 33:22.400
+ it's interesting that Herb Simon back in the early days of AI talked a lot about
+
+33:22.400 --> 33:26.960
+ means ends reasoning and reasoning back from the goal. There's a kind of an intuition that people
+
+33:26.960 --> 33:34.960
+ have that the number of the state space is big, the number of actions you could take is really big.
+
+33:35.760 --> 33:39.440
+ So if you say, here I sit and I want to search forward from where I am, what are all the things
+
+33:39.440 --> 33:45.040
+ I could do? That's just overwhelming. If you say, if you can reason at this other level and say,
+
+33:45.040 --> 33:49.520
+ here's what I'm hoping to achieve, what can I do to make that true that somehow the
+
+33:49.520 --> 33:54.000
+ branching is smaller? Now, what's interesting is that like in the AI planning community,
+
+33:54.000 --> 33:59.120
+ that hasn't worked out in the class of problems that they look at and the methods that they tend
+
+33:59.120 --> 34:04.400
+ to use, it hasn't turned out that it's better to go backward. It's still kind of my intuition
+
+34:04.400 --> 34:10.000
+ that it is, but I can't prove that to you right now. Right. I share your intuition, at least for us
+
+34:10.720 --> 34:19.920
+ mirror humans. Speaking of which, when you maybe now we take it and take a little step into that
+
+34:19.920 --> 34:27.280
+ philosophy circle, how hard would it, when you think about human life, you give those examples
+
+34:27.280 --> 34:32.400
+ often, how hard do you think it is to formulate human life as a planning problem or aspects of
+
+34:32.400 --> 34:37.600
+ human life? So when you look at robots, you're often trying to think about object manipulation,
+
+34:38.640 --> 34:46.240
+ tasks about moving a thing. When you take a slight step outside the room, let the robot
+
+34:46.240 --> 34:54.480
+ leave and he'll get lunch or maybe try to pursue more fuzzy goals. How hard do you think is that
+
+34:54.480 --> 35:00.720
+ problem? If you were to try to maybe put another way, try to formulate human life as a planning
+
+35:00.720 --> 35:05.680
+ problem. Well, that would be a mistake. I mean, it's not all a planning problem, right? I think
+
+35:05.680 --> 35:11.920
+ it's really, really important that we understand that you have to put together pieces and parts
+
+35:11.920 --> 35:18.640
+ that have different styles of reasoning and representation and learning. I think it seems
+
+35:18.640 --> 35:25.680
+ probably clear to anybody that it can't all be this or all be that. Brains aren't all like this
+
+35:25.680 --> 35:30.160
+ or all like that, right? They have different pieces and parts and substructure and so on.
+
+35:30.160 --> 35:34.400
+ So I don't think that there's any good reason to think that there's going to be like one true
+
+35:34.400 --> 35:39.600
+ algorithmic thing that's going to do the whole job. Just a bunch of pieces together,
+
+35:39.600 --> 35:48.160
+ design to solve a bunch of specific problems. Or maybe styles of problems. I mean,
+
+35:48.160 --> 35:52.880
+ there's probably some reasoning that needs to go on in image space. I think, again,
+
+35:55.840 --> 35:59.440
+ there's this model base versus model free idea, right? So in reinforcement learning,
+
+35:59.440 --> 36:06.000
+ people talk about, oh, should I learn? I could learn a policy just straight up a way of behaving.
+
+36:06.000 --> 36:11.360
+ I could learn it's popular in a value function. That's some kind of weird intermediate ground.
+
+36:13.360 --> 36:17.440
+ Or I could learn a transition model, which tells me something about the dynamics of the world.
+
+36:18.320 --> 36:22.560
+ If I take a, imagine that I learn a transition model and I couple it with a planner and I
+
+36:22.560 --> 36:29.520
+ draw a box around that, I have a policy again. It's just stored a different way, right?
+
+36:30.800 --> 36:34.560
+ But it's just as much of a policy as the other policy. It's just I've made, I think,
+
+36:34.560 --> 36:41.920
+ the way I see it is it's a time space trade off in computation, right? A more overt policy
+
+36:41.920 --> 36:47.680
+ representation. Maybe it takes more space, but maybe I can compute quickly what action I should
+
+36:47.680 --> 36:52.880
+ take. On the other hand, maybe a very compact model of the world dynamics plus a planner
+
+36:53.680 --> 36:58.240
+ lets me compute what action to take to just more slowly. There's no, I mean, I don't think,
+
+36:58.240 --> 37:04.240
+ there's no argument to be had. It's just like a question of what form of computation is best
+
+37:04.240 --> 37:12.720
+ for us. For the various sub problems. Right. So, and so like learning to do algebra manipulations
+
+37:12.720 --> 37:17.280
+ for some reason is, I mean, that's probably going to want naturally a sort of a different
+
+37:17.280 --> 37:22.640
+ representation than riding a unicycle. At the time constraints on the unicycle are serious.
+
+37:22.640 --> 37:28.640
+ The space is maybe smaller. I don't know. But so I could be the more human size of
+
+37:28.640 --> 37:36.240
+ falling in love, having a relationship that might be another another style of no idea how to model
+
+37:36.240 --> 37:43.280
+ that. Yeah, that's, that's first solve the algebra and the object manipulation. What do you think
+
+37:44.160 --> 37:50.480
+ is harder perception or planning perception? That's why I'm understanding that's why.
+
+37:51.920 --> 37:55.440
+ So what do you think is so hard about perception by understanding the world around you?
+
+37:55.440 --> 38:03.520
+ Well, I mean, I think the big question is representational. A hugely the question is
+
+38:03.520 --> 38:12.560
+ representation. So perception has made great strides lately, right? And we can classify images and we
+
+38:12.560 --> 38:17.760
+ can play certain kinds of games and predict how to steer the car and all this sort of stuff.
+
+38:17.760 --> 38:28.160
+ I don't think we have a very good idea of what perception should deliver, right? So if you
+
+38:28.160 --> 38:32.000
+ if you believe in modularity, okay, there's there's a very strong view which says
+
+38:34.640 --> 38:40.400
+ we shouldn't build in any modularity, we should make a giant gigantic neural network,
+
+38:40.400 --> 38:44.000
+ train it end to end to do the thing. And that's the best way forward.
+
+38:44.000 --> 38:51.280
+ And it's hard to argue with that except on a sample complexity basis, right? So you might say,
+
+38:51.280 --> 38:55.120
+ oh, well, if I want to do end to end reinforcement learning on this giant giant neural network,
+
+38:55.120 --> 39:00.800
+ it's going to take a lot of data and a lot of like broken robots and stuff. So
+
+39:02.640 --> 39:10.000
+ then the only answer is to say, okay, we have to build something in build in some structure
+
+39:10.000 --> 39:14.080
+ or some bias, we know from theory of machine learning, the only way to cut down the sample
+
+39:14.080 --> 39:20.320
+ complexity is to kind of cut down somehow cut down the hypothesis space, you can do that by
+
+39:20.320 --> 39:24.880
+ building in bias. There's all kinds of reason to think that nature built bias into humans.
+
+39:27.520 --> 39:32.800
+ Convolution is a bias, right? It's a very strong bias and it's a very critical bias.
+
+39:32.800 --> 39:39.520
+ So my view is that we should look for more things that are like convolution, but that address other
+
+39:39.520 --> 39:43.440
+ aspects of reasoning, right? So convolution helps us a lot with a certain kind of spatial
+
+39:43.440 --> 39:51.520
+ reasoning that's quite close to the imaging. I think there's other ideas like that,
+
+39:52.400 --> 39:58.240
+ maybe some amount of forward search, maybe some notions of abstraction, maybe the notion that
+
+39:58.240 --> 40:02.560
+ objects exist, actually, I think that's pretty important. And a lot of people won't give you
+
+40:02.560 --> 40:06.720
+ that to start with, right? So almost like a convolution in the
+
+40:08.640 --> 40:13.600
+ in the object semantic object space or some kind of some kind of ideas in there. That's right.
+
+40:13.600 --> 40:17.680
+ And people are like the graph, graph convolutions are an idea that are related to
+
+40:17.680 --> 40:26.240
+ relational representations. And so I think there are, so you, I've come far field from perception,
+
+40:26.240 --> 40:33.200
+ but I think, I think the thing that's going to make perception that kind of the next step is
+
+40:33.200 --> 40:38.000
+ actually understanding better what it should produce, right? So what are we going to do with
+
+40:38.000 --> 40:41.920
+ the output of it, right? It's fine when what we're going to do with the output is steer,
+
+40:41.920 --> 40:48.880
+ it's less clear when we're just trying to make one integrated intelligent agent,
+
+40:48.880 --> 40:53.520
+ what should the output of perception be? We have no idea. And how should that hook up to the other
+
+40:53.520 --> 41:00.240
+ stuff? We don't know. So I think the pressing question is, what kinds of structure can we
+
+41:00.240 --> 41:05.520
+ build in that are like the moral equivalent of convolution that will make a really awesome
+
+41:05.520 --> 41:10.240
+ superstructure that then learning can kind of progress on efficiently?
+
+41:10.240 --> 41:14.080
+ I agree. Very compelling description of actually where we stand with the perception from
+
+41:15.280 --> 41:19.120
+ you're teaching a course on embodying intelligence. What do you think it takes to
+
+41:19.120 --> 41:24.800
+ build a robot with human level intelligence? I don't know if we knew we would do it.
+
+41:27.680 --> 41:34.240
+ If you were to, I mean, okay, so do you think a robot needs to have a self awareness,
+
+41:36.000 --> 41:44.160
+ consciousness, fear of mortality? Or is it, is it simpler than that? Or is consciousness a simple
+
+41:44.160 --> 41:51.680
+ thing? Do you think about these notions? I don't think much about consciousness. Even most philosophers
+
+41:51.680 --> 41:56.560
+ who care about it will give you that you could have robots that are zombies, right, that behave
+
+41:56.560 --> 42:01.360
+ like humans but are not conscious. And I, at this moment, would be happy enough for that. So I'm not
+
+42:01.360 --> 42:06.240
+ really worried one way or the other. So the technical side, you're not thinking of the use of self
+
+42:06.240 --> 42:13.760
+ awareness? Well, but I, okay, but then what does self awareness mean? I mean, that you need to have
+
+42:13.760 --> 42:18.800
+ some part of the system that can observe other parts of the system and tell whether they're
+
+42:18.800 --> 42:24.560
+ working well or not. That seems critical. So does that count as, I mean, does that count as
+
+42:24.560 --> 42:30.560
+ self awareness or not? Well, it depends on whether you think that there's somebody at home who can
+
+42:30.560 --> 42:35.600
+ articulate whether they're self aware. But clearly, if I have like, you know, some piece of code
+
+42:35.600 --> 42:41.120
+ that's counting how many times this procedure gets executed, that's a kind of self awareness,
+
+42:41.120 --> 42:44.560
+ right? So there's a big spectrum. It's clear you have to have some of it.
+
+42:44.560 --> 42:48.800
+ Right. You know, we're quite far away on many dimensions, but is the direction of research
+
+42:49.600 --> 42:54.720
+ that's most compelling to you for, you know, trying to achieve human level intelligence
+
+42:54.720 --> 43:00.880
+ in our robots? Well, to me, I guess the thing that seems most compelling to me at the moment is this
+
+43:00.880 --> 43:10.320
+ question of what to build in and what to learn. I think we're, we don't, we're missing a bunch of
+
+43:10.320 --> 43:17.120
+ ideas. And, and we, you know, people, you know, don't you dare ask me how many years it's going
+
+43:17.120 --> 43:22.320
+ to be until that happens, because I won't even participate in the conversation. Because I think
+
+43:22.320 --> 43:26.240
+ we're missing ideas and I don't know how long it's going to take to find them. So I won't ask you
+
+43:26.240 --> 43:34.160
+ how many years, but maybe I'll ask you what it, when you will be sufficiently impressed that we've
+
+43:34.160 --> 43:41.280
+ achieved it. So what's a good test of intelligence? Do you like the Turing test and natural language
+
+43:41.280 --> 43:47.520
+ in the robotic space? Is there something where you would sit back and think, oh, that's pretty
+
+43:47.520 --> 43:52.800
+ impressive as a test, as a benchmark. Do you think about these kinds of problems?
+
+43:52.800 --> 43:58.480
+ No, I resist. I mean, I think all the time that we spend arguing about those kinds of things could
+
+43:58.480 --> 44:04.800
+ be better spent just making their robots work better. So you don't value competition. So I mean,
+
+44:04.800 --> 44:11.280
+ there's a nature of benchmark, benchmarks and data sets, or Turing test challenges, where
+
+44:11.280 --> 44:15.440
+ everybody kind of gets together and tries to build a better robot because they want to outcompete
+
+44:15.440 --> 44:21.360
+ each other, like the DARPA challenge with the autonomous vehicles. Do you see the value of that?
+
+44:23.600 --> 44:27.440
+ Or can get in the way? I think you can get in the way. I mean, some people, many people find it
+
+44:27.440 --> 44:34.880
+ motivating. And so that's good. I find it anti motivating personally. But I think you get an
+
+44:34.880 --> 44:41.200
+ interesting cycle where for a contest, a bunch of smart people get super motivated and they hack
+
+44:41.200 --> 44:47.120
+ their brains out. And much of what gets done is just hacks, but sometimes really cool ideas emerge.
+
+44:47.120 --> 44:53.360
+ And then that gives us something to chew on after that. So it's not a thing for me, but I don't
+
+44:53.360 --> 44:58.480
+ I don't regret that other people do it. Yeah, it's like you said, with everything else that
+
+44:58.480 --> 45:03.840
+ makes us good. So jumping topics a little bit, you started the journal machine learning research
+
+45:04.560 --> 45:09.920
+ and served as its editor in chief. How did the publication come about?
+
+45:11.680 --> 45:17.040
+ And what do you think about the current publishing model space in machine learning
+
+45:17.040 --> 45:23.040
+ artificial intelligence? Okay, good. So it came about because there was a journal called machine
+
+45:23.040 --> 45:29.840
+ learning, which still exists, which was owned by Clure. And there was I was on the editorial
+
+45:29.840 --> 45:33.520
+ board and we used to have these meetings annually where we would complain to Clure that
+
+45:33.520 --> 45:37.280
+ it was too expensive for the libraries and that people couldn't publish. And we would really
+
+45:37.280 --> 45:41.920
+ like to have some kind of relief on those fronts. And they would always sympathize,
+
+45:41.920 --> 45:49.120
+ but not do anything. So we just decided to make a new journal. And there was the Journal of AI
+
+45:49.120 --> 45:54.880
+ Research, which has was on the same model, which had been in existence for maybe five years or so,
+
+45:54.880 --> 46:01.920
+ and it was going on pretty well. So we just made a new journal. It wasn't I mean,
+
+46:03.600 --> 46:07.600
+ I don't know, I guess it was work, but it wasn't that hard. So basically the editorial board,
+
+46:07.600 --> 46:17.440
+ probably 75% of the editorial board of machine learning resigned. And we founded the new journal.
+
+46:17.440 --> 46:25.280
+ But it was sort of it was more open. Yeah, right. So it's completely open. It's open access.
+
+46:25.280 --> 46:31.600
+ Actually, I had a postdoc, George Conrad Harris, who wanted to call these journals free for all.
+
+46:33.520 --> 46:37.600
+ Because there were I mean, it both has no page charges and has no
+
+46:40.080 --> 46:45.520
+ access restrictions. And the reason and so lots of people, I mean, for there were,
+
+46:45.520 --> 46:50.240
+ there were people who are mad about the existence of this journal who thought it was a fraud or
+
+46:50.240 --> 46:55.200
+ something, it would be impossible, they said, to run a journal like this with basically,
+
+46:55.200 --> 46:59.200
+ I mean, for a long time, I didn't even have a bank account. I paid for the
+
+46:59.840 --> 47:06.640
+ lawyer to incorporate and the IP address. And it just didn't cost a couple hundred dollars a year
+
+47:06.640 --> 47:12.880
+ to run. It's a little bit more now, but not that much more. But that's because I think computer
+
+47:12.880 --> 47:19.920
+ scientists are competent and autonomous in a way that many scientists in other fields aren't.
+
+47:19.920 --> 47:23.920
+ I mean, at doing these kinds of things, we already type set around papers,
+
+47:23.920 --> 47:28.000
+ we all have students and people who can hack a website together in the afternoon.
+
+47:28.000 --> 47:32.960
+ So the infrastructure for us was like, not a problem, but for other people in other fields,
+
+47:32.960 --> 47:38.960
+ it's a harder thing to do. Yeah. And this kind of open access journal is nevertheless,
+
+47:38.960 --> 47:45.840
+ one of the most prestigious journals. So it's not like a prestige and it can be achieved
+
+47:45.840 --> 47:49.920
+ without any of the papers. Paper is not required for prestige, turns out. Yeah.
+
+47:50.640 --> 47:56.960
+ So on the review process side of actually a long time ago, I don't remember when I reviewed a paper
+
+47:56.960 --> 48:01.360
+ where you were also a reviewer and I remember reading your review and being influenced by it.
+
+48:01.360 --> 48:06.480
+ It was really well written. It influenced how I write feature reviews. You disagreed with me,
+
+48:06.480 --> 48:15.280
+ actually. And you made it my review, but much better. But nevertheless, the review process
+
+48:16.880 --> 48:23.600
+ has its flaws. And what do you think works well? How can it be improved?
+
+48:23.600 --> 48:27.600
+ So actually, when I started JMLR, I wanted to do something completely different.
+
+48:28.720 --> 48:34.800
+ And I didn't because it felt like we needed a traditional journal of record and so we just
+
+48:34.800 --> 48:40.800
+ made JMLR be almost like a normal journal, except for the open access parts of it, basically.
+
+48:43.200 --> 48:47.600
+ Increasingly, of course, publication is not even a sensible word. You can publish something by
+
+48:47.600 --> 48:54.400
+ putting it in an archive so I can publish everything tomorrow. So making stuff public is
+
+48:54.400 --> 49:04.800
+ there's no barrier. We still need curation and evaluation. I don't have time to read all of
+
+49:04.800 --> 49:20.480
+ archive. And you could argue that kind of social thumbs uping of articles suffices, right? You
+
+49:20.480 --> 49:25.440
+ might say, oh, heck with this, we don't need journals at all. We'll put everything on archive
+
+49:25.440 --> 49:30.400
+ and people will upvote and downvote the articles and then your CV will say, oh, man, he got a lot
+
+49:30.400 --> 49:44.000
+ of upvotes. So that's good. But I think there's still value in careful reading and commentary of
+
+49:44.000 --> 49:48.480
+ things. And it's hard to tell when people are upvoting and downvoting or arguing about your
+
+49:48.480 --> 49:55.440
+ paper on Twitter and Reddit, whether they know what they're talking about. So then I have the
+
+49:55.440 --> 50:01.360
+ second order problem of trying to decide whose opinions I should value and such. So I don't
+
+50:01.360 --> 50:06.240
+ know. If I had infinite time, which I don't, and I'm not going to do this because I really want to
+
+50:06.240 --> 50:11.920
+ make robots work, but if I felt inclined to do something more in a publication direction,
+
+50:12.880 --> 50:16.160
+ I would do this other thing, which I thought about doing the first time, which is to get
+
+50:16.160 --> 50:22.480
+ together some set of people whose opinions I value and who are pretty articulate. And I guess we
+
+50:22.480 --> 50:27.520
+ would be public, although we could be private, I'm not sure. And we would review papers. We wouldn't
+
+50:27.520 --> 50:31.600
+ publish them and you wouldn't submit them. We would just find papers and we would write reviews
+
+50:32.720 --> 50:39.120
+ and we would make those reviews public. And maybe if you, you know, so we're Leslie's friends who
+
+50:39.120 --> 50:45.200
+ review papers and maybe eventually if we, our opinion was sufficiently valued, like the opinion
+
+50:45.200 --> 50:50.800
+ of JMLR is valued, then you'd say on your CV that Leslie's friends gave my paper a five star reading
+
+50:50.800 --> 50:58.800
+ and that would be just as good as saying I got it accepted into this journal. So I think we
+
+50:58.800 --> 51:04.800
+ should have good public commentary and organize it in some way, but I don't really know how to
+
+51:04.800 --> 51:09.120
+ do it. It's interesting times. The way you describe it actually is really interesting. I mean,
+
+51:09.120 --> 51:15.040
+ we do it for movies, IMDB.com. There's experts, critics come in, they write reviews, but there's
+
+51:15.040 --> 51:20.960
+ also regular non critics humans write reviews and they're separated. I like open review.
+
+51:22.240 --> 51:31.600
+ The eye clear process, I think is interesting. It's a step in the right direction, but it's still
+
+51:31.600 --> 51:39.840
+ not as compelling as reviewing movies or video games. I mean, it sometimes almost, it might be
+
+51:39.840 --> 51:44.400
+ silly, at least from my perspective to say, but it boils down to the user interface, how fun and
+
+51:44.400 --> 51:50.400
+ easy it is to actually perform the reviews, how efficient, how much you as a reviewer get
+
+51:51.200 --> 51:56.560
+ street cred for being a good reviewer. Those human elements come into play.
+
+51:57.200 --> 52:03.600
+ No, it's a big investment to do a good review of a paper and the flood of papers is out of control.
+
+52:05.280 --> 52:08.960
+ There aren't 3,000 new, I don't know how many new movies are there in a year, I don't know,
+
+52:08.960 --> 52:15.440
+ but there's probably going to be less than how many machine learning papers there are in a year now.
+
+52:19.840 --> 52:22.320
+ Right, so I'm like an old person, so of course I'm going to say,
+
+52:23.520 --> 52:30.240
+ things are moving too fast, I'm a stick in the mud. So I can say that, but my particular flavor
+
+52:30.240 --> 52:38.240
+ of that is, I think the horizon for researchers has gotten very short, that students want to
+
+52:38.240 --> 52:46.000
+ publish a lot of papers and it's exciting and there's value in that and you get padded on the
+
+52:46.000 --> 52:58.320
+ head for it and so on. And some of that is fine, but I'm worried that we're driving out people who
+
+52:58.320 --> 53:05.280
+ would spend two years thinking about something. Back in my day, when we worked on our theses,
+
+53:05.280 --> 53:10.560
+ we did not publish papers, you did your thesis for years, you picked a hard problem and then you
+
+53:10.560 --> 53:16.320
+ worked and chewed on it and did stuff and wasted time and for a long time. And when it was roughly,
+
+53:16.320 --> 53:22.800
+ when it was done, you would write papers. And so I don't know how to, and I don't think that
+
+53:22.800 --> 53:26.720
+ everybody has to work in that mode, but I think there's some problems that are hard enough
+
+53:27.680 --> 53:31.680
+ that it's important to have a longer research horizon and I'm worried that
+
+53:31.680 --> 53:39.600
+ we don't incentivize that at all at this point. In this current structure. So what do you see
+
+53:41.440 --> 53:47.280
+ what are your hopes and fears about the future of AI and continuing on this theme? So AI has
+
+53:47.280 --> 53:53.440
+ gone through a few winters, ups and downs. Do you see another winter of AI coming?
+
+53:53.440 --> 54:02.480
+ Or are you more hopeful about making robots work, as you said? I think the cycles are inevitable,
+
+54:03.040 --> 54:10.080
+ but I think each time we get higher, right? I mean, it's like climbing some kind of
+
+54:10.080 --> 54:19.600
+ landscape with a noisy optimizer. So it's clear that the deep learning stuff has
+
+54:19.600 --> 54:25.760
+ made deep and important improvements. And so the high watermark is now higher. There's no question.
+
+54:25.760 --> 54:34.400
+ But of course, I think people are overselling and eventually investors, I guess, and other people
+
+54:34.400 --> 54:40.640
+ look around and say, well, you're not quite delivering on this grand claim and that wild
+
+54:40.640 --> 54:47.680
+ hypothesis. It's like probably it's going to crash something out and then it's okay. I mean,
+
+54:47.680 --> 54:54.000
+ it's okay. I mean, but I don't I can't imagine that there's like some awesome monotonic improvement
+
+54:54.000 --> 55:01.760
+ from here to human level AI. So in, you know, I have to ask this question, I probably anticipate
+
+55:01.760 --> 55:09.120
+ answers, the answers. But do you have a worry short term, a long term about the existential
+
+55:09.120 --> 55:18.880
+ threats of AI and maybe short term, less existential, but more robots taking away jobs?
+
+55:20.480 --> 55:28.000
+ Well, actually, let me talk a little bit about utility. Actually, I had an interesting conversation
+
+55:28.000 --> 55:32.480
+ with some military ethicists who wanted to talk to me about autonomous weapons.
+
+55:32.480 --> 55:39.360
+ And they're, they were interesting, smart, well educated guys who didn't know too much about AI or
+
+55:39.360 --> 55:43.600
+ machine learning. And the first question they asked me was, has your robot ever done something you
+
+55:43.600 --> 55:49.120
+ didn't expect? And I like burst out laughing because anybody who's ever done something other robot
+
+55:49.120 --> 55:54.720
+ right knows that they don't do much. And what I realized was that their model of how we program
+
+55:54.720 --> 55:59.440
+ a robot was completely wrong. Their model of how we could put program robot was like,
+
+55:59.440 --> 56:05.600
+ program robot was like, Lego Mindstorms, like, Oh, go forward a meter, turn left, take a picture,
+
+56:05.600 --> 56:11.120
+ do this, do that. And so if you have that model of programming, then it's true, it's kind of weird
+
+56:11.120 --> 56:16.240
+ that your robot would do something that you didn't anticipate. But the fact is, and actually,
+
+56:16.240 --> 56:21.840
+ so now this is my new educational mission, if I have to talk to non experts, I try to teach them
+
+56:22.720 --> 56:28.080
+ the idea that we don't operate, we operate at least one or maybe many levels of abstraction
+
+56:28.080 --> 56:33.280
+ about that. And we say, Oh, here's a hypothesis class, maybe it's a space of plans, or maybe it's a
+
+56:33.280 --> 56:38.400
+ space of classifiers, or whatever. But there's some set of answers and an objective function. And
+
+56:38.400 --> 56:44.800
+ then we work on some optimization method that tries to optimize a solution in that class.
+
+56:46.080 --> 56:50.560
+ And we don't know what solution is going to come out. Right. So I think it's important to
+
+56:50.560 --> 56:55.520
+ communicate that. So I mean, of course, probably people who listen to this, they know that lesson.
+
+56:55.520 --> 56:59.600
+ But I think it's really critical to communicate that lesson. And then lots of people are now
+
+56:59.600 --> 57:05.600
+ talking about, you know, the value alignment problem. So you want to be sure, as robots or
+
+57:06.480 --> 57:11.280
+ software systems get more competent, that their objectives are aligned with your objectives,
+
+57:11.280 --> 57:17.680
+ or that our objectives are compatible in some way, or we have a good way of mediating when they have
+
+57:17.680 --> 57:22.240
+ different objectives. And so I think it is important to start thinking in terms, like,
+
+57:22.240 --> 57:28.480
+ you don't have to be freaked out by the robot apocalypse to accept that it's important to think
+
+57:28.480 --> 57:33.760
+ about objective functions of value alignment. And that you have to really, everyone who's done
+
+57:33.760 --> 57:38.160
+ optimization knows that you have to be careful what you wish for that, you know, sometimes you get
+
+57:38.160 --> 57:45.280
+ the optimal solution. And you realize, man, that was that objective was wrong. So pragmatically,
+
+57:45.280 --> 57:51.360
+ in the shortest term, it seems to me that that those are really interesting and critical questions.
+
+57:51.360 --> 57:55.680
+ And the idea that we're going to go from being people who engineer algorithms to being people
+
+57:55.680 --> 58:00.800
+ who engineer objective functions, I think that's, that's definitely going to happen. And that's
+
+58:00.800 --> 58:03.360
+ going to change our thinking and methodology and stuff.
+
+58:03.360 --> 58:07.520
+ We're going to, you started at Stanford philosophy, that's wish you could be science,
+
+58:07.520 --> 58:13.840
+ and I will go back to philosophy maybe. Well, I mean, they're mixed together because, because,
+
+58:13.840 --> 58:18.240
+ as we also know, as machine learning people, right? When you design, in fact, this is the
+
+58:18.240 --> 58:23.360
+ lecture I gave in class today, when you design an objective function, you have to wear both hats.
+
+58:23.360 --> 58:28.320
+ There's the hat that says, what do I want? And there's the hat that says, but I know what my
+
+58:28.320 --> 58:34.240
+ optimizer can do to some degree. And I have to take that into account. So it's, it's always a
+
+58:34.240 --> 58:40.480
+ trade off. And we have to kind of be mindful of that. The part about taking people's jobs,
+
+58:40.480 --> 58:47.360
+ that I understand that that's important, I don't understand sociology or economics or people
+
+58:47.360 --> 58:51.840
+ very well. So I don't know how to think about that. So that's, yeah, so there might be a
+
+58:51.840 --> 58:56.640
+ sociological aspect there, the economic aspect that's very difficult to think about. Okay.
+
+58:56.640 --> 59:00.000
+ I mean, I think other people should be thinking about it, but I'm just, that's not my strength.
+
+59:00.000 --> 59:04.320
+ So what do you think is the most exciting area of research in the short term,
+
+59:04.320 --> 59:08.560
+ for the community and for your, for yourself? Well, so, I mean, there's this story I've been
+
+59:08.560 --> 59:16.480
+ telling about how to engineer intelligent robots. So that's what we want to do. We all kind of want
+
+59:16.480 --> 59:20.960
+ to do, well, I mean, some set of us want to do this. And the question is, what's the most effective
+
+59:20.960 --> 59:25.840
+ strategy? And we've tried, and there's a bunch of different things you could do at the extremes,
+
+59:25.840 --> 59:32.000
+ right? One super extreme is we do introspection and we write a program. Okay, that has not worked
+
+59:32.000 --> 59:37.360
+ out very well. Another extreme is we take a giant bunch of neural guru and we try and train it up to
+
+59:37.360 --> 59:43.040
+ do something. I don't think that's going to work either. So the question is, what's the middle
+
+59:43.040 --> 59:49.840
+ ground? And again, this isn't a theological question or anything like that. It's just,
+
+59:49.840 --> 59:57.040
+ like, how do, just how do we, what's the best way to make this work out? And I think it's clear,
+
+59:57.040 --> 1:00:01.840
+ it's a combination of learning, to me, it's clear, it's a combination of learning and not learning.
+
+1:00:02.400 --> 1:00:05.920
+ And what should that combination be? And what's the stuff we build in? So to me,
+
+1:00:05.920 --> 1:00:10.080
+ that's the most compelling question. And when you say engineer robots, you mean
+
+1:00:10.080 --> 1:00:15.600
+ engineering systems that work in the real world. That's the emphasis.
+
+1:00:17.600 --> 1:00:23.200
+ Last question, which robots or robot is your favorite from science fiction?
+
+1:00:24.480 --> 1:00:32.960
+ So you can go with Star Wars or RTD2, or you can go with more modern, maybe Hal.
+
+1:00:32.960 --> 1:00:37.040
+ No, sir, I don't think I have a favorite robot from science fiction.
+
+1:00:37.040 --> 1:00:45.520
+ This is, this is back to, you like to make robots work in the real world here, not, not in.
+
+1:00:45.520 --> 1:00:50.000
+ I mean, I love the process. And I care more about the process.
+
+1:00:50.000 --> 1:00:51.040
+ The engineering process.
+
+1:00:51.600 --> 1:00:55.760
+ Yeah. I mean, I do research because it's fun, not because I care about what we produce.
+
+1:00:57.520 --> 1:01:01.920
+ Well, that's, that's a beautiful note, actually. And Leslie, thank you so much for talking today.
+
+1:01:01.920 --> 1:01:07.920
+ Sure, it's been fun.
+

vtt/episode_018_small.vtt

@@ -0,0 +1,1871 @@
+WEBVTT
+
+00:00.000 --> 00:03.000
+ The following is a conversation with Elon Musk.
+
+00:03.000 --> 00:06.240
+ He's the CEO of Tesla, SpaceX, Neuralink,
+
+00:06.240 --> 00:09.200
+ and a cofounder of several other companies.
+
+00:09.200 --> 00:10.740
+ This conversation is part
+
+00:10.740 --> 00:13.200
+ of the artificial intelligence podcast.
+
+00:13.200 --> 00:15.640
+ The series includes leading researchers
+
+00:15.640 --> 00:19.320
+ in academia and industry, including CEOs and CTOs
+
+00:19.320 --> 00:24.080
+ of automotive, robotics, AI, and technology companies.
+
+00:24.080 --> 00:26.880
+ This conversation happened after the release of the paper
+
+00:26.880 --> 00:30.520
+ from our group at MIT on driver functional vigilance
+
+00:30.520 --> 00:32.880
+ during use of Tesla's autopilot.
+
+00:32.880 --> 00:34.560
+ The Tesla team reached out to me,
+
+00:34.560 --> 00:37.480
+ offering a podcast conversation with Mr. Musk.
+
+00:37.480 --> 00:40.640
+ I accepted with full control of questions I could ask
+
+00:40.640 --> 00:43.560
+ and the choice of what is released publicly.
+
+00:43.560 --> 00:46.840
+ I ended up editing out nothing of substance.
+
+00:46.840 --> 00:49.680
+ I've never spoken with Elon before this conversation,
+
+00:49.680 --> 00:51.720
+ publicly or privately.
+
+00:51.720 --> 00:54.360
+ Neither he nor his companies have any influence
+
+00:54.360 --> 00:57.840
+ on my opinion, nor on the rigor and integrity
+
+00:57.840 --> 00:59.760
+ of the scientific method that I practice
+
+00:59.760 --> 01:01.840
+ in my position at MIT.
+
+01:01.840 --> 01:04.640
+ Tesla has never financially supported my research
+
+01:04.640 --> 01:07.320
+ and I've never owned a Tesla vehicle.
+
+01:07.320 --> 01:10.160
+ I've never owned Tesla stock.
+
+01:10.160 --> 01:12.800
+ This podcast is not a scientific paper.
+
+01:12.800 --> 01:14.360
+ It is a conversation.
+
+01:14.360 --> 01:16.720
+ I respect Elon as I do all other leaders
+
+01:16.720 --> 01:18.680
+ and engineers I've spoken with.
+
+01:18.680 --> 01:21.440
+ We agree on some things and disagree on others.
+
+01:21.440 --> 01:23.480
+ My goal is always with these conversations
+
+01:23.480 --> 01:26.920
+ is to understand the way the guest sees the world.
+
+01:26.920 --> 01:28.600
+ One particular point of this agreement
+
+01:28.600 --> 01:30.640
+ in this conversation was the extent
+
+01:30.640 --> 01:33.240
+ to which camera based driver monitoring
+
+01:33.240 --> 01:36.120
+ will improve outcomes and for how long
+
+01:36.120 --> 01:39.120
+ it will remain relevant for AI assisted driving.
+
+01:39.960 --> 01:42.240
+ As someone who works on and is fascinated
+
+01:42.240 --> 01:45.200
+ by human centered artificial intelligence,
+
+01:45.200 --> 01:48.720
+ I believe that if implemented and integrated effectively,
+
+01:48.720 --> 01:51.840
+ camera based driver monitoring is likely to be of benefit
+
+01:51.840 --> 01:55.640
+ in both the short term and the long term.
+
+01:55.640 --> 01:59.240
+ In contrast, Elon and Tesla's focus
+
+01:59.240 --> 02:01.200
+ is on the improvement of autopilot
+
+02:01.200 --> 02:04.480
+ such that its statistical safety benefits
+
+02:04.480 --> 02:09.040
+ override any concern of human behavior and psychology.
+
+02:09.040 --> 02:12.040
+ Elon and I may not agree on everything
+
+02:12.040 --> 02:13.920
+ but I deeply respect the engineering
+
+02:13.920 --> 02:16.880
+ and innovation behind the efforts that he leads.
+
+02:16.880 --> 02:20.560
+ My goal here is to catalyze a rigorous, nuanced
+
+02:20.560 --> 02:23.520
+ and objective discussion in industry and academia
+
+02:23.520 --> 02:26.240
+ on AI assisted driving,
+
+02:26.240 --> 02:30.840
+ one that ultimately makes for a safer and better world.
+
+02:30.840 --> 02:34.600
+ And now here's my conversation with Elon Musk.
+
+02:35.600 --> 02:38.640
+ What was the vision, the dream of autopilot
+
+02:38.640 --> 02:41.400
+ when in the beginning the big picture system level
+
+02:41.400 --> 02:43.680
+ when it was first conceived
+
+02:43.680 --> 02:45.960
+ and started being installed in 2014
+
+02:45.960 --> 02:47.520
+ in the hardware and the cars?
+
+02:47.520 --> 02:49.760
+ What was the vision, the dream?
+
+02:49.760 --> 02:51.400
+ I would characterize the vision or dream
+
+02:51.400 --> 02:54.400
+ simply that there are obviously two
+
+02:54.400 --> 02:59.400
+ massive revolutions in the automobile industry.
+
+03:00.120 --> 03:04.440
+ One is the transition to electrification
+
+03:04.440 --> 03:06.400
+ and then the other is autonomy.
+
+03:07.720 --> 03:12.720
+ And it became obvious to me that in the future
+
+03:13.240 --> 03:16.240
+ any car that does not have autonomy
+
+03:16.240 --> 03:19.160
+ I would be about as useful as a horse.
+
+03:19.160 --> 03:22.040
+ Which is not to say that there's no use, it's just rare
+
+03:22.040 --> 03:23.640
+ and somewhat idiosyncratic
+
+03:23.640 --> 03:25.480
+ if somebody has a horse at this point.
+
+03:25.480 --> 03:28.000
+ It's just obvious that cars will drive themselves completely.
+
+03:28.000 --> 03:29.600
+ It's just a question of time
+
+03:29.600 --> 03:34.600
+ and if we did not participate in the autonomy revolution
+
+03:36.920 --> 03:40.840
+ then our cars would not be useful to people
+
+03:40.840 --> 03:43.680
+ relative to cars that are autonomous.
+
+03:43.680 --> 03:47.160
+ I mean an autonomous car is arguably worth
+
+03:47.160 --> 03:52.160
+ five to 10 times more than a car that which is not autonomous.
+
+03:53.760 --> 03:55.160
+ In the long term.
+
+03:55.160 --> 03:56.200
+ Turns out what you mean by long term,
+
+03:56.200 --> 03:59.520
+ but let's say at least for the next five years
+
+03:59.520 --> 04:00.520
+ perhaps 10 years.
+
+04:01.440 --> 04:04.080
+ So there are a lot of very interesting design choices
+
+04:04.080 --> 04:05.720
+ with autopilot early on.
+
+04:05.720 --> 04:09.960
+ First is showing on the instrument cluster
+
+04:09.960 --> 04:12.680
+ or in the Model 3 on the center stack display
+
+04:12.680 --> 04:15.720
+ what the combined sensor suite sees.
+
+04:15.720 --> 04:17.920
+ What was the thinking behind that choice?
+
+04:17.920 --> 04:18.960
+ Was there a debate?
+
+04:18.960 --> 04:20.480
+ What was the process?
+
+04:20.480 --> 04:24.840
+ The whole point of the display is to provide a health check
+
+04:24.840 --> 04:28.080
+ on the vehicle's perception of reality.
+
+04:28.080 --> 04:31.320
+ So the vehicle's taking information for a bunch of sensors
+
+04:31.320 --> 04:34.680
+ primarily cameras, but also radar and ultrasonics,
+
+04:34.680 --> 04:35.960
+ GPS and so forth.
+
+04:37.200 --> 04:42.200
+ And then that information is then rendered into vector space
+
+04:42.200 --> 04:46.360
+ and that with a bunch of objects with properties
+
+04:46.360 --> 04:49.920
+ like lane lines and traffic lights and other cars.
+
+04:49.920 --> 04:54.920
+ And then in vector space that is re rendered onto a display
+
+04:54.920 --> 04:57.400
+ so you can confirm whether the car knows
+
+04:57.400 --> 05:00.600
+ what's going on or not by looking out the window.
+
+05:01.600 --> 05:04.240
+ Right, I think that's an extremely powerful thing
+
+05:04.240 --> 05:06.480
+ for people to get an understanding
+
+05:06.480 --> 05:07.840
+ to become one with the system
+
+05:07.840 --> 05:10.400
+ and understanding what the system is capable of.
+
+05:10.400 --> 05:13.600
+ Now, have you considered showing more?
+
+05:13.600 --> 05:15.400
+ So if we look at the computer vision,
+
+05:16.400 --> 05:18.400
+ you know, like road segmentation, lane detection,
+
+05:18.400 --> 05:21.640
+ vehicle detection, object detection, underlying the system,
+
+05:21.640 --> 05:24.400
+ there is at the edges some uncertainty.
+
+05:24.400 --> 05:28.400
+ Have you considered revealing the parts
+
+05:28.400 --> 05:32.400
+ that the uncertainty in the system, the sort of problem
+
+05:32.400 --> 05:35.000
+ these associated with say image recognition
+
+05:35.000 --> 05:35.840
+ or something like that?
+
+05:35.840 --> 05:37.840
+ Yeah, so right now it shows like the vehicles
+
+05:37.840 --> 05:40.840
+ and the vicinity of very clean crisp image
+
+05:40.840 --> 05:43.840
+ and people do confirm that there's a car in front of me
+
+05:43.840 --> 05:45.840
+ and the system sees there's a car in front of me
+
+05:45.840 --> 05:47.840
+ but to help people build an intuition
+
+05:47.840 --> 05:51.840
+ of what computer vision is by showing some of the uncertainty.
+
+05:51.840 --> 05:53.840
+ Well, I think it's, in my car,
+
+05:53.840 --> 05:56.840
+ I always look at the sort of the debug view
+
+05:56.840 --> 05:58.840
+ and there's two debug views.
+
+05:58.840 --> 06:03.840
+ One is augmented vision, which I'm sure you've seen
+
+06:03.840 --> 06:07.840
+ where it's basically, we draw boxes and labels
+
+06:07.840 --> 06:10.840
+ around objects that are recognized.
+
+06:10.840 --> 06:14.840
+ And then there's what we call the visualizer,
+
+06:14.840 --> 06:16.840
+ which is basically a vector space representation
+
+06:16.840 --> 06:21.840
+ summing up the input from all sensors.
+
+06:21.840 --> 06:23.840
+ That does not show any pictures,
+
+06:23.840 --> 06:26.840
+ but it shows all of the,
+
+06:26.840 --> 06:32.840
+ it basically shows the cause view of the world in vector space.
+
+06:32.840 --> 06:36.840
+ But I think this is very difficult for normal people to understand.
+
+06:36.840 --> 06:38.840
+ They would not know what they're looking at.
+
+06:38.840 --> 06:40.840
+ So it's almost an HMI challenge.
+
+06:40.840 --> 06:42.840
+ The current things that are being displayed
+
+06:42.840 --> 06:46.840
+ is optimized for the general public understanding
+
+06:46.840 --> 06:48.840
+ of what the system is capable of.
+
+06:48.840 --> 06:50.840
+ It's like if you have no idea how computer vision works
+
+06:50.840 --> 06:52.840
+ or anything, you can still look at the screen
+
+06:52.840 --> 06:54.840
+ and see if the car knows what's going on.
+
+06:54.840 --> 06:57.840
+ And then if you're a development engineer
+
+06:57.840 --> 07:01.840
+ or if you have the development build like I do,
+
+07:01.840 --> 07:05.840
+ then you can see all the debug information.
+
+07:05.840 --> 07:10.840
+ But those would just be total diverse to most people.
+
+07:10.840 --> 07:13.840
+ What's your view on how to best distribute effort?
+
+07:13.840 --> 07:16.840
+ So there's three, I would say, technical aspects of autopilot
+
+07:16.840 --> 07:18.840
+ that are really important.
+
+07:18.840 --> 07:19.840
+ So it's the underlying algorithms,
+
+07:19.840 --> 07:21.840
+ like the neural network architecture.
+
+07:21.840 --> 07:23.840
+ There's the data that's trained on
+
+07:23.840 --> 07:25.840
+ and then there's the hardware development.
+
+07:25.840 --> 07:26.840
+ There may be others.
+
+07:26.840 --> 07:31.840
+ But so look, algorithm, data, hardware.
+
+07:31.840 --> 07:34.840
+ You only have so much money, only have so much time.
+
+07:34.840 --> 07:36.840
+ What do you think is the most important thing
+
+07:36.840 --> 07:39.840
+ to allocate resources to?
+
+07:39.840 --> 07:41.840
+ Do you see it as pretty evenly distributed
+
+07:41.840 --> 07:43.840
+ between those three?
+
+07:43.840 --> 07:46.840
+ We automatically get fast amounts of data
+
+07:46.840 --> 07:48.840
+ because all of our cars have
+
+07:50.840 --> 07:54.840
+ eight external facing cameras and radar
+
+07:54.840 --> 07:59.840
+ and usually 12 ultrasonic sensors, GPS, obviously,
+
+07:59.840 --> 08:03.840
+ and IMU.
+
+08:03.840 --> 08:08.840
+ And so we basically have a fleet that has,
+
+08:08.840 --> 08:11.840
+ we've got about 400,000 cars on the road
+
+08:11.840 --> 08:13.840
+ that have that level of data.
+
+08:13.840 --> 08:15.840
+ I think you keep quite close track of it, actually.
+
+08:15.840 --> 08:16.840
+ Yes.
+
+08:16.840 --> 08:19.840
+ So we're approaching half a million cars
+
+08:19.840 --> 08:22.840
+ on the road that have the full sensor suite.
+
+08:22.840 --> 08:26.840
+ So this is, I'm not sure how many other cars
+
+08:26.840 --> 08:28.840
+ on the road have this sensor suite,
+
+08:28.840 --> 08:31.840
+ but I'd be surprised if it's more than 5,000,
+
+08:31.840 --> 08:35.840
+ which means that we have 99% of all the data.
+
+08:35.840 --> 08:37.840
+ So there's this huge inflow of data.
+
+08:37.840 --> 08:39.840
+ Absolutely, massive inflow of data.
+
+08:39.840 --> 08:43.840
+ And then it's taken about three years,
+
+08:43.840 --> 08:46.840
+ but now we've finally developed our full self driving computer,
+
+08:46.840 --> 08:51.840
+ which can process
+
+08:51.840 --> 08:54.840
+ an order of magnitude as much as the NVIDIA system
+
+08:54.840 --> 08:56.840
+ that we currently have in the cars.
+
+08:56.840 --> 08:58.840
+ And it's really just to use it,
+
+08:58.840 --> 09:01.840
+ you unplug the NVIDIA computer and plug the Tesla computer in.
+
+09:01.840 --> 09:03.840
+ And that's it.
+
+09:03.840 --> 09:06.840
+ And it's, in fact, we're not even,
+
+09:06.840 --> 09:09.840
+ we're still exploring the boundaries of its capabilities,
+
+09:09.840 --> 09:11.840
+ but we're able to run the cameras at full frame rate,
+
+09:11.840 --> 09:14.840
+ full resolution, not even crop of the images,
+
+09:14.840 --> 09:19.840
+ and it's still got headroom, even on one of the systems.
+
+09:19.840 --> 09:22.840
+ The full self driving computer is really two computers,
+
+09:22.840 --> 09:25.840
+ two systems on a chip that are fully redundant.
+
+09:25.840 --> 09:28.840
+ So you could put a bolt through basically any part of that system
+
+09:28.840 --> 09:29.840
+ and it still works.
+
+09:29.840 --> 09:32.840
+ The redundancy, are they perfect copies of each other?
+
+09:32.840 --> 09:35.840
+ Or also it's purely for redundancy
+
+09:35.840 --> 09:37.840
+ as opposed to an arguing machine kind of architecture
+
+09:37.840 --> 09:39.840
+ where they're both making decisions.
+
+09:39.840 --> 09:41.840
+ This is purely for redundancy.
+
+09:41.840 --> 09:44.840
+ I think it's more like, if you have a twin engine aircraft,
+
+09:44.840 --> 09:46.840
+ commercial aircraft,
+
+09:46.840 --> 09:51.840
+ this system will operate best if both systems are operating,
+
+09:51.840 --> 09:55.840
+ but it's capable of operating safely on one.
+
+09:55.840 --> 09:59.840
+ So, but as it is right now, we can just run,
+
+09:59.840 --> 10:03.840
+ we haven't even hit the edge of performance,
+
+10:03.840 --> 10:08.840
+ so there's no need to actually distribute
+
+10:08.840 --> 10:12.840
+ functionality across both SoCs.
+
+10:12.840 --> 10:16.840
+ We can actually just run a full duplicate on each one.
+
+10:16.840 --> 10:20.840
+ You haven't really explored or hit the limit of the system?
+
+10:20.840 --> 10:21.840
+ Not yet, hit the limit now.
+
+10:21.840 --> 10:26.840
+ So the magic of deep learning is that it gets better with data.
+
+10:26.840 --> 10:28.840
+ You said there's a huge inflow of data,
+
+10:28.840 --> 10:33.840
+ but the thing about driving the really valuable data
+
+10:33.840 --> 10:35.840
+ to learn from is the edge cases.
+
+10:35.840 --> 10:42.840
+ So how do you, I mean, I've heard you talk somewhere about
+
+10:42.840 --> 10:46.840
+ autopilot disengagement as being an important moment of time to use.
+
+10:46.840 --> 10:51.840
+ Is there other edge cases or perhaps can you speak to those edge cases,
+
+10:51.840 --> 10:53.840
+ what aspects of them might be valuable,
+
+10:53.840 --> 10:55.840
+ or if you have other ideas,
+
+10:55.840 --> 10:59.840
+ how to discover more and more and more edge cases in driving?
+
+10:59.840 --> 11:01.840
+ Well, there's a lot of things that I learned.
+
+11:01.840 --> 11:05.840
+ There are certainly edge cases where I say somebody's on autopilot
+
+11:05.840 --> 11:07.840
+ and they take over.
+
+11:07.840 --> 11:12.840
+ And then, okay, that's a trigger that goes to a system that says,
+
+11:12.840 --> 11:14.840
+ okay, do they take over for convenience
+
+11:14.840 --> 11:18.840
+ or do they take over because the autopilot wasn't working properly?
+
+11:18.840 --> 11:21.840
+ There's also, like let's say we're trying to figure out
+
+11:21.840 --> 11:26.840
+ what is the optimal spline for traversing an intersection.
+
+11:26.840 --> 11:30.840
+ Then the ones where there are no interventions
+
+11:30.840 --> 11:32.840
+ and are the right ones.
+
+11:32.840 --> 11:36.840
+ So you then say, okay, when it looks like this, do the following.
+
+11:36.840 --> 11:40.840
+ And then you get the optimal spline for a complex,
+
+11:40.840 --> 11:44.840
+ now getting a complex intersection.
+
+11:44.840 --> 11:48.840
+ So that's for, there's kind of the common case.
+
+11:48.840 --> 11:51.840
+ You're trying to capture a huge amount of samples
+
+11:51.840 --> 11:54.840
+ of a particular intersection, how one thing went right.
+
+11:54.840 --> 11:58.840
+ And then there's the edge case where, as you said,
+
+11:58.840 --> 12:01.840
+ not for convenience, but something didn't go exactly right.
+
+12:01.840 --> 12:04.840
+ Somebody took over, somebody asserted manual control from autopilot.
+
+12:04.840 --> 12:08.840
+ And really, like the way to look at this is view all input is error.
+
+12:08.840 --> 12:11.840
+ If the user had to do input, it does something.
+
+12:11.840 --> 12:13.840
+ All input is error.
+
+12:13.840 --> 12:15.840
+ That's a powerful line to think of it that way,
+
+12:15.840 --> 12:17.840
+ because it may very well be error.
+
+12:17.840 --> 12:19.840
+ But if you want to exit the highway,
+
+12:19.840 --> 12:22.840
+ or if you want to, it's a navigation decision
+
+12:22.840 --> 12:24.840
+ that all autopilot is not currently designed to do,
+
+12:24.840 --> 12:26.840
+ then the driver takes over.
+
+12:26.840 --> 12:28.840
+ How do you know the difference?
+
+12:28.840 --> 12:30.840
+ Yeah, that's going to change with navigate and autopilot,
+
+12:30.840 --> 12:33.840
+ which we've just released, and without stall confirm.
+
+12:33.840 --> 12:36.840
+ So the navigation, like lane change based,
+
+12:36.840 --> 12:39.840
+ like asserting control in order to do a lane change,
+
+12:39.840 --> 12:43.840
+ or exit a freeway, or doing highway interchange,
+
+12:43.840 --> 12:47.840
+ the vast majority of that will go away with the release
+
+12:47.840 --> 12:49.840
+ that just went out.
+
+12:49.840 --> 12:52.840
+ Yeah, I don't think people quite understand
+
+12:52.840 --> 12:54.840
+ how big of a step that is.
+
+12:54.840 --> 12:55.840
+ Yeah, they don't.
+
+12:55.840 --> 12:57.840
+ If you drive the car, then you do.
+
+12:57.840 --> 12:59.840
+ So you still have to keep your hands on the steering wheel
+
+12:59.840 --> 13:02.840
+ currently when it does the automatic lane change?
+
+13:02.840 --> 13:04.840
+ What are...
+
+13:04.840 --> 13:07.840
+ So there's these big leaps through the development of autopilot
+
+13:07.840 --> 13:09.840
+ through its history,
+
+13:09.840 --> 13:12.840
+ and what stands out to you as the big leaps?
+
+13:12.840 --> 13:14.840
+ I would say this one,
+
+13:14.840 --> 13:19.840
+ navigate and autopilot without having to confirm,
+
+13:19.840 --> 13:20.840
+ is a huge leap.
+
+13:20.840 --> 13:21.840
+ It is a huge leap.
+
+13:21.840 --> 13:24.840
+ It also automatically overtakes slow cars.
+
+13:24.840 --> 13:30.840
+ So it's both navigation and seeking the fastest lane.
+
+13:30.840 --> 13:36.840
+ So it'll overtake a slow cause and exit the freeway
+
+13:36.840 --> 13:39.840
+ and take highway interchanges.
+
+13:39.840 --> 13:46.840
+ And then we have traffic light recognition,
+
+13:46.840 --> 13:49.840
+ which is introduced initially as a warning.
+
+13:49.840 --> 13:51.840
+ I mean, on the development version that I'm driving,
+
+13:51.840 --> 13:55.840
+ the car fully stops and goes at traffic lights.
+
+13:55.840 --> 13:57.840
+ So those are the steps, right?
+
+13:57.840 --> 13:59.840
+ You just mentioned something sort of
+
+13:59.840 --> 14:02.840
+ including a step towards full autonomy.
+
+14:02.840 --> 14:07.840
+ What would you say are the biggest technological roadblocks
+
+14:07.840 --> 14:09.840
+ to full cell driving?
+
+14:09.840 --> 14:10.840
+ Actually, I don't think...
+
+14:10.840 --> 14:11.840
+ I think we just...
+
+14:11.840 --> 14:13.840
+ the full cell driving computer that we just...
+
+14:13.840 --> 14:14.840
+ that has a...
+
+14:14.840 --> 14:16.840
+ what we call the FSD computer.
+
+14:16.840 --> 14:20.840
+ That's now in production.
+
+14:20.840 --> 14:25.840
+ So if you order any Model SRX or any Model 3
+
+14:25.840 --> 14:28.840
+ that has the full cell driving package,
+
+14:28.840 --> 14:31.840
+ you'll get the FSD computer.
+
+14:31.840 --> 14:36.840
+ That's important to have enough base computation.
+
+14:36.840 --> 14:40.840
+ Then refining the neural net and the control software.
+
+14:40.840 --> 14:44.840
+ But all of that can just be provided as an over there update.
+
+14:44.840 --> 14:46.840
+ The thing that's really profound,
+
+14:46.840 --> 14:50.840
+ and where I'll be emphasizing at the...
+
+14:50.840 --> 14:52.840
+ that investor day that we're having focused on autonomy,
+
+14:52.840 --> 14:55.840
+ is that the cars currently being produced,
+
+14:55.840 --> 14:57.840
+ or the hardware currently being produced,
+
+14:57.840 --> 15:00.840
+ is capable of full cell driving.
+
+15:00.840 --> 15:03.840
+ But capable is an interesting word because...
+
+15:03.840 --> 15:05.840
+ Like the hardware is.
+
+15:05.840 --> 15:08.840
+ And as we refine the software,
+
+15:08.840 --> 15:11.840
+ the capabilities will increase dramatically
+
+15:11.840 --> 15:13.840
+ and then the reliability will increase dramatically
+
+15:13.840 --> 15:15.840
+ and then it will receive regulatory approval.
+
+15:15.840 --> 15:18.840
+ So essentially buying a car today is an investment in the future.
+
+15:18.840 --> 15:21.840
+ You're essentially buying...
+
+15:21.840 --> 15:25.840
+ I think the most profound thing is that
+
+15:25.840 --> 15:27.840
+ if you buy a Tesla today,
+
+15:27.840 --> 15:29.840
+ I believe you are buying an appreciating asset,
+
+15:29.840 --> 15:32.840
+ not a depreciating asset.
+
+15:32.840 --> 15:34.840
+ So that's a really important statement there
+
+15:34.840 --> 15:36.840
+ because if hardware is capable enough,
+
+15:36.840 --> 15:39.840
+ that's the hard thing to upgrade usually.
+
+15:39.840 --> 15:40.840
+ Exactly.
+
+15:40.840 --> 15:43.840
+ So then the rest is a software problem.
+
+15:43.840 --> 15:47.840
+ Yes. Software has no marginal cost, really.
+
+15:47.840 --> 15:51.840
+ But what's your intuition on the software side?
+
+15:51.840 --> 15:55.840
+ How hard are the remaining steps
+
+15:55.840 --> 15:58.840
+ to get it to where...
+
+15:58.840 --> 16:02.840
+ you know, the experience,
+
+16:02.840 --> 16:05.840
+ not just the safety, but the full experience
+
+16:05.840 --> 16:08.840
+ is something that people would enjoy.
+
+16:08.840 --> 16:12.840
+ I think people would enjoy it very much on the highways.
+
+16:12.840 --> 16:16.840
+ It's a total game changer for quality of life,
+
+16:16.840 --> 16:20.840
+ for using Tesla autopilot on the highways.
+
+16:20.840 --> 16:24.840
+ So it's really just extending that functionality to city streets,
+
+16:24.840 --> 16:28.840
+ adding in the traffic light recognition,
+
+16:28.840 --> 16:31.840
+ navigating complex intersections,
+
+16:31.840 --> 16:36.840
+ and then being able to navigate complicated parking lots
+
+16:36.840 --> 16:39.840
+ so the car can exit a parking space
+
+16:39.840 --> 16:45.840
+ and come and find you even if it's in a complete maze of a parking lot.
+
+16:45.840 --> 16:51.840
+ And then you can just drop you off and find a parking spot by itself.
+
+16:51.840 --> 16:53.840
+ Yeah, in terms of enjoyability
+
+16:53.840 --> 16:57.840
+ and something that people would actually find a lot of use from,
+
+16:57.840 --> 17:00.840
+ the parking lot is a really...
+
+17:00.840 --> 17:03.840
+ it's rich of annoyance when you have to do it manually,
+
+17:03.840 --> 17:07.840
+ so there's a lot of benefit to be gained from automation there.
+
+17:07.840 --> 17:11.840
+ So let me start injecting the human into this discussion a little bit.
+
+17:11.840 --> 17:14.840
+ So let's talk about full autonomy.
+
+17:14.840 --> 17:17.840
+ If you look at the current level four vehicles,
+
+17:17.840 --> 17:19.840
+ being Tesla and road like Waymo and so on,
+
+17:19.840 --> 17:22.840
+ they're only technically autonomous.
+
+17:22.840 --> 17:25.840
+ They're really level two systems
+
+17:25.840 --> 17:28.840
+ with just a different design philosophy
+
+17:28.840 --> 17:31.840
+ because there's always a safety driver in almost all cases
+
+17:31.840 --> 17:33.840
+ and they're monitoring the system.
+
+17:33.840 --> 17:37.840
+ Maybe Tesla's full self driving
+
+17:37.840 --> 17:41.840
+ is still for a time to come,
+
+17:41.840 --> 17:44.840
+ requiring supervision of the human being.
+
+17:44.840 --> 17:47.840
+ So its capabilities are powerful enough to drive,
+
+17:47.840 --> 17:50.840
+ but nevertheless requires the human to still be supervising
+
+17:50.840 --> 17:56.840
+ just like a safety driver is in a other fully autonomous vehicles.
+
+17:56.840 --> 18:01.840
+ I think it will require detecting hands on wheel
+
+18:01.840 --> 18:08.840
+ or at least six months or something like that from here.
+
+18:08.840 --> 18:11.840
+ Really it's a question of like,
+
+18:11.840 --> 18:15.840
+ from a regulatory standpoint,
+
+18:15.840 --> 18:19.840
+ how much safer than a person does autopilot need to be
+
+18:19.840 --> 18:24.840
+ for it to be okay to not monitor the car?
+
+18:24.840 --> 18:27.840
+ And this is a debate that one can have.
+
+18:27.840 --> 18:31.840
+ But you need a large amount of data
+
+18:31.840 --> 18:34.840
+ so you can prove with high confidence,
+
+18:34.840 --> 18:36.840
+ statistically speaking,
+
+18:36.840 --> 18:39.840
+ that the car is dramatically safer than a person
+
+18:39.840 --> 18:42.840
+ and that adding in the person monitoring
+
+18:42.840 --> 18:45.840
+ does not materially affect the safety.
+
+18:45.840 --> 18:49.840
+ So it might need to be like two or three hundred percent safer than a person.
+
+18:49.840 --> 18:51.840
+ And how do you prove that?
+
+18:51.840 --> 18:53.840
+ Incidence per mile.
+
+18:53.840 --> 18:56.840
+ So crashes and fatalities.
+
+18:56.840 --> 18:58.840
+ Yeah, fatalities would be a factor,
+
+18:58.840 --> 19:00.840
+ but there are just not enough fatalities
+
+19:00.840 --> 19:03.840
+ to be statistically significant at scale.
+
+19:03.840 --> 19:06.840
+ But there are enough crashes,
+
+19:06.840 --> 19:10.840
+ there are far more crashes than there are fatalities.
+
+19:10.840 --> 19:15.840
+ So you can assess what is the probability of a crash,
+
+19:15.840 --> 19:19.840
+ then there's another step which probability of injury
+
+19:19.840 --> 19:21.840
+ and probability of permanent injury
+
+19:21.840 --> 19:23.840
+ and probability of death.
+
+19:23.840 --> 19:27.840
+ And all of those need to be much better than a person
+
+19:27.840 --> 19:32.840
+ by at least perhaps two hundred percent.
+
+19:32.840 --> 19:36.840
+ And you think there's the ability to have a healthy discourse
+
+19:36.840 --> 19:39.840
+ with the regulatory bodies on this topic?
+
+19:39.840 --> 19:43.840
+ I mean, there's no question that regulators pay
+
+19:43.840 --> 19:48.840
+ disproportionate amount of attention to that which generates press.
+
+19:48.840 --> 19:50.840
+ This is just an objective fact.
+
+19:50.840 --> 19:52.840
+ And Tesla generates a lot of press.
+
+19:52.840 --> 19:56.840
+ So that, you know, in the United States,
+
+19:56.840 --> 20:00.840
+ there's I think almost 40,000 automotive deaths per year.
+
+20:00.840 --> 20:03.840
+ But if there are four in Tesla,
+
+20:03.840 --> 20:06.840
+ they'll probably receive a thousand times more press
+
+20:06.840 --> 20:08.840
+ than anyone else.
+
+20:08.840 --> 20:10.840
+ So the psychology of that is actually fascinating.
+
+20:10.840 --> 20:12.840
+ I don't think we'll have enough time to talk about that,
+
+20:12.840 --> 20:16.840
+ but I have to talk to you about the human side of things.
+
+20:16.840 --> 20:20.840
+ So myself and our team at MIT recently released a paper
+
+20:20.840 --> 20:24.840
+ on functional vigilance of drivers while using autopilot.
+
+20:24.840 --> 20:27.840
+ This is work we've been doing since autopilot was first
+
+20:27.840 --> 20:30.840
+ released publicly over three years ago,
+
+20:30.840 --> 20:34.840
+ collecting video driver faces and driver body.
+
+20:34.840 --> 20:38.840
+ So I saw that you tweeted a quote from the abstract
+
+20:38.840 --> 20:43.840
+ so I can at least guess that you've glanced at it.
+
+20:43.840 --> 20:46.840
+ Can I talk you through what we found?
+
+20:46.840 --> 20:51.840
+ Okay, so it appears that in the data that we've collected
+
+20:51.840 --> 20:54.840
+ that drivers are maintaining functional vigilance
+
+20:54.840 --> 20:57.840
+ such that we're looking at 18,000 disengagement
+
+20:57.840 --> 21:02.840
+ from autopilot, 18,900 and annotating were they able
+
+21:02.840 --> 21:05.840
+ to take over control in a timely manner?
+
+21:05.840 --> 21:07.840
+ So they were there present looking at the road
+
+21:07.840 --> 21:09.840
+ to take over control.
+
+21:09.840 --> 21:14.840
+ Okay, so this goes against what many would predict
+
+21:14.840 --> 21:18.840
+ from the body of literature on vigilance with automation.
+
+21:18.840 --> 21:21.840
+ Now the question is, do you think these results
+
+21:21.840 --> 21:23.840
+ hold across the broader population?
+
+21:23.840 --> 21:26.840
+ So ours is just a small subset.
+
+21:26.840 --> 21:30.840
+ Do you think one of the criticism is that there's
+
+21:30.840 --> 21:34.840
+ a small minority of drivers that may be highly responsible
+
+21:34.840 --> 21:37.840
+ where their vigilance decrement would increase
+
+21:37.840 --> 21:39.840
+ with autopilot use?
+
+21:39.840 --> 21:41.840
+ I think this is all really going to be swept.
+
+21:41.840 --> 21:46.840
+ I mean, the system's improving so much so fast
+
+21:46.840 --> 21:50.840
+ that this is going to be a mood point very soon
+
+21:50.840 --> 21:56.840
+ where vigilance is, if something's many times safer
+
+21:56.840 --> 22:00.840
+ than a person, then adding a person does,
+
+22:00.840 --> 22:04.840
+ the effect on safety is limited.
+
+22:04.840 --> 22:09.840
+ And in fact, it could be negative.
+
+22:09.840 --> 22:11.840
+ That's really interesting.
+
+22:11.840 --> 22:16.840
+ So the fact that a human may, some percent of the population
+
+22:16.840 --> 22:20.840
+ may exhibit a vigilance decrement will not affect
+
+22:20.840 --> 22:22.840
+ overall statistics numbers of safety.
+
+22:22.840 --> 22:27.840
+ No, in fact, I think it will become very, very quickly,
+
+22:27.840 --> 22:29.840
+ maybe even towards the end of this year,
+
+22:29.840 --> 22:32.840
+ but I'd say I'd be shocked if it's not next year,
+
+22:32.840 --> 22:36.840
+ at the latest, that having a human intervene
+
+22:36.840 --> 22:39.840
+ will increase safety.
+
+22:39.840 --> 22:40.840
+ Decrease.
+
+22:40.840 --> 22:42.840
+ I can imagine if you're an elevator.
+
+22:42.840 --> 22:45.840
+ Now, it used to be that there were elevator operators
+
+22:45.840 --> 22:47.840
+ and you couldn't go on an elevator by yourself
+
+22:47.840 --> 22:51.840
+ and work the lever to move between floors.
+
+22:51.840 --> 22:56.840
+ And now, nobody wants an elevator operator
+
+22:56.840 --> 23:00.840
+ because the automated elevator that stops the floors
+
+23:00.840 --> 23:03.840
+ is much safer than the elevator operator.
+
+23:03.840 --> 23:05.840
+ And in fact, it would be quite dangerous
+
+23:05.840 --> 23:07.840
+ if someone with a lever that can move
+
+23:07.840 --> 23:09.840
+ the elevator between floors.
+
+23:09.840 --> 23:12.840
+ So that's a really powerful statement
+
+23:12.840 --> 23:14.840
+ and a really interesting one.
+
+23:14.840 --> 23:16.840
+ But I also have to ask, from a user experience
+
+23:16.840 --> 23:18.840
+ and from a safety perspective,
+
+23:18.840 --> 23:20.840
+ one of the passions for me algorithmically
+
+23:20.840 --> 23:25.840
+ is camera based detection of sensing the human,
+
+23:25.840 --> 23:27.840
+ but detecting what the driver is looking at,
+
+23:27.840 --> 23:29.840
+ cognitive load, body pose.
+
+23:29.840 --> 23:31.840
+ On the computer vision side, that's a fascinating problem,
+
+23:31.840 --> 23:34.840
+ but there's many in industry who believe
+
+23:34.840 --> 23:37.840
+ you have to have camera based driver monitoring.
+
+23:37.840 --> 23:39.840
+ Do you think this could be benefit gained
+
+23:39.840 --> 23:41.840
+ from driver monitoring?
+
+23:41.840 --> 23:45.840
+ If you have a system that's out or below
+
+23:45.840 --> 23:49.840
+ human level reliability, then driver monitoring makes sense.
+
+23:49.840 --> 23:51.840
+ But if your system is dramatically better,
+
+23:51.840 --> 23:53.840
+ more reliable than a human,
+
+23:53.840 --> 23:58.840
+ then driver monitoring is not help much.
+
+23:58.840 --> 24:03.840
+ And like I said, you wouldn't want someone into...
+
+24:03.840 --> 24:05.840
+ You wouldn't want someone in the elevator.
+
+24:05.840 --> 24:07.840
+ If you're in an elevator, do you really want someone
+
+24:07.840 --> 24:09.840
+ with a big lever, some random person operating
+
+24:09.840 --> 24:11.840
+ in the elevator between floors?
+
+24:11.840 --> 24:13.840
+ I wouldn't trust that.
+
+24:13.840 --> 24:16.840
+ I would rather have the buttons.
+
+24:16.840 --> 24:19.840
+ Okay, you're optimistic about the pace
+
+24:19.840 --> 24:21.840
+ of improvement of the system.
+
+24:21.840 --> 24:23.840
+ From what you've seen with the full self driving car,
+
+24:23.840 --> 24:25.840
+ computer.
+
+24:25.840 --> 24:27.840
+ The rate of improvement is exponential.
+
+24:27.840 --> 24:30.840
+ So one of the other very interesting design choices
+
+24:30.840 --> 24:34.840
+ early on that connects to this is the operational
+
+24:34.840 --> 24:37.840
+ design domain of autopilot.
+
+24:37.840 --> 24:41.840
+ So where autopilot is able to be turned on.
+
+24:41.840 --> 24:46.840
+ So contrast another vehicle system that we're studying
+
+24:46.840 --> 24:48.840
+ is the Cadillac SuperCrew system.
+
+24:48.840 --> 24:51.840
+ That's in terms of ODD, very constrained to this particular
+
+24:51.840 --> 24:54.840
+ kinds of highways, well mapped, tested,
+
+24:54.840 --> 24:58.840
+ but it's much narrower than the ODD of Tesla vehicles.
+
+24:58.840 --> 25:00.840
+ What's...
+
+25:00.840 --> 25:02.840
+ It's like ADD.
+
+25:02.840 --> 25:04.840
+ Yeah.
+
+25:04.840 --> 25:07.840
+ That's good. That's a good line.
+
+25:07.840 --> 25:10.840
+ What was the design decision
+
+25:10.840 --> 25:13.840
+ in that different philosophy of thinking where...
+
+25:13.840 --> 25:15.840
+ There's pros and cons.
+
+25:15.840 --> 25:20.840
+ What we see with a wide ODD is Tesla drivers are able
+
+25:20.840 --> 25:23.840
+ to explore more the limitations of the system,
+
+25:23.840 --> 25:26.840
+ at least early on, and they understand together
+
+25:26.840 --> 25:28.840
+ the instrument cluster display.
+
+25:28.840 --> 25:30.840
+ They start to understand what are the capabilities.
+
+25:30.840 --> 25:32.840
+ So that's a benefit.
+
+25:32.840 --> 25:37.840
+ The con is you're letting drivers use it basically anywhere.
+
+25:37.840 --> 25:41.840
+ Well, anyways, I could detect lanes with confidence.
+
+25:41.840 --> 25:46.840
+ Was there a philosophy design decisions that were challenging
+
+25:46.840 --> 25:48.840
+ that were being made there?
+
+25:48.840 --> 25:53.840
+ Or from the very beginning, was that done on purpose
+
+25:53.840 --> 25:55.840
+ with intent?
+
+25:55.840 --> 25:58.840
+ Frankly, it's pretty crazy letting people drive
+
+25:58.840 --> 26:02.840
+ a two ton death machine manually.
+
+26:02.840 --> 26:04.840
+ That's crazy.
+
+26:04.840 --> 26:06.840
+ In the future, people will be like,
+
+26:06.840 --> 26:09.840
+ I can't believe anyone was just allowed to drive
+
+26:09.840 --> 26:12.840
+ one of these two ton death machines
+
+26:12.840 --> 26:14.840
+ and they just drive wherever they wanted,
+
+26:14.840 --> 26:16.840
+ just like elevators.
+
+26:16.840 --> 26:18.840
+ You just move the elevator with the lever wherever you want.
+
+26:18.840 --> 26:21.840
+ It can stop at halfway between floors if you want.
+
+26:21.840 --> 26:24.840
+ It's pretty crazy.
+
+26:24.840 --> 26:29.840
+ So it's going to seem like a mad thing in the future
+
+26:29.840 --> 26:32.840
+ that people were driving cars.
+
+26:32.840 --> 26:35.840
+ So I have a bunch of questions about the human psychology,
+
+26:35.840 --> 26:37.840
+ about behavior and so on.
+
+26:37.840 --> 26:39.840
+ I don't know.
+
+26:39.840 --> 26:45.840
+ Because you have faith in the AI system,
+
+26:45.840 --> 26:50.840
+ not faith, but both on the hardware side
+
+26:50.840 --> 26:52.840
+ and the deep learning approach of learning from data
+
+26:52.840 --> 26:55.840
+ will make it just far safer than humans.
+
+26:55.840 --> 26:57.840
+ Yeah, exactly.
+
+26:57.840 --> 27:00.840
+ Recently, there are a few hackers who tricked autopilot
+
+27:00.840 --> 27:03.840
+ to act in unexpected ways with adversarial examples.
+
+27:03.840 --> 27:06.840
+ So we all know that neural network systems
+
+27:06.840 --> 27:08.840
+ are very sensitive to minor disturbances
+
+27:08.840 --> 27:10.840
+ to these adversarial examples on input.
+
+27:10.840 --> 27:13.840
+ Do you think it's possible to defend against something like this
+
+27:13.840 --> 27:15.840
+ for the industry?
+
+27:15.840 --> 27:17.840
+ Sure.
+
+27:17.840 --> 27:22.840
+ Can you elaborate on the confidence behind that answer?
+
+27:22.840 --> 27:27.840
+ Well, a neural net is just like a basic bunch of matrix math.
+
+27:27.840 --> 27:30.840
+ You have to be like a very sophisticated,
+
+27:30.840 --> 27:32.840
+ somebody who really understands neural nets
+
+27:32.840 --> 27:37.840
+ and basically reverse engineer how the matrix is being built
+
+27:37.840 --> 27:42.840
+ and then create a little thing that just exactly causes
+
+27:42.840 --> 27:44.840
+ the matrix math to be slightly off.
+
+27:44.840 --> 27:48.840
+ But it's very easy to then block that by having
+
+27:48.840 --> 27:51.840
+ basically anti negative recognition.
+
+27:51.840 --> 27:55.840
+ It's like if the system sees something that looks like a matrix hack
+
+27:55.840 --> 28:01.840
+ excluded, it's such an easy thing to do.
+
+28:01.840 --> 28:05.840
+ So learn both on the valid data and the invalid data.
+
+28:05.840 --> 28:07.840
+ So basically learn on the adversarial examples
+
+28:07.840 --> 28:09.840
+ to be able to exclude them.
+
+28:09.840 --> 28:12.840
+ Yeah, you basically want to both know what is a car
+
+28:12.840 --> 28:15.840
+ and what is definitely not a car.
+
+28:15.840 --> 28:18.840
+ You train for this is a car and this is definitely not a car.
+
+28:18.840 --> 28:20.840
+ Those are two different things.
+
+28:20.840 --> 28:23.840
+ People have no idea neural nets really.
+
+28:23.840 --> 28:25.840
+ They probably think neural nets involves like, you know,
+
+28:25.840 --> 28:28.840
+ fishing net or something.
+
+28:28.840 --> 28:35.840
+ So as you know, taking a step beyond just Tesla and autopilot,
+
+28:35.840 --> 28:39.840
+ current deep learning approaches still seem in some ways
+
+28:39.840 --> 28:44.840
+ to be far from general intelligence systems.
+
+28:44.840 --> 28:49.840
+ Do you think the current approaches will take us to general intelligence
+
+28:49.840 --> 28:55.840
+ or do totally new ideas need to be invented?
+
+28:55.840 --> 28:59.840
+ I think we're missing a few key ideas for general intelligence,
+
+28:59.840 --> 29:04.840
+ general, artificial general intelligence.
+
+29:04.840 --> 29:08.840
+ But it's going to be upon us very quickly
+
+29:08.840 --> 29:11.840
+ and then we'll need to figure out what shall we do
+
+29:11.840 --> 29:15.840
+ if we even have that choice.
+
+29:15.840 --> 29:18.840
+ But it's amazing how people can't differentiate between, say,
+
+29:18.840 --> 29:22.840
+ the narrow AI that, you know, allows a car to figure out
+
+29:22.840 --> 29:25.840
+ what a lane line is and, you know,
+
+29:25.840 --> 29:29.840
+ and navigate streets versus general intelligence.
+
+29:29.840 --> 29:32.840
+ Like these are just very different things.
+
+29:32.840 --> 29:35.840
+ Like your toaster and your computer are both machines,
+
+29:35.840 --> 29:38.840
+ but one's much more sophisticated than another.
+
+29:38.840 --> 29:43.840
+ You're confident with Tesla you can create the world's best toaster.
+
+29:43.840 --> 29:45.840
+ The world's best toaster, yes.
+
+29:45.840 --> 29:48.840
+ The world's best self driving.
+
+29:48.840 --> 29:51.840
+ I'm, yes.
+
+29:51.840 --> 29:54.840
+ To me, right now, this seems game set match.
+
+29:54.840 --> 29:57.840
+ I don't, I mean, that's, I don't want to be complacent or overconfident,
+
+29:57.840 --> 29:59.840
+ but that's what it appears.
+
+29:59.840 --> 30:02.840
+ That is just literally what it, how it appears right now.
+
+30:02.840 --> 30:06.840
+ It could be wrong, but it appears to be the case
+
+30:06.840 --> 30:10.840
+ that Tesla is vastly ahead of everyone.
+
+30:10.840 --> 30:13.840
+ Do you think we will ever create an AI system
+
+30:13.840 --> 30:17.840
+ that we can love and loves us back in a deep meaningful way
+
+30:17.840 --> 30:20.840
+ like in the movie, Her?
+
+30:20.840 --> 30:23.840
+ I think AI will be capable of convincing you
+
+30:23.840 --> 30:25.840
+ to fall in love with it very well.
+
+30:25.840 --> 30:28.840
+ And that's different than us humans?
+
+30:28.840 --> 30:31.840
+ You know, we start getting into a metaphysical question
+
+30:31.840 --> 30:35.840
+ and do emotions and thoughts exist in a different realm than the physical.
+
+30:35.840 --> 30:37.840
+ And maybe they do, maybe they don't.
+
+30:37.840 --> 30:39.840
+ I don't know, but from a physics standpoint,
+
+30:39.840 --> 30:43.840
+ I tend to think of things, you know,
+
+30:43.840 --> 30:47.840
+ like physics was my main sort of training.
+
+30:47.840 --> 30:50.840
+ And from a physics standpoint,
+
+30:50.840 --> 30:52.840
+ essentially, if it loves you in a way
+
+30:52.840 --> 30:57.840
+ that you can't tell whether it's real or not, it is real.
+
+30:57.840 --> 30:59.840
+ That's a physics view of love.
+
+30:59.840 --> 31:04.840
+ If you cannot prove that it does not,
+
+31:04.840 --> 31:07.840
+ if there's no test that you can apply
+
+31:07.840 --> 31:14.840
+ that would make it allow you to tell the difference,
+
+31:14.840 --> 31:16.840
+ then there is no difference.
+
+31:16.840 --> 31:20.840
+ And it's similar to seeing our world as simulation.
+
+31:20.840 --> 31:22.840
+ There may not be a test to tell the difference
+
+31:22.840 --> 31:24.840
+ between what the real world and the simulation.
+
+31:24.840 --> 31:26.840
+ And therefore, from a physics perspective,
+
+31:26.840 --> 31:28.840
+ it might as well be the same thing.
+
+31:28.840 --> 31:29.840
+ Yes.
+
+31:29.840 --> 31:32.840
+ There may be ways to test whether it's a simulation.
+
+31:32.840 --> 31:35.840
+ There might be, I'm not saying there aren't,
+
+31:35.840 --> 31:38.840
+ but you could certainly imagine that a simulation could correct
+
+31:38.840 --> 31:40.840
+ that once an entity in the simulation
+
+31:40.840 --> 31:42.840
+ found a way to detect the simulation,
+
+31:42.840 --> 31:44.840
+ it could either restart, you know,
+
+31:44.840 --> 31:47.840
+ pause the simulation, start a new simulation,
+
+31:47.840 --> 31:52.840
+ or do one of many other things that then corrects for that error.
+
+31:52.840 --> 31:58.840
+ So when maybe you or somebody else creates an AGI system
+
+31:58.840 --> 32:02.840
+ and you get to ask her one question,
+
+32:02.840 --> 32:16.840
+ what would that question be?
+
+32:16.840 --> 32:21.840
+ What's outside the simulation?
+
+32:21.840 --> 32:23.840
+ Milan, thank you so much for talking today.
+
+32:23.840 --> 32:52.840
+ All right, thank you.
+

vtt/episode_023_small.vtt

@@ -0,0 +1,2177 @@
+WEBVTT
+
+00:00.000 --> 00:04.720
+ The following is a conversation with Gavin Miller, he's the head of Adobe Research.
+
+00:04.720 --> 00:08.960
+ Adobe has empowered artists, designers, and creative minds from all professions,
+
+00:08.960 --> 00:14.320
+ working in the digital medium for over 30 years with software such as Photoshop, Illustrator,
+
+00:14.320 --> 00:20.560
+ Premiere, After Effects, InDesign, Audition, Software that work with images, video, and audio.
+
+00:21.200 --> 00:25.920
+ Adobe Research is working to define the future evolution of these products in a way
+
+00:25.920 --> 00:31.360
+ that makes the life of creatives easier, automates the tedious tasks, and gives more and more time
+
+00:31.360 --> 00:36.880
+ to operate in the idea space instead of pixel space. This is where the cutting edge, deep
+
+00:36.880 --> 00:41.360
+ learning methods of the past decade can really shine more than perhaps any other application.
+
+00:42.240 --> 00:47.840
+ Gavin is the embodiment of combining tech and creativity. Outside of Adobe Research,
+
+00:47.840 --> 00:53.600
+ he writes poetry and builds robots, both things that are near and dear to my heart as well.
+
+00:53.600 --> 00:59.200
+ This conversation is part of the Artificial Intelligence Podcast. If you enjoy it, subscribe
+
+00:59.200 --> 01:05.360
+ on YouTube, iTunes, or simply connect with me on Twitter at Lex Friedman's spelled F R I D.
+
+01:06.000 --> 01:09.600
+ And now here's my conversation with Gavin Miller.
+
+01:11.120 --> 01:15.920
+ You're head of Adobe Research, leading a lot of innovative efforts and applications of AI,
+
+01:15.920 --> 01:23.200
+ creating images, video, audio, language, but you're also yourself an artist, a poet,
+
+01:23.200 --> 01:28.640
+ a writer, and even a roboticist. So while I promise to everyone listening,
+
+01:28.640 --> 01:32.880
+ that I will not spend the entire time we have together reading your poetry, which I love.
+
+01:33.440 --> 01:39.200
+ I have to sprinkle it in at least a little bit. So some of them are pretty deep and profound,
+
+01:39.200 --> 01:43.520
+ and some are light and silly. Let's start with a few lines from the silly variety.
+
+01:43.520 --> 01:56.800
+ You write in a beautiful parody, both Edith Piaf's and my web at Frank Sinatra.
+
+01:56.800 --> 02:06.400
+ So it opens with, and now dessert is near. It's time to pay the final total. I've tried to slim
+
+02:06.400 --> 02:14.800
+ all year, but my diets have been anecdotal. So where does that love for poetry come from
+
+02:14.800 --> 02:20.880
+ for you? And if we dissect your mind, how does it all fit together in the bigger puzzle of Dr.
+
+02:20.880 --> 02:27.440
+ Gavin Miller? Well, interesting you chose that one. That was a poem I wrote when I'd been to
+
+02:27.440 --> 02:32.400
+ my doctor and he said, you really need to lose some weight and go on a diet. And whilst the
+
+02:32.400 --> 02:37.200
+ rational part of my brain wanted to do that, the irrational part of my brain was protesting and
+
+02:37.200 --> 02:42.400
+ sort of embraced the opposite idea. I regret nothing, hence. Yes, exactly. Taken to an extreme,
+
+02:42.400 --> 02:49.600
+ I thought it would be funny. Obviously, it's a serious topic for some people. But I think,
+
+02:49.600 --> 02:53.920
+ for me, I've always been interested in writing since I was in high school, as well as doing
+
+02:53.920 --> 02:58.960
+ technology and invention. And sometimes the parallel strands in your life that carry on,
+
+02:58.960 --> 03:05.120
+ and one is more about your private life and one's more about your technological career.
+
+03:05.680 --> 03:10.640
+ And then at sort of happy moments along the way, sometimes the two things touch, one idea informs
+
+03:10.640 --> 03:17.040
+ the other. And we can talk about that as we go. Do you think you're writing the art, the poetry
+
+03:17.040 --> 03:23.440
+ contribute indirectly or directly to your research, to your work in Adobe? Well, sometimes it does if
+
+03:23.440 --> 03:30.000
+ I say, imagine a future in a science fiction kind of way. And then once it exists on paper,
+
+03:30.000 --> 03:37.520
+ I think, well, why shouldn't I just build that? There was an example where when realistic voice
+
+03:37.520 --> 03:42.800
+ synthesis first started in the 90s at Apple, where I worked in research. I was done by a friend of mine.
+
+03:44.000 --> 03:48.640
+ I sort of sat down and started writing a poem which each line I would enter into the voice
+
+03:48.640 --> 03:54.240
+ synthesizer and see how it sounded and sort of wrote it for that voice. And at the time,
+
+03:55.040 --> 04:00.160
+ the agents weren't very sophisticated. So they'd sort of add random intonation. And I kind of made
+
+04:00.160 --> 04:06.560
+ up the poem to sort of match the tone of the voice. And it sounded slightly sad and depressed. So I
+
+04:06.560 --> 04:12.720
+ pretended it was a poem written by an intelligent agent, sort of telling the user to go home and
+
+04:12.720 --> 04:16.560
+ leave them alone. But at the same time, they were lonely and wanted to have company and learn from
+
+04:16.560 --> 04:21.760
+ what the user was saying. And at the time, it was way beyond anything that AI could possibly do.
+
+04:21.760 --> 04:27.520
+ But, you know, since then, it's becoming more within the bounds of possibility.
+
+04:29.040 --> 04:34.800
+ And then at the same time, I had a project at home where I did sort of a smart home. This was
+
+04:34.800 --> 04:40.960
+ probably 93, 94. And I had the talking voice who'd remind me when I walked in the door of what
+
+04:40.960 --> 04:45.600
+ things I had to do. I had buttons on my washing machine because I was a bachelor and I'd leave
+
+04:45.600 --> 04:49.920
+ the clothes in there for three days and they'd go moldy. So as I got up in the morning, I would say,
+
+04:49.920 --> 04:56.640
+ don't forget the washing and so on. I made photographic photo albums that used light
+
+04:56.640 --> 05:01.040
+ sensors to know which page you were looking at would send that over wireless radio to the agent
+
+05:01.040 --> 05:05.760
+ who would then play sounds that matched the image she were looking at in the book. So I was kind of
+
+05:05.760 --> 05:10.480
+ in love with this idea of magical realism and whether it was possible to do that with technology.
+
+05:10.480 --> 05:16.080
+ So that was a case where the sort of the agent sort of intrigued me from a literary point of
+
+05:16.080 --> 05:22.880
+ view and became a personality. I think more recently, I've also written plays and when
+
+05:22.880 --> 05:27.440
+ plays you write dialogue and obviously you write a fixed set of dialogue that follows a linear
+
+05:27.440 --> 05:33.360
+ narrative. But with modern agents, as you design a personality or a capability for conversation,
+
+05:33.360 --> 05:37.760
+ you're sort of thinking of, I kind of have imaginary dialogue in my head. And then I think,
+
+05:37.760 --> 05:43.680
+ what would it take not only to have that be real, but for it to really know what it's talking about.
+
+05:44.240 --> 05:49.440
+ So it's easy to fall into the uncanny valley with AI where it says something it doesn't really
+
+05:49.440 --> 05:54.560
+ understand, but it sounds good to the person. But you rapidly realize that it's kind of just
+
+05:55.520 --> 06:00.000
+ stimulus response. It doesn't really have real world knowledge about the thing it's describing.
+
+06:00.640 --> 06:06.320
+ And so when you get to that point, it really needs to have multiple ways of talking about
+
+06:06.320 --> 06:10.560
+ the same concept. So it sounds as though it really understands it. Now, what really understanding
+
+06:10.560 --> 06:16.160
+ means is in the eye of the beholder, right? But if it only has one way of referring to something,
+
+06:16.160 --> 06:21.200
+ it feels like it's a canned response. But if it can reason about it, or you can go at it from
+
+06:21.200 --> 06:25.600
+ multiple angles and give a similar kind of response that people would, then it starts to
+
+06:26.400 --> 06:30.480
+ seem more like there's something there that's sentient.
+
+06:31.040 --> 06:35.600
+ You can say the same thing, multiple things from different perspectives. I mean, with the
+
+06:35.600 --> 06:40.000
+ automatic image captioning that I've seen the work that you're doing, there's elements of that,
+
+06:40.000 --> 06:46.000
+ right? Being able to generate different kinds of... Right. So one in my team, there's a lot of work on
+
+06:46.640 --> 06:52.000
+ turning a medium from one form to another, whether it's auto tagging imagery or making up full
+
+06:52.000 --> 06:57.840
+ sentences about what's in the image, then changing the sentence, finding another image that matches
+
+06:57.840 --> 07:04.720
+ the new sentence or vice versa. And in the modern world of GANs, you sort of give it a description
+
+07:04.720 --> 07:11.360
+ and it synthesizes an asset that matches the description. So I've sort of gone on a journey.
+
+07:11.360 --> 07:16.560
+ My early days in my career were about 3D computer graphics, the sort of pioneering work sort of
+
+07:16.560 --> 07:22.400
+ before movies had special effects done with 3D graphics and sort of rode that revolution. And
+
+07:22.400 --> 07:26.720
+ that was very much like the renaissance where people would model light and color and shape
+
+07:26.720 --> 07:32.160
+ and everything. And now we're kind of in another wave where it's more impressionistic and it's
+
+07:32.160 --> 07:38.240
+ sort of the idea of something can be used to generate an image directly, which is sort of the
+
+07:38.240 --> 07:45.520
+ new frontier in computer image generation using AI algorithms. So the creative process is more in
+
+07:45.520 --> 07:49.840
+ the space of ideas or becoming more in the space of ideas versus in the raw pixels?
+
+07:50.720 --> 07:55.280
+ Well, it's interesting. It depends. I think at Adobe, we really want to span the entire range
+
+07:55.280 --> 08:01.040
+ from really, really good, what you might call low level tools by low level, as close to say analog
+
+08:01.040 --> 08:07.040
+ workflows as possible. So what we do there is we make up systems that do really realistic oil
+
+08:07.040 --> 08:11.600
+ paint and watercolor simulation. So if you want every bristle to behave as it would in the real
+
+08:11.600 --> 08:17.760
+ world and leave a beautiful analog trail of water and then flow after you've made the brushstroke,
+
+08:17.760 --> 08:21.600
+ you can do that. And that's really important for people who want to create something
+
+08:22.720 --> 08:28.160
+ really expressive or really novel because they have complete control. And then a certain other
+
+08:28.160 --> 08:35.600
+ task become automated. It frees the artists up to focus on the inspiration and less of the perspiration.
+
+08:35.600 --> 08:43.920
+ So thinking about different ideas, obviously, once you finish the design, there's a lot of work to
+
+08:43.920 --> 08:49.840
+ say do it for all the different aspect ratio of phones or websites and so on. And that used to
+
+08:49.840 --> 08:55.040
+ take up an awful lot of time for artists. It still does for many, what we call content velocity.
+
+08:55.040 --> 09:01.360
+ And one of the targets of AI is actually to reason about from the first example of what are the
+
+09:01.360 --> 09:06.960
+ likely intent for these other formats, maybe if you change the language to German and the words
+
+09:06.960 --> 09:12.160
+ are longer, how do you reflow everything so that it looks nicely artistic in that way.
+
+09:12.160 --> 09:17.360
+ And so the person can focus on the really creative bit in the middle, which is what is the look and
+
+09:17.360 --> 09:21.200
+ style and feel and what's the message and what's the story and the human element.
+
+09:21.200 --> 09:27.920
+ So I think creativity is changing. So that's one way in which we're trying to just make it easier
+
+09:27.920 --> 09:32.880
+ and faster and cheaper to do so that there can be more of it, more demand, because it's less
+
+09:32.880 --> 09:39.200
+ expensive. So everyone wants beautiful artwork for everything from a school website to Hollywood movie.
+
+09:40.800 --> 09:46.480
+ On the other side, as some of these things have automatic versions of them, people will
+
+09:46.480 --> 09:52.480
+ possibly change role from being the hands on artist and to being either the art director or
+
+09:52.480 --> 09:57.920
+ the conceptual artist. And then the computer will be a partner to help create polished examples of
+
+09:57.920 --> 10:02.720
+ the idea that they're exploring. Let's talk about Adobe products versus AI and Adobe products.
+
+10:04.000 --> 10:11.200
+ Just so you know where I'm coming from, I'm a huge fan of Photoshop for images premiere for video,
+
+10:11.200 --> 10:17.200
+ audition for audio. I'll probably use Photoshop to create the thumbnail for this video, premiere
+
+10:17.200 --> 10:24.800
+ to edit the video, audition to do the audio. That said, everything I do is really manually. And I
+
+10:24.800 --> 10:30.640
+ set up, I use this old school kinesis keyboard and I have auto hotkey that just it's really about
+
+10:30.640 --> 10:37.280
+ optimizing the flow of just making sure there's as few clicks as possible. So just being extremely
+
+10:37.280 --> 10:43.920
+ efficient. It's something you started to speak to. So before we get into the fun, sort of awesome
+
+10:43.920 --> 10:48.880
+ deep learning things, where does AI, if you could speak a little more to it AI or just
+
+10:48.880 --> 10:57.280
+ automation in general, do you see in the coming months and years or in general prior in 2018
+
+10:58.560 --> 11:03.840
+ fitting into making the life, the low level pixel work flow easier?
+
+11:03.840 --> 11:10.160
+ Yeah, that's a great question. So we have a very rich array of algorithms already in Photoshop,
+
+11:10.160 --> 11:17.600
+ just classical procedural algorithms as well as ones based on data. In some cases, they end up
+
+11:17.600 --> 11:22.560
+ with a large number of sliders and degrees of freedom. So one way in which AI can help is just
+
+11:22.560 --> 11:27.520
+ an auto button which comes up with default settings based on the content itself rather than
+
+11:27.520 --> 11:34.080
+ default values for the tool. At that point, you then start tweaking. So that's that's a very kind of
+
+11:34.080 --> 11:39.600
+ make life easier for people whilst making use of common sense from other example images.
+
+11:39.600 --> 11:40.960
+ So like smart defaults.
+
+11:40.960 --> 11:47.760
+ Smart defaults, absolutely. Another one is something we've spent a lot of work over the last
+
+11:47.760 --> 11:53.360
+ 20 years. I've been at Adobe 19 thinking about selection, for instance, where
+
+11:53.360 --> 11:58.720
+ you know, with a quick select, you would look at color boundaries and figure out how to sort of
+
+11:58.720 --> 12:02.480
+ flood fill into regions that you thought were physically connected in the real world.
+
+12:03.360 --> 12:08.080
+ But that algorithm had no visual common sense about what a cat looks like or a dog. It would just do
+
+12:08.080 --> 12:14.080
+ it based on rules of thumb, which were applied to graph theory. And it was a big improvement over
+
+12:14.080 --> 12:19.600
+ the previous work we had sort of almost click every everything by hand or if it just did similar
+
+12:19.600 --> 12:24.880
+ colors, it would do little tiny regions that wouldn't be connected. But in the future,
+
+12:24.880 --> 12:31.120
+ using neural nets to actually do a great job with say a single click or even in the case of
+
+12:31.120 --> 12:36.160
+ well known categories like people or animals, no click, where you just say select the object and
+
+12:36.160 --> 12:41.440
+ it just knows the dominant object as a person in the middle of the photograph. Those kinds of things
+
+12:41.440 --> 12:47.520
+ are really valuable if they can be robust enough to give you good quality results.
+
+12:47.520 --> 12:50.720
+ Or they can be a great start for like tweaking it.
+
+12:50.720 --> 12:56.080
+ So for example, background removal, like one thing I'll, in a thumbnail,
+
+12:56.080 --> 13:00.240
+ I'll take a picture of you right now and essentially remove the background behind you.
+
+13:00.240 --> 13:06.480
+ And I want to make that as easy as possible. You don't have flowing hair, like rich at the
+
+13:06.480 --> 13:11.040
+ moment. Rich sort of. I had it in the past, it may come again in the future, but for now.
+
+13:12.320 --> 13:16.160
+ So that sometimes makes it a little more challenging to remove the background.
+
+13:16.160 --> 13:23.680
+ How difficult do you think is that problem for AI for basically making the quick selection tool
+
+13:23.680 --> 13:26.960
+ smarter and smarter and smarter? Well, we have a lot of research on that already.
+
+13:28.400 --> 13:34.160
+ If you want a sort of quick, cheap and cheerful, look, I'm pretending I'm in Hawaii,
+
+13:34.160 --> 13:38.400
+ but it's sort of a joke, then you don't need perfect boundaries. And you can do that today
+
+13:38.400 --> 13:44.960
+ with a single click for the algorithms we have. We have other algorithms where with a little bit
+
+13:44.960 --> 13:48.560
+ more guidance on the boundaries, like you might need to touch it up a little bit.
+
+13:49.920 --> 13:56.480
+ We have other algorithms that can pull a nice mat from a crude selection. So we have combinations
+
+13:56.480 --> 14:04.000
+ of tools that can do all of that. And at our recent Max conference at AB Max, we demonstrated how
+
+14:04.880 --> 14:09.680
+ very quickly just by drawing a simple polygon around the object of interest, we could not
+
+14:09.680 --> 14:17.920
+ only do it for a single still, but we could pull at least a selection mask from a moving target,
+
+14:17.920 --> 14:23.360
+ like a person dancing in front of a brick wall or something. And so it's going from hours to
+
+14:23.360 --> 14:29.360
+ a few seconds for workflows that are really nice. And then you might go in and touch up a little.
+
+14:30.240 --> 14:33.040
+ So that's a really interesting question. You mentioned the word robust.
+
+14:33.040 --> 14:40.000
+ You know, there's like a journey for an idea, right? And what you presented probably at Max
+
+14:41.360 --> 14:46.320
+ has elements of just sort of it inspires the concept, it can work pretty well in a majority
+
+14:46.320 --> 14:51.520
+ of cases. But how do you make something that works? Well, in majority of cases, how do you make
+
+14:51.520 --> 14:56.480
+ something that works, maybe in all cases, or it becomes a robust tool?
+
+14:56.480 --> 15:01.760
+ There are a couple of things. So that really touches on the difference between academic research
+
+15:01.760 --> 15:06.640
+ and industrial research. So in academic research, it's really about who's the person to have the
+
+15:06.640 --> 15:12.480
+ great new idea that shows promise. And we certainly love to be those people too. But
+
+15:13.120 --> 15:17.840
+ we have sort of two forms of publishing. One is academic peer review, which we do a lot of,
+
+15:17.840 --> 15:24.720
+ and we have great success there as much as some universities. But then we also have shipping,
+
+15:24.720 --> 15:30.160
+ which is a different type of, and then we get customer review, as well as, you know, product
+
+15:30.160 --> 15:37.840
+ critics. And that might be a case where it's not about being perfect every single time, but
+
+15:37.840 --> 15:43.280
+ perfect enough at the time, plus a mechanism to intervene and recover where you do have mistakes.
+
+15:43.280 --> 15:47.120
+ So we have the luxury of very talented customers. We don't want them to be
+
+15:48.720 --> 15:55.200
+ overly taxed doing it every time. But if they can go in and just take it from 99 to 100,
+
+15:55.200 --> 16:01.440
+ 100 with the touch of a mouse or something, then for the professional end, that's something
+
+16:01.440 --> 16:06.320
+ that we definitely want to support as well. And for them, it went from having to do that
+
+16:06.320 --> 16:13.360
+ tedious task all the time to much less often. So I think that gives us an out. If it had to be
+
+16:13.360 --> 16:18.640
+ 100% automatic all the time, then that would delay the time at which we could get to market.
+
+16:18.640 --> 16:26.000
+ So on that thread, maybe you can untangle something. Again, I'm sort of just speaking to
+
+16:26.000 --> 16:36.000
+ my own experience. Maybe that is the most useful idea. So I think Photoshop, as an example or premiere,
+
+16:37.680 --> 16:44.240
+ has a lot of amazing features that I haven't touched. And so what's the, in terms of AI,
+
+16:44.240 --> 16:54.080
+ helping make my life or the life of creatives easier? How this collaboration between human
+
+16:54.080 --> 16:58.960
+ and machine, how do you learn to collaborate better? How do you learn the new algorithms?
+
+17:00.000 --> 17:04.240
+ Is it something that where you have to watch tutorials and you have to watch videos and so
+
+17:04.240 --> 17:10.400
+ on? Or do you ever think, do you think about the experience itself through exploration being
+
+17:10.400 --> 17:18.320
+ the teacher? We absolutely do. So I'm glad that you brought this up. We sort of think about
+
+17:18.320 --> 17:22.080
+ two things. One is helping the person in the moment to do the task that they need to do. But
+
+17:22.080 --> 17:26.960
+ the other is thinking more holistically about their journey learning a tool. And when it's like,
+
+17:26.960 --> 17:31.120
+ think of it as Adobe University, where you use the tool long enough, you become an expert.
+
+17:31.120 --> 17:34.960
+ And not necessarily an expert in everything. It's like living in a city. You don't necessarily
+
+17:34.960 --> 17:40.160
+ know every street, but you know, the important ones you need to get to. So we have projects in
+
+17:40.160 --> 17:45.360
+ research, which actually look at the thousands of hours of tutorials online and try to understand
+
+17:45.360 --> 17:51.040
+ what's being taught in them. And then we had one publication at CHI where it was looking at,
+
+17:52.560 --> 17:57.360
+ given the last three or four actions you did, what did other people in tutorials do next?
+
+17:57.360 --> 18:01.680
+ So if you want some inspiration for what you might do next, or you just want to watch the
+
+18:01.680 --> 18:06.800
+ tutorial and see, learn from people who are doing similar workflows to you, you can without having
+
+18:06.800 --> 18:13.360
+ to go and search on keywords and everything. So really trying to use the context of your use of
+
+18:13.360 --> 18:18.640
+ the app to make intelligent suggestions, either about choices that you might make,
+
+18:20.800 --> 18:26.480
+ or in a more assistive way where it could say, if you did this next, we could show you. And that's
+
+18:26.480 --> 18:31.360
+ basically the frontier that we're exploring now, which is, if we really deeply understand the
+
+18:31.360 --> 18:38.480
+ domain in which designers and creative people work, can we combine that with AI and pattern
+
+18:38.480 --> 18:47.520
+ matching of behavior to make intelligent suggestions, either through verbal possibilities or just
+
+18:47.520 --> 18:53.840
+ showing the results of if you try this. And that's really the sort of, I was in a meeting today
+
+18:53.840 --> 18:58.880
+ thinking about these things. So it's still a grand challenge. We'd all love
+
+18:58.880 --> 19:05.920
+ an artist over one shoulder and a teacher over the other, right? And we hope to get there. And
+
+19:05.920 --> 19:10.640
+ the right thing to do is to give enough at each stage that it's useful in itself, but it builds
+
+19:10.640 --> 19:19.120
+ a foundation for the next level of expectation. Are you aware of this gigantic medium of YouTube
+
+19:19.120 --> 19:26.240
+ that's creating just a bunch of creative people, both artists and teachers of different kinds?
+
+19:26.240 --> 19:31.440
+ Absolutely. And the more we can understand those media types, both visually and in terms of
+
+19:31.440 --> 19:37.200
+ transcripts and words, the more we can bring the wisdom that they embody into the guidance that's
+
+19:37.200 --> 19:42.960
+ embedded in the tool. That would be brilliant to remove the barrier from having to yourself type
+
+19:42.960 --> 19:49.600
+ in the keyword, searching, so on. Absolutely. And then in the longer term, an interesting
+
+19:49.600 --> 19:54.000
+ discussion is, does it ultimately not just assist with learning the interface we have,
+
+19:54.000 --> 19:59.200
+ but does it modify the interface to be simpler? Or do you fragment into a variety of tools,
+
+19:59.200 --> 20:05.600
+ each of which has a different level of visibility of the functionality? I like to say that if you
+
+20:05.600 --> 20:12.640
+ add a feature to a GUI, you have to have yet more visual complexity confronting the new user.
+
+20:12.640 --> 20:17.120
+ Whereas if you have an assistant with a new skill, if you know they have it, so you know
+
+20:17.120 --> 20:23.280
+ to ask for it, then it's sort of additive without being more intimidating. So we definitely think
+
+20:23.280 --> 20:29.120
+ about new users and how to onboard them. Many actually value the idea of being able to master
+
+20:29.120 --> 20:34.720
+ that complex interface and keyboard shortcuts, like you were talking about earlier, because
+
+20:35.520 --> 20:39.680
+ with great familiarity, it becomes a musical instrument for expressing your visual ideas.
+
+20:40.480 --> 20:45.840
+ And other people just want to get something done quickly in the simplest way possible,
+
+20:45.840 --> 20:50.400
+ and that's where a more assistive version of the same technology might be useful,
+
+20:50.400 --> 20:54.560
+ maybe on a different class of device, which is more in context for capture, say,
+
+20:55.920 --> 21:01.680
+ whereas somebody who's in a deep post production workflow maybe want to be on a laptop or a big
+
+21:01.680 --> 21:10.560
+ screen desktop and have more knobs and dials to really express the subtlety of what they want to do.
+
+21:12.160 --> 21:16.320
+ So there's so many exciting applications of computer vision and machine learning
+
+21:16.320 --> 21:21.920
+ that Adobe is working on, like scene stitching, sky replacement, foreground,
+
+21:21.920 --> 21:26.880
+ background removal, spatial object based image search, automatic image captioning,
+
+21:26.880 --> 21:31.280
+ like we mentioned, project cloak, project deep fill filling in parts of the images,
+
+21:31.920 --> 21:38.640
+ project scribbler, style transfer video, style transfer faces and video with Project Puppetron,
+
+21:38.640 --> 21:49.040
+ best name ever. Can you talk through a favorite or some of them or examples that pop in mind?
+
+21:49.040 --> 21:54.800
+ I'm sure I'll be able to provide links to other ones we don't talk about because there's visual
+
+21:54.800 --> 22:00.640
+ elements to all of them that are exciting. Why they're interesting for different reasons might
+
+22:00.640 --> 22:06.640
+ be a good way to go. So I think sky replace is interesting because we talked about selection
+
+22:06.640 --> 22:11.440
+ being sort of an atomic operation. It's almost like a, if you think of an assembly language,
+
+22:11.440 --> 22:17.840
+ it's like a single instruction. Whereas sky replace is a compound action where you automatically
+
+22:17.840 --> 22:22.720
+ select the sky, you look for stock content that matches the geometry of the scene.
+
+22:24.000 --> 22:27.600
+ You try to have variety in your choices so that you do coverage of different moods.
+
+22:28.160 --> 22:35.600
+ It then mats in the sky behind the foreground, but then importantly it uses the foreground
+
+22:35.600 --> 22:40.560
+ of the other image that you just searched on to recolor the foreground of the image that
+
+22:40.560 --> 22:47.840
+ you're editing. So if you say go from a midday sky to an evening sky, it will actually add
+
+22:47.840 --> 22:53.760
+ sort of an orange glow to the foreground objects as well. I was a big fan in college of Magritte
+
+22:53.760 --> 22:59.600
+ and he has a number of paintings where it's surrealism because he'll like do a composite,
+
+22:59.600 --> 23:03.440
+ but the foreground building will be at night and the sky will be during the day. There's one
+
+23:03.440 --> 23:09.120
+ called The Empire of Light which was on my wall in college and we're trying not to do surrealism.
+
+23:09.120 --> 23:15.680
+ It can be a choice, but we'd rather have it be natural by default rather than it looking fake
+
+23:15.680 --> 23:19.760
+ and then you have to do a whole bunch of post production to fix it. So that's a case where
+
+23:19.760 --> 23:25.120
+ we're kind of capturing an entire workflow into a single action and doing it in about a second
+
+23:25.120 --> 23:30.720
+ rather than a minute or two. And when you do that, you can not just do it once, but you can do it
+
+23:30.720 --> 23:36.960
+ for say like 10 different backgrounds and then you're almost back to this inspiration idea of
+
+23:36.960 --> 23:41.840
+ I don't know quite what I want, but I'll know it when I see it. And you can just explore the
+
+23:41.840 --> 23:47.200
+ design space as close to final production value as possible. And then when you really pick one,
+
+23:47.200 --> 23:51.120
+ you might go back and slightly tweak the selection mask just to make it perfect and
+
+23:51.120 --> 23:54.320
+ do that kind of polish that professionals like to bring to their work.
+
+23:54.320 --> 24:01.040
+ So then there's this idea of, as you mentioned, the sky replacing it to different stock images of
+
+24:01.040 --> 24:07.040
+ the sky. In general, you have this idea or it could be on your disk or whatever. But making even
+
+24:07.040 --> 24:12.400
+ more intelligent choices about ways to search stock images which is really interesting. It's
+
+24:12.400 --> 24:19.120
+ kind of spatial. Absolutely. Right. So that was something we called Concept Canvas. So normally
+
+24:19.120 --> 24:26.240
+ when you do say an image search, I assume it's just based on text. You would give the keywords
+
+24:26.240 --> 24:30.080
+ of the things you want to be in the image and it would find the nearest one that had those tags.
+
+24:32.720 --> 24:37.200
+ For many tasks, you really want to be able to say I want a big person in the middle or in a
+
+24:37.200 --> 24:41.280
+ dog to the right and umbrella above the left because you want to leave space for the text or
+
+24:41.280 --> 24:48.640
+ whatever. And so Concept Canvas lets you assign spatial regions to the keywords. And then we've
+
+24:48.640 --> 24:54.560
+ already preindexed the images to know where the important concepts are in the picture. So we then
+
+24:54.560 --> 25:01.200
+ go through that index matching to assets. And even though it's just another form of search,
+
+25:01.200 --> 25:06.480
+ because you're doing spatial design or layout, it starts to feel like design. You sort of feel
+
+25:06.480 --> 25:13.120
+ oddly responsible for the image that comes back as if you invented it a little bit. So it's a good
+
+25:13.120 --> 25:18.960
+ example where giving enough control starts to make people have a sense of ownership over the
+
+25:18.960 --> 25:23.280
+ outcome of the event. And then we also have technologies in Photoshop where you physically
+
+25:23.280 --> 25:29.440
+ can move the dog in post as well. But for Concept Canvas, it was just a very fast way to sort of
+
+25:29.440 --> 25:38.560
+ loop through and be able to lay things out. In terms of being able to remove objects from a scene
+
+25:38.560 --> 25:45.920
+ and fill in the background automatically. So that's extremely exciting. And that's
+
+25:45.920 --> 25:51.200
+ so neural networks are stepping in there. I just talked this week with Ian Goodfellow.
+
+25:51.200 --> 25:55.360
+ So the GANS for doing that is definitely one approach.
+
+25:55.360 --> 25:59.440
+ So is that a really difficult problem? Is it as difficult as it looks,
+
+25:59.440 --> 26:06.080
+ again, to take it to a robust product level? Well, there are certain classes of image for
+
+26:06.080 --> 26:10.800
+ which the traditional algorithms like Content Aware Fill work really well. Like if you have
+
+26:10.800 --> 26:15.200
+ a naturalistic texture like a gravel path or something, because it's patch based, it will
+
+26:15.200 --> 26:19.840
+ make up a very plausible looking intermediate thing and fill in the hole. And then we use some
+
+26:20.960 --> 26:25.200
+ algorithms to sort of smooth out the lighting so you don't see any brightness contrasts in that
+
+26:25.200 --> 26:29.680
+ region or you've gradually ramped from dark to light if it straddles the boundary.
+
+26:29.680 --> 26:37.600
+ Where it gets complicated is if you have to infer invisible structure behind the person in front.
+
+26:37.600 --> 26:41.920
+ And that really requires a common sense knowledge of the world to know what,
+
+26:42.480 --> 26:47.040
+ if I see three quarters of a house, do I have a rough sense of what the rest of the house looks
+
+26:47.040 --> 26:51.840
+ like? If you just fill it in with patches, it can end up sort of doing things that make sense
+
+26:51.840 --> 26:56.480
+ locally. But you look at the global structure and it looks like it's just sort of crumpled or messed
+
+26:56.480 --> 27:02.800
+ up. And so what GANs and neural nets bring to the table is this common sense learned from the
+
+27:02.800 --> 27:10.640
+ training set. And the challenge right now is that the generative methods that can make up
+
+27:10.640 --> 27:14.960
+ missing holes using that kind of technology are still only stable at low resolutions.
+
+27:15.520 --> 27:19.680
+ And so you either need to then go from a low resolution to a high resolution using some other
+
+27:19.680 --> 27:24.720
+ algorithm or we need to push the state of the art and it's still in research to get to that point.
+
+27:24.720 --> 27:30.720
+ Right. Of course, if you show it something, say it's trained on houses and then you show it in
+
+27:30.720 --> 27:37.360
+ octopus, it's not going to do a very good job of showing common sense about octopuses. So
+
+27:39.360 --> 27:45.600
+ again, you're asking about how you know that it's ready for prime time. You really need a very
+
+27:45.600 --> 27:52.880
+ diverse training set of images. And ultimately, that may be a case where you put it out there
+
+27:52.880 --> 28:00.400
+ with some guard rails where you might do a detector which looks at the image and
+
+28:00.960 --> 28:05.280
+ sort of estimates its own competence of how well a job could this algorithm do.
+
+28:05.920 --> 28:10.400
+ So eventually, there may be this idea of what we call an ensemble of experts where
+
+28:11.120 --> 28:15.440
+ any particular expert is specialized in certain things and then there's sort of a
+
+28:15.440 --> 28:19.360
+ either they vote to say how confident they are about what to do. This is sort of more future
+
+28:19.360 --> 28:24.080
+ looking or there's some dispatcher which says you're good at houses, you're good at trees.
+
+28:27.120 --> 28:31.520
+ So I mean, all this adds up to a lot of work because each of those models will be a whole
+
+28:31.520 --> 28:38.320
+ bunch of work. But I think over time, you'd gradually fill out the set and initially focus
+
+28:38.320 --> 28:41.520
+ on certain workflows and then sort of branch out as you get more capable.
+
+28:41.520 --> 28:48.640
+ So you mentioned workflows and have you considered maybe looking far into the future?
+
+28:50.000 --> 28:57.680
+ First of all, using the fact that there is a huge amount of people that use Photoshop,
+
+28:57.680 --> 29:03.520
+ for example, they have certain workflows, being able to collect the information by which
+
+29:04.880 --> 29:09.440
+ they basically get information about their workflows, about what they need,
+
+29:09.440 --> 29:15.120
+ what can the ways to help them, whether it is houses or octopus that people work on more.
+
+29:16.000 --> 29:23.440
+ Basically getting a beat on what kind of data is needed to be annotated and collected for people
+
+29:23.440 --> 29:26.320
+ to build tools that actually work well for people.
+
+29:26.320 --> 29:31.680
+ Absolutely. And this is a big topic and the whole world of AI is what data can you gather and why.
+
+29:33.200 --> 29:39.120
+ At one level, the way to think about it is we not only want to train our customers in how to use
+
+29:39.120 --> 29:44.160
+ our products, but we want them to teach us what's important and what's useful. At the same time,
+
+29:44.160 --> 29:51.120
+ we want to respect their privacy and obviously we wouldn't do things without their explicit permission.
+
+29:52.800 --> 29:57.440
+ And I think the modern spirit of the age around this is you have to demonstrate to somebody
+
+29:57.440 --> 30:02.720
+ how they're benefiting from sharing their data with the tool. Either it's helping in the short
+
+30:02.720 --> 30:07.120
+ term to understand their intent so you can make better recommendations or if they're
+
+30:07.120 --> 30:11.840
+ there friendly to your cause or you're tall or they want to help you evolve quickly because
+
+30:11.840 --> 30:16.320
+ they depend on you for their livelihood, they may be willing to share some of their
+
+30:17.360 --> 30:23.360
+ workflows or choices with the dataset to be then trained.
+
+30:24.720 --> 30:30.560
+ There are technologies for looking at learning without necessarily storing all the information
+
+30:30.560 --> 30:36.080
+ permanently so that you can learn on the fly but not keep a record of what somebody did.
+
+30:36.080 --> 30:38.720
+ So, we're definitely exploring all of those possibilities.
+
+30:38.720 --> 30:45.520
+ And I think Adobe exists in a space where Photoshop, if I look at the data I've created
+
+30:45.520 --> 30:51.840
+ in OWN, I'm less comfortable sharing data with social networks than I am with Adobe because
+
+30:51.840 --> 31:01.360
+ there's just exactly as you said, there's an obvious benefit for sharing the data that I use
+
+31:01.360 --> 31:05.440
+ to create in Photoshop because it's helping improve the workflow in the future.
+
+31:05.440 --> 31:06.080
+ Right.
+
+31:06.080 --> 31:09.440
+ As opposed to it's not clear what the benefit is in social networks.
+
+31:10.000 --> 31:14.000
+ It's nice of you to say that. I mean, I think there are some professional workflows where
+
+31:14.000 --> 31:17.360
+ people might be very protective of what they're doing such as if I was preparing
+
+31:18.240 --> 31:22.640
+ evidence for a legal case, I wouldn't want any of that, you know,
+
+31:22.640 --> 31:25.440
+ phoning home to help train the algorithm or anything.
+
+31:26.560 --> 31:30.720
+ There may be other cases where people say having a trial version or they're doing some,
+
+31:30.720 --> 31:35.280
+ I'm not saying we're doing this today, but there's a future scenario where somebody has a more
+
+31:35.280 --> 31:40.400
+ permissive relationship with Adobe where they explicitly say, I'm fine, I'm only doing hobby
+
+31:40.400 --> 31:48.000
+ projects or things which are non confidential and in exchange for some benefit tangible or
+
+31:48.000 --> 31:51.200
+ otherwise, I'm willing to share very fine grain data.
+
+31:51.840 --> 31:57.920
+ So, another possible scenario is to capture relatively crude high level things from more
+
+31:57.920 --> 32:02.160
+ people and then more detailed knowledge from people who are willing to participate.
+
+32:02.160 --> 32:07.280
+ We do that today with explicit customer studies where, you know, we go and visit somebody and
+
+32:07.280 --> 32:10.640
+ ask them to try the tool and we human observe what they're doing.
+
+32:12.000 --> 32:15.760
+ In the future, to be able to do that enough to be able to train an algorithm,
+
+32:16.320 --> 32:20.240
+ we'd need a more systematic process, but we'd have to do it very consciously because
+
+32:21.200 --> 32:24.560
+ one of the things people treasure about Adobe is a sense of trust
+
+32:24.560 --> 32:28.880
+ and we don't want to endanger that through overly aggressive data collection.
+
+32:28.880 --> 32:35.520
+ So, we have a Chief Privacy Officer and it's definitely front and center of thinking about AI
+
+32:35.520 --> 32:39.920
+ rather than an afterthought. Well, when you start that program, sign me up.
+
+32:39.920 --> 32:41.040
+ Okay, happy to.
+
+32:42.800 --> 32:48.640
+ Is there other projects that you wanted to mention that I didn't perhaps that pop into mind?
+
+32:48.640 --> 32:51.840
+ Well, you covered the number. I think you mentioned Project Puppetron.
+
+32:51.840 --> 32:58.480
+ I think that one is interesting because you might think of Adobe as only thinking in 2D
+
+32:59.760 --> 33:04.800
+ and that's a good example where we're actually thinking more three dimensionally about how to
+
+33:04.800 --> 33:09.440
+ assign features to faces so that we can, you know, if you take, so what Puppetron does,
+
+33:09.440 --> 33:16.320
+ it takes either a still or a video of a person talking and then it can take a painting of somebody
+
+33:16.320 --> 33:20.160
+ else and then apply the style of the painting to the person who's talking in the video.
+
+33:20.160 --> 33:29.600
+ And it's unlike a sort of screen door post filter effect that you sometimes see online.
+
+33:30.320 --> 33:36.080
+ It really looks as though it's sort of somehow attached or reflecting the motion of the face.
+
+33:36.080 --> 33:40.320
+ And so that's the case where even to do a 2D workflow like stylization,
+
+33:40.320 --> 33:44.160
+ you really need to infer more about the 3D structure of the world.
+
+33:44.160 --> 33:48.560
+ And I think as 3D computer vision algorithms get better,
+
+33:48.560 --> 33:52.960
+ initially they'll focus on particular domains like faces where you have a lot of
+
+33:52.960 --> 33:57.680
+ prior knowledge about structure and you can maybe have a parameterized template that you fit to the
+
+33:57.680 --> 34:03.600
+ image. But over time, this should be possible for more general content. And it might even be
+
+34:03.600 --> 34:09.360
+ invisible to the user that you're doing 3D reconstruction but under the hood, but it might
+
+34:09.360 --> 34:15.200
+ then let you do edits much more reliably or correctly than you would otherwise.
+
+34:15.200 --> 34:20.800
+ And you know, the face is a very important application, right?
+
+34:20.800 --> 34:22.480
+ So making things work.
+
+34:22.480 --> 34:26.640
+ And a very sensitive one. If you do something uncanny, it's very disturbing.
+
+34:26.640 --> 34:30.080
+ That's right. You have to get it. You have to get it right.
+
+34:30.080 --> 34:39.040
+ So in the space of augmented reality and virtual reality, what do you think is the role of AR and
+
+34:39.040 --> 34:45.360
+ VR and in the content we consume as people as consumers and the content we create as creators?
+
+34:45.360 --> 34:47.920
+ No, that's a great question. Let me think about this a lot too.
+
+34:48.720 --> 34:55.360
+ So I think VR and AR serve slightly different purposes. So VR can really transport you to an
+
+34:55.360 --> 35:02.400
+ entire immersive world, no matter what your personal situation is. To that extent, it's a bit like
+
+35:02.400 --> 35:06.720
+ a really, really widescreen television where it sort of snaps you out of your context and puts you
+
+35:06.720 --> 35:13.360
+ in a new one. And I think it's still evolving in terms of the hardware I actually worked on,
+
+35:13.360 --> 35:17.680
+ VR in the 90s, trying to solve the latency and sort of nausea problem, which we did,
+
+35:17.680 --> 35:23.040
+ but it was very expensive and a bit early. There's a new wave of that now. I think
+
+35:23.600 --> 35:27.600
+ and increasingly those devices are becoming all in one rather than something that's tethered to a
+
+35:27.600 --> 35:34.480
+ box. I think the market seems to be bifurcating into things for consumers and things for professional
+
+35:34.480 --> 35:40.080
+ use cases, like for architects and people designing where your product is a building and you really
+
+35:40.080 --> 35:45.200
+ want to experience it better than looking at a scale model or a drawing, I think,
+
+35:45.920 --> 35:50.320
+ or even than a video. So I think for that, where you need a sense of scale and spatial
+
+35:50.320 --> 35:59.600
+ relationships, it's great. I think AR holds the promise of sort of taking digital assets off the
+
+35:59.600 --> 36:03.600
+ screen and putting them in context in the real world on the table in front of you on the wall
+
+36:03.600 --> 36:10.400
+ behind you. And that has the corresponding need that the assets need to adapt to the physical
+
+36:10.400 --> 36:15.680
+ context in which they're being placed. I mean, it's a bit like having a live theater troupe come
+
+36:15.680 --> 36:20.880
+ to your house and put on Hamlet. My mother had a friend who used to do this at Stately Homes in
+
+36:20.880 --> 36:26.640
+ England for the National Trust. And they would adapt the scenes and even they'd walk the audience
+
+36:26.640 --> 36:32.880
+ through the rooms to see the action based on the country house they found themselves in for two
+
+36:32.880 --> 36:38.720
+ days. And I think AR will have the same issue that if you have a tiny table in a big living room
+
+36:38.720 --> 36:44.960
+ or something, it'll try to figure out what can you change and what's fixed. And there's a little
+
+36:44.960 --> 36:52.560
+ bit of a tension between fidelity, where if you captured Senior Eye of doing a fantastic ballet,
+
+36:52.560 --> 36:56.640
+ you'd want it to be sort of exactly reproduced. And maybe all you could do is scale it down.
+
+36:56.640 --> 37:03.760
+ Whereas somebody telling you a story might be walking around the room doing some gestures
+
+37:03.760 --> 37:07.040
+ and that could adapt to the room in which they were telling the story.
+
+37:07.760 --> 37:12.800
+ And do you think fidelity is that important in that space or is it more about the storytelling?
+
+37:12.800 --> 37:17.840
+ I think it may depend on the characteristic of the media. If it's a famous celebrity,
+
+37:17.840 --> 37:22.480
+ then it may be that you want to catch every nuance and they don't want to be reanimated by some
+
+37:22.480 --> 37:30.800
+ algorithm. It could be that if it's really a loveable frog telling you a story and it's
+
+37:30.800 --> 37:34.320
+ about a princess and a frog, then it doesn't matter if the frog moves in a different way.
+
+37:35.520 --> 37:38.640
+ I think a lot of the ideas that have sort of grown up in the game world will
+
+37:39.520 --> 37:45.120
+ now come into the broader commercial sphere once they're needing adaptive characters in AR.
+
+37:45.120 --> 37:51.920
+ Are you thinking of engineering tools that allow creators to create in the augmented world,
+
+37:52.480 --> 37:56.000
+ basically making a Photoshop for the augmented world?
+
+37:57.360 --> 38:02.560
+ Well, we have shown a few demos of sort of taking a Photoshop layer stack and then expanding it into
+
+38:02.560 --> 38:08.640
+ 3D. That's actually been shown publicly as one example in AR. Where we're particularly excited
+
+38:08.640 --> 38:17.120
+ at the moment is in 3D. 3D design is still a very challenging space. We believe that it's
+
+38:17.120 --> 38:22.800
+ a worthwhile experiment to try to figure out if AR or immersive makes 3D design more spontaneous.
+
+38:23.360 --> 38:26.960
+ Can you give me an example of 3D design just like applications?
+
+38:26.960 --> 38:32.080
+ Well, literally, a simple one would be laying out objects, right? On a conventional screen,
+
+38:32.080 --> 38:35.680
+ you'd sort of have a plan view and a side view and a perspective view and you sort of be dragging
+
+38:35.680 --> 38:39.440
+ it around with the mouse and if you're not careful, it would go through the wall and all that.
+
+38:39.440 --> 38:46.400
+ Whereas if you were really laying out objects, say in a VR headset, you could literally move
+
+38:46.400 --> 38:50.720
+ your head to see a different viewpoint. They'd be in stereo, so you'd have a sense of depth
+
+38:50.720 --> 38:57.040
+ because you're already wearing the depth glasses, right? So it would be those sort of big gross
+
+38:57.040 --> 39:01.120
+ motor, move things around, kind of skills seem much more spontaneous just like they are in the
+
+39:01.120 --> 39:08.320
+ real world. The frontier for us, I think, is whether that same medium can be used to do fine
+
+39:08.320 --> 39:14.880
+ grain design tasks, like very accurate constraints on, say, a CAD model or something. That may be
+
+39:14.880 --> 39:19.120
+ better done on a desktop, but it may just be a matter of inventing the right UI.
+
+39:20.160 --> 39:27.600
+ So we're hopeful that because there will be this potential explosion of demand for 3D assets
+
+39:27.600 --> 39:32.560
+ that's driven by AR and more real time animation on conventional screens,
+
+39:34.880 --> 39:40.800
+ those tools will also help with, or those devices will help with designing the content as well.
+
+39:40.800 --> 39:46.480
+ You've mentioned quite a few interesting sort of new ideas. At the same time, there's old
+
+39:46.480 --> 39:51.280
+ timers like me that are stuck in their old ways. I think I'm the old timer.
+
+39:51.280 --> 39:58.560
+ Okay. All right. But the opposed all change at all costs. When you're thinking about
+
+39:58.560 --> 40:05.440
+ creating new interfaces, do you feel the burden of just this giant user base that loves the
+
+40:05.440 --> 40:13.760
+ current product? So anything new you do that any new idea comes at a cost that you'll be resisted?
+
+40:13.760 --> 40:21.200
+ Well, I think if you have to trade off control for convenience, then our existing user base
+
+40:21.200 --> 40:27.680
+ would definitely be offended by that. I think if there are some things where you have more convenience
+
+40:27.680 --> 40:34.160
+ and just as much control, that may be more welcome. We do think about not breaking well known
+
+40:34.160 --> 40:40.640
+ metaphors for things. So things should sort of make sense. Photoshop has never been a static
+
+40:40.640 --> 40:46.640
+ target. It's always been evolving and growing. And to some extent, there's been a lot of brilliant
+
+40:46.640 --> 40:50.640
+ thought along the way of how it works today. So we don't want to just throw all that out.
+
+40:51.840 --> 40:55.600
+ If there's a fundamental breakthrough, like a single click is good enough to select an object
+
+40:55.600 --> 41:01.760
+ rather than having to do lots of strokes, that actually fits in quite nicely to the existing
+
+41:01.760 --> 41:07.840
+ tool set, either as an optional mode or as a starting point. I think where we're looking at
+
+41:07.840 --> 41:13.440
+ radical simplicity where you could encapsulate an entire workflow with a much simpler UI,
+
+41:14.000 --> 41:18.640
+ then sometimes that's easier to do in the context of either a different device like a
+
+41:18.640 --> 41:25.120
+ mobile device where the affordances are naturally different or in a tool that's targeted at a
+
+41:25.120 --> 41:31.040
+ different workflow where it's about spontaneity and velocity rather than precision. And we have
+
+41:31.040 --> 41:36.720
+ projects like Rush, which can let you do professional quality video editing for a certain class of
+
+41:36.720 --> 41:47.440
+ media output that is targeted very differently in terms of users and the experience. And ideally,
+
+41:47.440 --> 41:54.160
+ people would go, if I'm feeling like doing Premiere, big project, I'm doing a four part
+
+41:54.160 --> 41:59.200
+ television series. That's definitely a premier thing. But if I want to do something to show my
+
+41:59.200 --> 42:05.200
+ recent vacation, maybe I'll just use Rush because I can do it in the half an hour. I have free at
+
+42:05.200 --> 42:12.480
+ home rather than the four hours I need to do it at work. And for the use cases which we can do well,
+
+42:12.480 --> 42:16.960
+ it really is much faster to get the same output. But the more professional tools obviously have
+
+42:16.960 --> 42:21.520
+ a much richer toolkit and more flexibility in what they can do.
+
+42:21.520 --> 42:26.400
+ And then at the same time, with the flexibility and control, I like this idea of smart defaults,
+
+42:27.040 --> 42:33.520
+ of using AI to coach you to like what Google has, I'm feeling lucky button.
+
+42:33.520 --> 42:37.360
+ Right. Or one button kind of gives you a pretty good
+
+42:38.160 --> 42:41.440
+ set of settings. And then you almost, that's almost an educational tool.
+
+42:42.000 --> 42:43.360
+ Absolutely. Yeah.
+
+42:43.360 --> 42:49.920
+ To show, because sometimes when you have all this control, you're not sure about the
+
+42:51.040 --> 42:55.600
+ correlation between the different bars that control different elements of the image and so on.
+
+42:55.600 --> 43:00.400
+ And sometimes there's a degree of, you don't know what the optimal is.
+
+43:00.400 --> 43:05.360
+ And then some things are sort of on demand like help, right?
+
+43:05.360 --> 43:06.160
+ Help, yeah.
+
+43:06.160 --> 43:09.600
+ I'm stuck. I need to know what to look for. I'm not quite sure what it's called.
+
+43:10.400 --> 43:13.920
+ And something that was proactively making helpful suggestions or,
+
+43:14.800 --> 43:20.640
+ you know, you could imagine a make a suggestion button where you'd use all of that knowledge
+
+43:20.640 --> 43:25.120
+ of workflows and everything to maybe suggest something to go and learn about or just to try
+
+43:25.120 --> 43:31.280
+ or show the answer. And maybe it's not one intelligent to pothole, but it's like a variety
+
+43:31.280 --> 43:33.760
+ of defaults. And then you go, oh, I like that one.
+
+43:33.760 --> 43:34.960
+ Yeah. Yeah.
+
+43:34.960 --> 43:35.680
+ Several options.
+
+43:37.200 --> 43:39.520
+ So back to poetry.
+
+43:39.520 --> 43:40.640
+ Ah, yes.
+
+43:40.640 --> 43:42.400
+ We're going to interleave.
+
+43:43.440 --> 43:48.000
+ So first few lines of a recent poem of yours before I ask the next question.
+
+43:48.000 --> 43:55.840
+ Yeah. This is about the smartphone. Today left my phone at home and went down to the sea.
+
+43:57.120 --> 44:01.600
+ The sand was soft, the ocean glass, but I was still just me.
+
+44:02.480 --> 44:08.240
+ So this is a poem about you leaving your phone behind and feeling quite liberated because of it.
+
+44:08.960 --> 44:14.800
+ So this is kind of a difficult topic and let's see if we can talk about it, figure it out.
+
+44:14.800 --> 44:21.120
+ But so with the help of AI, more and more, we can create versions of ourselves, versions of
+
+44:21.120 --> 44:31.920
+ reality that are in some ways more beautiful than actual reality. And some of the creative effort
+
+44:31.920 --> 44:38.880
+ there is part of creating this illusion. So of course, this is inevitable, but how do you think
+
+44:38.880 --> 44:44.320
+ we should adjust this human beings to live in this digital world that's partly artificial,
+
+44:44.320 --> 44:51.520
+ that's better than the world that we lived in a hundred years ago when you didn't have
+
+44:51.520 --> 44:55.840
+ Instagram and Facebook versions of ourselves and the online.
+
+44:55.840 --> 44:58.880
+ Oh, this is sort of showing off better versions of ourselves.
+
+44:58.880 --> 45:04.880
+ We're using the tooling of modifying the images or even with artificial intelligence
+
+45:04.880 --> 45:12.880
+ ideas of deep fakes and creating adjusted or fake versions of ourselves and reality.
+
+45:14.080 --> 45:17.600
+ I think it's an interesting question. You're all sort of historical bent on this.
+
+45:19.360 --> 45:24.720
+ I actually wonder if 18th century aristocrats who commissioned famous painters to paint portraits
+
+45:24.720 --> 45:28.960
+ of them had portraits that were slightly nicer than they actually looked in practice.
+
+45:28.960 --> 45:29.680
+ Well played, sir.
+
+45:29.680 --> 45:34.720
+ So human desire to put your best foot forward has always been true.
+
+45:37.440 --> 45:42.240
+ I think it's interesting. You sort of framed it in two ways. One is if we can imagine alternate
+
+45:42.240 --> 45:47.280
+ realities and visualize them, is that a good or bad thing? In the old days, you do it with
+
+45:47.280 --> 45:52.560
+ storytelling and words and poetry, which still resides sometimes on websites. But
+
+45:52.560 --> 46:00.640
+ we've become a very visual culture in particular. In the 19th century, we were very much a text
+
+46:00.640 --> 46:07.280
+ based culture. People would read long tracks. Political speeches were very long. Nowadays,
+
+46:07.280 --> 46:15.280
+ everything's very kind of quick and visual and snappy. I think it depends on how harmless your
+
+46:15.280 --> 46:23.120
+ intent. A lot of it's about intent. So if you have a somewhat flattering photo that you pick
+
+46:23.120 --> 46:30.960
+ out of the photos that you have in your inbox to say, this is what I look like, it's probably fine.
+
+46:32.160 --> 46:37.040
+ If someone's going to judge you by how you look, then they'll decide soon enough when they meet
+
+46:37.040 --> 46:45.120
+ you whether the reality. I think where it can be harmful is if people hold themselves up to an
+
+46:45.120 --> 46:49.520
+ impossible standard, which they then feel bad about themselves for not meeting. I think that's
+
+46:49.520 --> 46:58.400
+ definitely can be an issue. But I think the ability to imagine and visualize an alternate
+
+46:58.400 --> 47:04.800
+ reality, which sometimes which you then go off and build later, can be a wonderful thing too.
+
+47:04.800 --> 47:10.720
+ People can imagine architectural styles, which they then have a startup, make a fortune and then
+
+47:10.720 --> 47:17.680
+ build a house that looks like their favorite video game. Is that a terrible thing? I think
+
+47:18.720 --> 47:24.560
+ I used to worry about exploration actually, that part of the joy of going to the moon
+
+47:24.560 --> 47:30.320
+ when I was a tiny child, I remember it, and grainy black and white, was to know what it would look
+
+47:30.320 --> 47:35.120
+ like when you got there. And I think now we have such good graphics for knowing, for visualizing
+
+47:35.120 --> 47:40.960
+ experience before it happens, that I slightly worry that it may take the edge off actually wanting
+
+47:40.960 --> 47:46.160
+ to go. Because we've seen it on TV, we kind of, oh, by the time we finally get to Mars,
+
+47:46.160 --> 47:53.200
+ we're going, oh yeah, it's Mars, that's what it looks like. But then the outer exploration,
+
+47:53.200 --> 47:58.480
+ I mean, I think Pluto was a fantastic recent discovery where nobody had any idea what it
+
+47:58.480 --> 48:04.800
+ looked like and it was just breathtakingly varied and beautiful. So I think expanding
+
+48:04.800 --> 48:10.800
+ the ability of the human toolkit to imagine and communicate on balance is a good thing.
+
+48:10.800 --> 48:15.920
+ I think there are abuses, we definitely take them seriously and try to discourage them.
+
+48:17.440 --> 48:22.960
+ I think there's a parallel side where the public needs to know what's possible through events like
+
+48:22.960 --> 48:30.720
+ this, right? So that you don't believe everything you read and print anymore, and it may over time
+
+48:30.720 --> 48:35.440
+ become true of images as well. Or you need multiple sets of evidence to really believe
+
+48:35.440 --> 48:40.240
+ something rather than a single media asset. So I think it's a constantly evolving thing.
+
+48:40.240 --> 48:45.680
+ It's been true forever. There's a famous story about Anne of Cleves and Henry VIII where,
+
+48:47.760 --> 48:52.720
+ luckily for Anne, they didn't get married, right? So, or they got married and
+
+48:53.840 --> 48:54.560
+ What's the story?
+
+48:54.560 --> 48:59.200
+ Oh, so Holbein went and painted a picture and then Henry VIII wasn't pleased and, you know,
+
+48:59.200 --> 49:03.520
+ history doesn't record whether Anne was pleased, but I think she was pleased not
+
+49:03.520 --> 49:07.920
+ to be married more than a day or something. So I mean, this has gone on for a long time,
+
+49:07.920 --> 49:13.280
+ but I think it's just part of the magnification of human capability.
+
+49:14.640 --> 49:20.560
+ You've kind of built up an amazing research environment here, research culture, research
+
+49:20.560 --> 49:25.600
+ lab, and you've written that the secret to a thriving research lab is interns. Can you unpack
+
+49:25.600 --> 49:29.520
+ that a little bit? Oh, absolutely. So a couple of reasons.
+
+49:31.360 --> 49:36.080
+ As you see, looking at my personal history, there are certain ideas you bond with at a certain
+
+49:36.080 --> 49:40.880
+ stage of your career and you tend to keep revisiting them through time. If you're lucky,
+
+49:40.880 --> 49:44.880
+ you pick one that doesn't just get solved in the next five years, and then you're sort of out of
+
+49:44.880 --> 49:49.840
+ luck. So I think a constant influx of new people brings new ideas with it.
+
+49:49.840 --> 49:56.800
+ From the point of view of industrial research, because a big part of what we do is really taking
+
+49:56.800 --> 50:03.360
+ those ideas to the point where they can ship us very robust features, you end up investing a lot
+
+50:03.360 --> 50:08.640
+ in a particular idea. And if you're not careful, people can get too conservative in what they
+
+50:08.640 --> 50:15.280
+ choose to do next, knowing that the product teams will want it. And interns let you explore the more
+
+50:15.280 --> 50:22.160
+ fanciful or unproven ideas in a relatively lightweight way, ideally leading to new publications for
+
+50:22.160 --> 50:28.080
+ the intern and for the researcher. And it gives you then a portfolio from which to draw which idea
+
+50:28.080 --> 50:32.720
+ am I going to then try to take all the way through to being robust in the next year or two to ship.
+
+50:34.000 --> 50:37.520
+ So it sort of becomes part of the funnel. It's also a great way for us to
+
+50:38.160 --> 50:42.640
+ identify future full time researchers, many of our greatest researchers were former interns.
+
+50:42.640 --> 50:48.800
+ It builds a bridge to university departments so we can get to know and build an enduring relationship
+
+50:48.800 --> 50:53.840
+ with the professors and we often do academic give funds to as well as an acknowledgement of the
+
+50:53.840 --> 51:01.360
+ value the interns add and their own collaborations. So it's sort of a virtuous cycle. And then the
+
+51:01.360 --> 51:06.960
+ long term legacy of a great research lab hopefully will be not only the people who stay but the ones
+
+51:06.960 --> 51:11.520
+ who move through and then go off and carry that same model to other companies.
+
+51:11.520 --> 51:17.440
+ And so we believe strongly in industrial research and how it can complement academia and
+
+51:17.440 --> 51:21.920
+ we hope that this model will continue to propagate and be invested in by other companies,
+
+51:21.920 --> 51:26.800
+ which makes it harder for us to recruit, of course, but you know, that's the sign of success
+
+51:26.800 --> 51:30.320
+ and a rising tide lifts all ships in that sense.
+
+51:31.040 --> 51:38.080
+ And where's the idea of born with the interns? Is there brainstorming? Is there discussions
+
+51:38.080 --> 51:46.240
+ about, you know, like what the ideas come from? Yeah, as I'm asking the question, I
+
+51:46.240 --> 51:51.760
+ realized how dumb it is, but I'm hoping you have a better answer than a question I ask at the
+
+51:51.760 --> 51:59.280
+ beginning of every summer. So what will happen is we'll send out a call for interns. They'll
+
+52:00.000 --> 52:04.240
+ we'll have a number of resumes come in, people will contact the candidates, talk to them about
+
+52:04.240 --> 52:09.920
+ their interests. They'll usually try to find some somebody who has a reasonably good match to what
+
+52:09.920 --> 52:14.320
+ they're already doing, or just has a really interesting domain that they've been pursuing in
+
+52:14.320 --> 52:20.640
+ their PhD. And we think we'd love to do one of those projects too. And then the intern stays in
+
+52:20.640 --> 52:27.360
+ touch with the mentors, we call them. And then they come and in the first at the end of two weeks,
+
+52:27.360 --> 52:32.000
+ they have to decide. So they'll often have a general sense by the time they arrive.
+
+52:32.000 --> 52:36.080
+ And we'll have internal discussions about what are all the general
+
+52:36.800 --> 52:41.120
+ ideas that we're wanting to pursue to see whether two people have the same idea and maybe they
+
+52:41.120 --> 52:47.040
+ should talk and all that. But then once the intern actually arrives, sometimes the idea goes linearly
+
+52:47.040 --> 52:51.120
+ and sometimes it takes a giant left turn and we go, that sounded good. But when we thought about
+
+52:51.120 --> 52:54.880
+ it, there's this other project or it's already been done and we found this paper that we were
+
+52:54.880 --> 53:02.240
+ scooped. But we have this other great idea. So it's pretty flexible at the beginning. One of the
+
+53:02.240 --> 53:08.080
+ questions for research labs is who's deciding what to do, and then who's to blame if it goes
+
+53:08.080 --> 53:15.600
+ wrong, who gets the credit if it goes right. And so in Adobe, we push the needle very much towards
+
+53:15.600 --> 53:22.960
+ freedom of choice of projects by the researchers and the interns. But then we reward people based
+
+53:22.960 --> 53:28.000
+ on impact. So if the projects ultimately end up impacting the products and having papers and so
+
+53:28.000 --> 53:34.400
+ on. And so your alternative model just to be clear is that you have one lab director who thinks he's
+
+53:34.400 --> 53:38.720
+ a genius and tells everybody what to do, takes all the credit if it goes well, blames everybody
+
+53:38.720 --> 53:44.800
+ else if it goes badly. So we don't want that model. And this helps new ideas percolate up.
+
+53:45.440 --> 53:49.840
+ The art of running such a lab is that there are strategic priorities for the company
+
+53:49.840 --> 53:55.520
+ and there are areas where we do want to invest in pressing problems. And so it's a little bit of a
+
+53:55.520 --> 54:01.360
+ trickle down and filter up meets in the middle. And so you don't tell people you have to do X,
+
+54:01.360 --> 54:07.360
+ but you say X would be particularly appreciated this year. And then people reinterpret X through
+
+54:07.360 --> 54:12.720
+ the filter of things they want to do and they're interested in. And miraculously, it usually comes
+
+54:12.720 --> 54:18.640
+ together very well. One thing that really helps is Adobe has a really broad portfolio of products.
+
+54:18.640 --> 54:24.960
+ So if we have a good idea, there's usually a product team that is intrigued or interested.
+
+54:26.000 --> 54:31.520
+ So it means we don't have to qualify things too much ahead of time. Once in a while, the product
+
+54:31.520 --> 54:36.880
+ teams sponsor an extra intern because they have a particular problem that they really care about,
+
+54:36.880 --> 54:41.440
+ in which case it's a little bit more, we really need one of these. And then we sort of say,
+
+54:41.440 --> 54:45.920
+ great, I get an extra intern. We find an intern who thinks that's a great problem. But that's not
+
+54:45.920 --> 54:49.520
+ the typical model. That's sort of the icing on the cake as far as the budget's concerned.
+
+54:51.440 --> 54:55.920
+ And all of the above end up being important. It's really hard to predict at the beginning of the
+
+54:55.920 --> 55:01.200
+ summer, which we all have high hopes of all of the intern projects. But ultimately, some of them
+
+55:01.200 --> 55:06.480
+ pay off and some of them sort of are a nice paper, but don't turn into a feature. Others turn out
+
+55:06.480 --> 55:12.080
+ not to be as novel as we thought, but they'd be a great feature, but not a paper. And then others,
+
+55:12.080 --> 55:16.400
+ we make a little bit of progress and we realize how much we don't know. And maybe we revisit that
+
+55:16.400 --> 55:22.320
+ problem several years in a row until it finally we have a breakthrough. And then it becomes more
+
+55:22.320 --> 55:30.320
+ on track to impact a product. Jumping back to a big overall view of Adobe Research, what are you
+
+55:30.320 --> 55:37.360
+ looking forward to in 2019 and beyond? What is, you mentioned there's a giant suite of products,
+
+55:37.360 --> 55:45.920
+ giant suite of products, giant suite of ideas, new interns, a large team of researchers.
+
+55:46.960 --> 55:54.400
+ Where do you think the future holds? In terms of the technological breakthroughs?
+
+55:54.400 --> 56:00.960
+ Technological breakthroughs, especially ones that will make it into product will get to
+
+56:00.960 --> 56:05.920
+ impact the world. So I think the creative or the analytics assistance that we talked about where
+
+56:05.920 --> 56:10.320
+ they're constantly trying to figure out what you're trying to do and how can they be helpful and make
+
+56:10.320 --> 56:16.160
+ useful suggestions is a really hot topic. And it's very unpredictable as to when it'll be ready,
+
+56:16.160 --> 56:20.720
+ but I'm really looking forward to seeing how much progress we make against that. I think
+
+56:22.480 --> 56:28.640
+ some of the core technologies like generative adversarial networks are immensely promising
+
+56:28.640 --> 56:34.720
+ and seeing how quickly those become practical for mainstream use cases at high resolution with
+
+56:34.720 --> 56:39.840
+ really good quality is also exciting. And they also have this sort of strange way of even the
+
+56:39.840 --> 56:45.120
+ things they do oddly are odd in an interesting way. So it can look like dreaming or something.
+
+56:46.160 --> 56:55.040
+ So that's fascinating. I think internally we have a Sensei platform, which is a way in which
+
+56:55.040 --> 57:01.760
+ we're pooling our neural net and other intelligence models into a central platform, which can then be
+
+57:01.760 --> 57:07.040
+ leveraged by multiple product teams at once. So we're in the middle of transitioning from a,
+
+57:07.040 --> 57:11.040
+ you know, once you have a good idea, you pick a product team to work with and you sort of hand
+
+57:11.040 --> 57:17.520
+ design it for that use case to a more sort of Henry Ford, stand it up in a standard way, which
+
+57:17.520 --> 57:22.880
+ can be accessed in a standard way, which should mean that the time between a good idea and impacting
+
+57:22.880 --> 57:28.960
+ our products will be greatly shortened. And when one product has a good idea, many of the other
+
+57:28.960 --> 57:34.240
+ products can just leverage it too. So it's sort of an economy of scale. So that's more about the
+
+57:34.240 --> 57:39.680
+ how then the what, but that combination of this sort of renaissance in AI, there's a comparable
+
+57:39.680 --> 57:45.280
+ one in graphics with real time ray tracing and other really exciting emerging technologies.
+
+57:45.280 --> 57:49.920
+ And when these all come together, you'll sort of basically be dancing with light, right? Where
+
+57:49.920 --> 57:56.080
+ you'll have real time shadows, reflections, and as if it's a real world in front of you, but then
+
+57:56.080 --> 58:00.880
+ with all these magical properties brought by AI where it sort of anticipates or modifies itself
+
+58:00.880 --> 58:05.040
+ in ways that make sense based on how it understands the creative task you're trying to do.
+
+58:06.320 --> 58:12.320
+ That's a really exciting future for creative for myself too, the creator. So first of all,
+
+58:12.320 --> 58:17.760
+ I work in autonomous vehicles. I'm a roboticist. I love robots. And I think you have a fascination
+
+58:17.760 --> 58:23.920
+ with snakes, both natural and artificial robots. I share your fascination. I mean, their movement
+
+58:23.920 --> 58:31.760
+ is beautiful, adaptable. The adaptability is fascinating. There are, I looked it up, 2900
+
+58:31.760 --> 58:37.200
+ species of snakes in the world. Wow. The 175 venomous, some are tiny, some are huge.
+
+58:38.880 --> 58:44.560
+ Saw that there's one that's 25 feet in some cases. So what's the most interesting thing
+
+58:44.560 --> 58:52.000
+ that you connect with in terms of snakes, both natural and artificial? Why, what was the connection
+
+58:52.000 --> 58:58.000
+ with robotics AI in this particular form of a robot? Well, it actually came out of my work
+
+58:58.000 --> 59:02.880
+ in the 80s on computer animation, where I started doing things like cloth simulation and
+
+59:02.880 --> 59:07.360
+ other kind of soft body simulation. And you'd sort of drop it and it would bounce,
+
+59:07.360 --> 59:10.880
+ then it would just sort of stop moving. And I thought, well, what if you animate the spring
+
+59:10.880 --> 59:16.160
+ lengths and simulate muscles? And the simplest object I could do that for was an earthworm.
+
+59:16.160 --> 59:21.680
+ So I actually did a paper in 1988 on called the motion dynamics of snakes and worms. And I
+
+59:21.680 --> 59:27.760
+ read the physiology literature on both Hale snakes and worms move and then did some of the early
+
+59:27.760 --> 59:34.640
+ computer animation examples of that. So your interest in robotics started with graphics?
+
+59:34.640 --> 59:40.960
+ Came out of simulation and graphics. When I moved from Alias to Apple, we actually did a
+
+59:40.960 --> 59:46.160
+ movie called Her Majesty's Secret Serpent, which is about a secret agent snake that parachutes in
+
+59:46.160 --> 59:50.320
+ and captures a film canister from a satellite, which tells you how old fashioned we were thinking
+
+59:50.320 --> 59:57.760
+ back then, sort of classic 19 sort of 50s or 60s Bond movie kind of thing. And at the same time,
+
+59:57.760 --> 1:00:03.120
+ I'd always made radio control ships when I was a child and from scratch. And I thought, well,
+
+1:00:03.120 --> 1:00:08.800
+ how can it be to build a real one? And so then started what turned out to be like a 15 year
+
+1:00:08.800 --> 1:00:14.320
+ obsession with trying to build better snake robots. And the first one that I built just sort of
+
+1:00:14.320 --> 1:00:19.520
+ slithered sideways, but didn't actually go forward, then added wheels and building things in real
+
+1:00:19.520 --> 1:00:25.840
+ life makes you honest about the friction. The thing that appeals to me is I love creating the
+
+1:00:25.840 --> 1:00:30.800
+ illusion of life, which is what drove me to drove me to animation. And if you have a robot with
+
+1:00:30.800 --> 1:00:36.320
+ enough degrees of coordinated freedom that move in a kind of biological way, then it starts to
+
+1:00:36.320 --> 1:00:42.080
+ cross the Yankani Valley and to see me like a creature rather than a thing. And I certainly got
+
+1:00:42.080 --> 1:00:50.320
+ that with the early snakes by S3, I had it able to sidewind as well as go directly forward. My
+
+1:00:50.320 --> 1:00:54.240
+ wife to be suggested that it would be the ring bearer at our wedding. So it actually went down
+
+1:00:54.240 --> 1:00:59.360
+ the aisle carrying the rings and got in the local paper for that, which was really fun.
+
+1:01:00.160 --> 1:01:07.200
+ And this was all done as a hobby. And then I at the time that can onboard compute was incredibly
+
+1:01:07.200 --> 1:01:11.840
+ limited. It was sort of yes, you should explain that these snakes, the whole idea is that you would
+
+1:01:11.840 --> 1:01:18.640
+ you're trying to run it autonomously. Autonomously, on board right. And so
+
+1:01:19.760 --> 1:01:25.280
+ the very first one, I actually built the controller from discrete logic, because I used to do LSI,
+
+1:01:25.280 --> 1:01:30.640
+ you know, circuits and things when I was a teenager. And then the second and third one,
+
+1:01:30.640 --> 1:01:35.120
+ the 8 bit microprocessors were available with like a whole 256 bytes of RAM,
+
+1:01:36.000 --> 1:01:39.840
+ which you could just about squeeze in. So they were radio controlled rather than autonomous.
+
+1:01:39.840 --> 1:01:44.480
+ And really, we're more about the physic physicality and coordinated motion.
+
+1:01:46.560 --> 1:01:51.520
+ I've occasionally taken a side step into if only I could make it cheaply enough,
+
+1:01:51.520 --> 1:01:59.040
+ bake a great toy, which has been a lesson in how clockwork is its own magical realm that
+
+1:01:59.040 --> 1:02:03.680
+ you venture into and learn things about backlash and other things you don't take into account as
+
+1:02:03.680 --> 1:02:07.600
+ a computer scientist, which is why what seemed like a good idea doesn't work. So it's quite
+
+1:02:07.600 --> 1:02:14.160
+ humbling. And then more recently, I've been building S9, which is a much better engineered
+
+1:02:14.160 --> 1:02:17.760
+ version of S3 where the motors wore out and it doesn't work anymore. And you can't buy
+
+1:02:17.760 --> 1:02:24.640
+ replacements, which is sad given that it was such a meaningful one. S5 was about twice as long and
+
+1:02:25.760 --> 1:02:32.960
+ look much more biologically inspired. I, unlike the typical roboticist, I taper my snakes.
+
+1:02:33.520 --> 1:02:37.040
+ There are good mechanical reasons to do that, but it also makes them look more biological,
+
+1:02:37.040 --> 1:02:43.280
+ although it means every segment's unique rather than a repetition, which is why most engineers
+
+1:02:43.280 --> 1:02:49.840
+ don't do it. It actually saves weight and leverage and everything. And that one is currently on
+
+1:02:49.840 --> 1:02:54.480
+ display at the International Spy Museum in Washington, DC. None of it has done any spying.
+
+1:02:56.080 --> 1:03:00.000
+ It was on YouTube and it got its own conspiracy theory where people thought that it wasn't real
+
+1:03:00.000 --> 1:03:04.160
+ because they work at Adobe, it must be fake graphics. And people would write to me, tell me
+
+1:03:04.160 --> 1:03:11.200
+ it's real. They say the background doesn't move and it's like, it's on a tripod. So that one,
+
+1:03:11.200 --> 1:03:16.960
+ but you can see the real thing. So it really is true. And then the latest one is the first one
+
+1:03:16.960 --> 1:03:21.280
+ where I could put a Raspberry Pi, which leads to all sorts of terrible jokes about pythons and
+
+1:03:21.280 --> 1:03:29.920
+ things. But this one can have onboard compute. And then where my hobby work and my work worker
+
+1:03:29.920 --> 1:03:36.400
+ converging is you can now add vision accelerator chips, which can evaluate neural nets and do
+
+1:03:36.400 --> 1:03:41.600
+ object recognition and everything. So both for the snakes and more recently for the spider that
+
+1:03:41.600 --> 1:03:48.640
+ I've been working on, having desktop level compute is now opening up a whole world of
+
+1:03:49.200 --> 1:03:54.880
+ true autonomy with onboard compute, onboard batteries, and still having that sort of
+
+1:03:54.880 --> 1:04:01.680
+ biomimetic quality that appeals to children in particular. They are really drawn to them and
+
+1:04:01.680 --> 1:04:08.960
+ adults think they look creepy, but children actually think they look charming. And I gave a
+
+1:04:08.960 --> 1:04:14.880
+ series of lectures at Girls Who Code to encourage people to take an interest in technology. And
+
+1:04:14.880 --> 1:04:19.200
+ at the moment, I'd say they're still more expensive than the value that they add,
+
+1:04:19.200 --> 1:04:22.560
+ which is why they're a great hobby for me, but they're not really a great product.
+
+1:04:22.560 --> 1:04:30.000
+ It makes me think about doing that very early thing I did at Alias with changing the muscle
+
+1:04:30.000 --> 1:04:35.760
+ rest lengths. If I could do that with a real artificial muscle material, then the next snake
+
+1:04:35.760 --> 1:04:40.960
+ ideally would use that rather than motors and gearboxes and everything. It would be lighter,
+
+1:04:40.960 --> 1:04:49.200
+ much stronger, and more continuous and smooth. So I like to say being in research as a license
+
+1:04:49.200 --> 1:04:54.640
+ to be curious, and I have the same feeling with my hobby yet. It forced me to read biology and
+
+1:04:54.640 --> 1:04:59.680
+ be curious about things that otherwise would have just been natural geographic special.
+
+1:04:59.680 --> 1:05:04.320
+ Suddenly, I'm thinking, how does that snake move? Can I copy it? I look at the trails that
+
+1:05:04.320 --> 1:05:08.640
+ side winding snakes leave in sand and see if my snake robots would do the same thing.
+
+1:05:10.240 --> 1:05:13.840
+ Out of something inanimate, I like why you put a try to bring life into it and beauty.
+
+1:05:13.840 --> 1:05:18.240
+ Absolutely. And then ultimately, give it a personality, which is where the intelligent
+
+1:05:18.240 --> 1:05:25.040
+ agent research will converge with the vision and voice synthesis to give it a sense of having
+
+1:05:25.040 --> 1:05:29.840
+ not necessarily human level intelligence. I think the Turing test is such a high bar, it's
+
+1:05:30.480 --> 1:05:36.080
+ a little bit self defeating, but having one that you can have a meaningful conversation with,
+
+1:05:36.880 --> 1:05:43.040
+ especially if you have a reasonably good sense of what you can say. So not trying to have it so
+
+1:05:43.040 --> 1:05:50.080
+ a stranger could walk up and have one, but so as a pet owner or a robot pet owner, you could
+
+1:05:50.080 --> 1:05:53.200
+ know what it thinks about and what it can reason about.
+
+1:05:53.200 --> 1:05:58.800
+ Or sometimes just meaningful interaction. If you have the kind of interaction you have with a dog,
+
+1:05:58.800 --> 1:06:02.080
+ sometimes you might have a conversation, but it's usually one way.
+
+1:06:02.080 --> 1:06:02.640
+ Absolutely.
+
+1:06:02.640 --> 1:06:06.000
+ And nevertheless, it feels like a meaningful connection.
+
+1:06:06.800 --> 1:06:12.720
+ And one of the things that I'm trying to do in the sample audio that will play you is beginning
+
+1:06:12.720 --> 1:06:17.680
+ to get towards the point where the reasoning system can explain why it knows something or
+
+1:06:17.680 --> 1:06:22.320
+ why it thinks something. And that again, creates the sense that it really does know what it's
+
+1:06:22.320 --> 1:06:29.840
+ talking about, but also for debugging. As you get more and more elaborate behavior, it's like,
+
+1:06:29.840 --> 1:06:37.120
+ why did you decide to do that? How do you know that? I think the robot's really
+
+1:06:37.120 --> 1:06:42.560
+ my muse for helping me think about the future of AI and what to invent next.
+
+1:06:42.560 --> 1:06:47.040
+ So even at Adobe, that's mostly operating in the digital world.
+
+1:06:47.040 --> 1:06:47.440
+ Correct.
+
+1:06:47.440 --> 1:06:54.480
+ Do you ever, do you see a future where Adobe even expands into the more physical world perhaps?
+
+1:06:54.480 --> 1:07:02.080
+ So bringing life not into animations, but bringing life into physical objects with whether it's,
+
+1:07:02.080 --> 1:07:08.640
+ well, I'd have to say at the moment it's a twinkle in my eye. I think the more likely thing is that
+
+1:07:08.640 --> 1:07:14.240
+ we will bring virtual objects into the physical world through augmented reality.
+
+1:07:14.240 --> 1:07:21.360
+ And many of the ideas that might take five years to build a robot to do, you can do in a few weeks
+
+1:07:21.360 --> 1:07:29.120
+ with digital assets. So I think when really intelligent robots finally become commonplace,
+
+1:07:29.120 --> 1:07:34.000
+ they won't be that surprising because we'll have been living with those personalities in the virtual
+
+1:07:34.000 --> 1:07:38.640
+ sphere for a long time. And then they'll just say, oh, it's Siri with legs or Alexa,
+
+1:07:39.520 --> 1:07:47.760
+ Alexa on hooves or something. So I can see that welcoming. And for now, it's still an adventure
+
+1:07:47.760 --> 1:07:53.760
+ and we don't know quite what the experience will be like. And it's really exciting to sort of see
+
+1:07:53.760 --> 1:07:59.680
+ all of these different strands of my career converge. Yeah, in interesting ways. And it is
+
+1:07:59.680 --> 1:08:07.920
+ definitely a fun adventure. So let me end with my favorite poem, the last few lines of my favorite
+
+1:08:07.920 --> 1:08:14.560
+ poem of yours that ponders mortality. And in some sense, immortality, as our ideas live through
+
+1:08:14.560 --> 1:08:21.120
+ the ideas of others through the work of others, it ends with, do not weep or mourn. It was enough
+
+1:08:21.120 --> 1:08:28.000
+ the little atomies permitted just a single dance. Scatter them as deep as your eyes can see. I'm
+
+1:08:28.000 --> 1:08:34.320
+ content they'll have another chance, sweeping more centered parts along to join a jostling,
+
+1:08:34.320 --> 1:08:41.440
+ lifting throng as others danced in me. Beautiful poem. Beautiful way to end it. Gavin, thank you
+
+1:08:41.440 --> 1:08:45.920
+ so much for talking today. And thank you for inspiring and empowering millions of people
+
+1:08:45.920 --> 1:08:50.960
+ like myself for creating amazing stuff. Oh, thank you. It's a great conversation.
+

vtt/episode_024_small.vtt

@@ -0,0 +1,2675 @@
+WEBVTT
+
+00:00.000 --> 00:02.880
+ The following is a conversation with Rosalind Picard.
+
+00:02.880 --> 00:04.560
+ She's a professor at MIT,
+
+00:04.560 --> 00:06.880
+ director of the Effective Computing Research Group
+
+00:06.880 --> 00:08.360
+ at the MIT Media Lab,
+
+00:08.360 --> 00:12.440
+ and cofounder of two companies, Effectiva and Empatica.
+
+00:12.440 --> 00:13.560
+ Over two decades ago,
+
+00:13.560 --> 00:15.440
+ she launched a field of effective computing
+
+00:15.440 --> 00:17.560
+ with her book of the same name.
+
+00:17.560 --> 00:20.040
+ This book described the importance of emotion
+
+00:20.040 --> 00:23.040
+ in artificial and natural intelligence.
+
+00:23.040 --> 00:25.320
+ The vital role of emotional communication
+
+00:25.320 --> 00:28.520
+ has to the relationship between people in general
+
+00:28.520 --> 00:30.920
+ and human robot interaction.
+
+00:30.920 --> 00:34.040
+ I really enjoy talking with Ros over so many topics,
+
+00:34.040 --> 00:36.440
+ including emotion, ethics, privacy,
+
+00:36.440 --> 00:39.720
+ wearable computing and her recent research in epilepsy
+
+00:39.720 --> 00:42.600
+ and even love and meaning.
+
+00:42.600 --> 00:44.000
+ This conversation is part
+
+00:44.000 --> 00:46.040
+ of the Artificial Intelligence Podcast.
+
+00:46.040 --> 00:48.200
+ If you enjoy it, subscribe on YouTube,
+
+00:48.200 --> 00:50.760
+ iTunes or simply connect with me on Twitter
+
+00:50.760 --> 00:54.000
+ at Lex Freedman, spelled F R I D.
+
+00:54.000 --> 00:58.000
+ And now here's my conversation with Rosalind Picard.
+
+00:59.520 --> 01:00.760
+ More than 20 years ago,
+
+01:00.760 --> 01:03.360
+ you've coined the term Effective Computing
+
+01:03.360 --> 01:05.400
+ and led a lot of research in this area.
+
+01:05.400 --> 01:08.800
+ Since then, as I understand the goal is to make the machine
+
+01:08.800 --> 01:12.400
+ detect and interpret the emotional state of a human being
+
+01:12.400 --> 01:14.240
+ and adopt the behavior of the machine
+
+01:14.240 --> 01:16.160
+ based on the emotional state.
+
+01:16.160 --> 01:19.680
+ So how is your understanding of the problems
+
+01:19.680 --> 01:24.680
+ based defined by effective computing changed in the past 24 years?
+
+01:25.400 --> 01:28.960
+ So it's the scope, the applications, the challenges,
+
+01:28.960 --> 01:32.160
+ what's involved, how has that evolved over the years?
+
+01:32.160 --> 01:35.400
+ Yeah, actually, originally when I defined the term
+
+01:35.400 --> 01:38.080
+ Effective Computing, it was a bit broader
+
+01:38.080 --> 01:41.320
+ than just recognizing and responding intelligently
+
+01:41.320 --> 01:44.560
+ to human emotion, although those are probably the two pieces
+
+01:44.560 --> 01:47.160
+ that we've worked on the hardest.
+
+01:47.160 --> 01:50.680
+ The original concept also encompassed machines
+
+01:50.680 --> 01:53.520
+ that would have mechanisms that functioned
+
+01:53.520 --> 01:55.680
+ like human emotion does inside them.
+
+01:55.680 --> 01:59.000
+ It would be any computing that relates to arises from
+
+01:59.000 --> 02:01.600
+ or deliberately influences human emotion.
+
+02:02.600 --> 02:05.160
+ So the human computer interaction part
+
+02:05.160 --> 02:07.880
+ is the part that people tend to see.
+
+02:07.880 --> 02:11.000
+ Like if I'm really ticked off at my computer
+
+02:11.000 --> 02:13.480
+ and I'm scowling at it and I'm cursing at it
+
+02:13.480 --> 02:15.720
+ and it just keeps acting smiling and happy
+
+02:15.720 --> 02:17.480
+ like that little paperclip used to do,
+
+02:17.480 --> 02:22.160
+ like dancing, winking, that kind of thing
+
+02:22.160 --> 02:24.640
+ just makes you even more frustrated.
+
+02:24.640 --> 02:29.080
+ And I thought that stupid thing needs to see my affect
+
+02:29.080 --> 02:30.600
+ and if it's gonna be intelligent,
+
+02:30.600 --> 02:33.000
+ which Microsoft researchers had worked really hard on,
+
+02:33.000 --> 02:35.120
+ it actually had some of the most sophisticated AI in it
+
+02:35.120 --> 02:37.960
+ at the time, that thing's gonna actually be smart.
+
+02:37.960 --> 02:41.560
+ It needs to respond to me and you
+
+02:41.560 --> 02:45.320
+ and we can send it very different signals.
+
+02:45.320 --> 02:47.160
+ So by the way, just a quick interruption.
+
+02:47.160 --> 02:52.160
+ The Clippy, maybe it's in Word 95 and 98,
+
+02:52.360 --> 02:54.360
+ I don't remember when it was born,
+
+02:54.360 --> 02:58.280
+ but many people, do you find yourself with that reference
+
+02:58.280 --> 03:00.720
+ that people recognize what you're talking about still
+
+03:00.720 --> 03:01.640
+ to this point?
+
+03:01.640 --> 03:05.160
+ I don't expect the newest students to these days,
+
+03:05.160 --> 03:07.160
+ but I've mentioned it to a lot of audiences
+
+03:07.160 --> 03:09.200
+ like how many of you know this Clippy thing
+
+03:09.200 --> 03:11.680
+ and still the majority of people seem to know it.
+
+03:11.680 --> 03:15.280
+ So Clippy kind of looks at maybe natural language processing
+
+03:15.280 --> 03:19.200
+ where you are typing and tries to help you complete, I think.
+
+03:19.200 --> 03:22.440
+ I don't even remember what Clippy was except annoying.
+
+03:22.440 --> 03:25.760
+ Yeah, some people actually liked it.
+
+03:25.760 --> 03:27.480
+ I would hear those stories.
+
+03:27.480 --> 03:28.400
+ You miss it?
+
+03:28.400 --> 03:31.240
+ Well, I miss the annoyance.
+
+03:31.240 --> 03:35.680
+ They felt like there's an element, somebody was there
+
+03:35.680 --> 03:37.560
+ and we're in it together and they were annoying.
+
+03:37.560 --> 03:40.800
+ It's like a puppy that just doesn't get it.
+
+03:40.800 --> 03:42.120
+ They keep stripping up the couch kind of thing.
+
+03:42.120 --> 03:44.920
+ And in fact, they could have done it smarter like a puppy.
+
+03:44.920 --> 03:47.960
+ If they had done, like if when you yelled at it
+
+03:47.960 --> 03:50.760
+ or cursed at it, if it had put its little ears back
+
+03:50.760 --> 03:52.880
+ and its tail down and shirked off,
+
+03:52.880 --> 03:55.840
+ probably people would have wanted it back, right?
+
+03:55.840 --> 03:58.560
+ But instead when you yelled at it, what did it do?
+
+03:58.560 --> 04:01.200
+ It smiled, it winked, it danced, right?
+
+04:01.200 --> 04:03.160
+ If somebody comes to my office and I yell at them,
+
+04:03.160 --> 04:04.720
+ they started smiling, winking and dancing.
+
+04:04.720 --> 04:06.720
+ I'm like, I never want to see you again.
+
+04:06.720 --> 04:08.480
+ So Bill Gates got a standing ovation
+
+04:08.480 --> 04:10.080
+ when he said it was going away
+
+04:10.080 --> 04:12.320
+ because people were so ticked.
+
+04:12.320 --> 04:14.960
+ It was so emotionally unintelligent, right?
+
+04:14.960 --> 04:18.080
+ It was intelligent about whether you were writing a letter
+
+04:18.080 --> 04:20.800
+ or what kind of help you needed for that context.
+
+04:20.800 --> 04:23.360
+ It was completely unintelligent about,
+
+04:23.360 --> 04:25.720
+ hey, if you're annoying your customer,
+
+04:25.720 --> 04:28.320
+ don't smile in their face when you do it.
+
+04:28.320 --> 04:32.280
+ So that kind of mismatch was something
+
+04:32.280 --> 04:35.000
+ the developers just didn't think about.
+
+04:35.000 --> 04:39.440
+ And intelligence at the time was really all about math
+
+04:39.440 --> 04:44.920
+ and language and chess and games,
+
+04:44.920 --> 04:47.880
+ problems that could be pretty well defined.
+
+04:47.880 --> 04:51.440
+ Social emotional interaction is much more complex than chess
+
+04:51.440 --> 04:55.960
+ or Go or any of the games that people are trying to solve.
+
+04:55.960 --> 04:58.640
+ And in order to understand that required skills
+
+04:58.640 --> 05:00.240
+ that most people in computer science
+
+05:00.240 --> 05:02.520
+ actually were lacking personally.
+
+05:02.520 --> 05:03.760
+ Well, let's talk about computer science
+
+05:03.760 --> 05:06.320
+ if things gotten better since the work,
+
+05:06.320 --> 05:09.480
+ since the message, since you've really launched a field
+
+05:09.480 --> 05:11.280
+ with a lot of research work in the space,
+
+05:11.280 --> 05:14.080
+ I still find as a person like yourself
+
+05:14.080 --> 05:16.640
+ who's deeply passionate about human beings
+
+05:16.640 --> 05:18.840
+ and yet in computer science,
+
+05:18.840 --> 05:23.040
+ there still seems to be a lack of, sorry to say,
+
+05:23.040 --> 05:26.760
+ empathy in us computer scientists.
+
+05:26.760 --> 05:28.800
+ Yeah, well, or hasn't gotten better.
+
+05:28.800 --> 05:30.680
+ Let's just say there's a lot more variety
+
+05:30.680 --> 05:32.320
+ among computer scientists these days.
+
+05:32.320 --> 05:34.960
+ Computer scientists are a much more diverse group today
+
+05:34.960 --> 05:37.600
+ than they were 25 years ago.
+
+05:37.600 --> 05:39.000
+ And that's good.
+
+05:39.000 --> 05:41.760
+ We need all kinds of people to become computer scientists
+
+05:41.760 --> 05:45.600
+ so that computer science reflects more what society needs.
+
+05:45.600 --> 05:49.080
+ And there's brilliance among every personality type.
+
+05:49.080 --> 05:52.000
+ So it need not be limited to people
+
+05:52.000 --> 05:54.120
+ who prefer computers to other people.
+
+05:54.120 --> 05:55.400
+ How hard do you think it is?
+
+05:55.400 --> 05:58.600
+ How your view of how difficult it is to recognize emotion
+
+05:58.600 --> 06:03.960
+ or to create a deeply emotional intelligent interaction
+
+06:03.960 --> 06:07.440
+ has it gotten easier or harder as you've explored it further?
+
+06:07.440 --> 06:10.040
+ And how far away are we from cracking this?
+
+06:12.440 --> 06:16.040
+ If you think of the Turing test solving the intelligence,
+
+06:16.040 --> 06:18.760
+ looking at the Turing test for emotional intelligence?
+
+06:20.720 --> 06:25.600
+ I think it is as difficult as I thought it was gonna be.
+
+06:25.600 --> 06:29.280
+ I think my prediction of its difficulty is spot on.
+
+06:29.280 --> 06:33.160
+ I think the time estimates are always hard
+
+06:33.160 --> 06:37.360
+ because they're always a function of society's love
+
+06:37.360 --> 06:39.560
+ and hate of a particular topic.
+
+06:39.560 --> 06:44.560
+ If society gets excited and you get thousands
+
+06:44.560 --> 06:49.080
+ of researchers working on it for a certain application,
+
+06:49.080 --> 06:52.080
+ that application gets solved really quickly.
+
+06:52.080 --> 06:54.360
+ The general intelligence,
+
+06:54.360 --> 06:58.200
+ the computer's complete lack of ability
+
+06:58.200 --> 07:03.200
+ to have awareness of what it's doing,
+
+07:03.560 --> 07:05.560
+ the fact that it's not conscious,
+
+07:05.560 --> 07:08.640
+ the fact that there's no signs of it becoming conscious,
+
+07:08.640 --> 07:11.880
+ the fact that it doesn't read between the lines,
+
+07:11.880 --> 07:15.080
+ those kinds of things that we have to teach it explicitly,
+
+07:15.080 --> 07:17.520
+ what other people pick up implicitly.
+
+07:17.520 --> 07:20.440
+ We don't see that changing yet.
+
+07:20.440 --> 07:23.600
+ There aren't breakthroughs yet that lead us to believe
+
+07:23.600 --> 07:25.360
+ that that's gonna go any faster,
+
+07:25.360 --> 07:28.680
+ which means that it's still gonna be kind of stuck
+
+07:28.680 --> 07:31.320
+ with a lot of limitations
+
+07:31.320 --> 07:34.080
+ where it's probably only gonna do the right thing
+
+07:34.080 --> 07:37.200
+ in very limited, narrow, prespecified contexts
+
+07:37.200 --> 07:40.960
+ where we can prescribe pretty much
+
+07:40.960 --> 07:42.880
+ what's gonna happen there.
+
+07:42.880 --> 07:44.920
+ So I don't see the,
+
+07:47.000 --> 07:48.040
+ it's hard to predict the date
+
+07:48.040 --> 07:51.800
+ because when people don't work on it, it's infinite.
+
+07:51.800 --> 07:55.240
+ When everybody works on it, you get a nice piece of it
+
+07:55.240 --> 07:58.600
+ well solved in a short amount of time.
+
+07:58.600 --> 08:01.560
+ I actually think there's a more important issue right now
+
+08:01.560 --> 08:04.560
+ than the difficulty of it.
+
+08:04.560 --> 08:07.440
+ And that's causing some of us to put the brakes on a little bit.
+
+08:07.440 --> 08:09.360
+ Usually we're all just like step on the gas,
+
+08:09.360 --> 08:11.160
+ let's go faster.
+
+08:11.160 --> 08:14.200
+ This is causing us to pull back and put the brakes on.
+
+08:14.200 --> 08:18.680
+ And that's the way that some of this technology
+
+08:18.680 --> 08:21.200
+ is being used in places like China right now.
+
+08:21.200 --> 08:24.520
+ And that worries me so deeply
+
+08:24.520 --> 08:27.800
+ that it's causing me to pull back myself
+
+08:27.800 --> 08:30.080
+ on a lot of the things that we could be doing
+
+08:30.080 --> 08:33.680
+ and try to get the community to think a little bit more
+
+08:33.680 --> 08:36.040
+ about, okay, if we're gonna go forward with that,
+
+08:36.040 --> 08:39.320
+ how can we do it in a way that puts in place safeguards
+
+08:39.320 --> 08:41.120
+ that protects people?
+
+08:41.120 --> 08:43.560
+ So the technology we're referring to is
+
+08:43.560 --> 08:46.400
+ just when a computer senses the human being,
+
+08:46.400 --> 08:47.720
+ like the human face.
+
+08:47.720 --> 08:48.560
+ Yeah, yeah.
+
+08:48.560 --> 08:51.840
+ And so there's a lot of exciting things there
+
+08:51.840 --> 08:53.960
+ like forming a deep connection with the human being.
+
+08:53.960 --> 08:55.440
+ So what are your worries?
+
+08:55.440 --> 08:56.880
+ How that could go wrong?
+
+08:58.000 --> 08:59.520
+ Is it in terms of privacy?
+
+08:59.520 --> 09:03.000
+ Is it in terms of other kinds of more subtle things?
+
+09:03.000 --> 09:04.320
+ Let's dig into privacy.
+
+09:04.320 --> 09:07.800
+ So here in the US, if I'm watching a video
+
+09:07.800 --> 09:11.920
+ of say a political leader and in the US
+
+09:11.920 --> 09:13.640
+ we're quite free as we all know
+
+09:13.640 --> 09:17.920
+ to even criticize the president of the United States, right?
+
+09:17.920 --> 09:19.440
+ Here that's not a shocking thing.
+
+09:19.440 --> 09:22.720
+ It happens about every five seconds, right?
+
+09:22.720 --> 09:27.720
+ But in China, what happens if you criticize
+
+09:27.880 --> 09:31.080
+ the leader of the government, right?
+
+09:31.080 --> 09:34.320
+ And so people are very careful not to do that.
+
+09:34.320 --> 09:37.880
+ However, what happens if you're simply watching a video
+
+09:37.880 --> 09:41.000
+ and you make a facial expression
+
+09:41.000 --> 09:45.280
+ that shows a little bit of skepticism, right?
+
+09:45.280 --> 09:48.200
+ Well, and here we're completely free to do that.
+
+09:48.200 --> 09:50.680
+ In fact, we're free to fly off the handle
+
+09:50.680 --> 09:54.720
+ and say anything we want, usually.
+
+09:54.720 --> 09:56.560
+ I mean, there are some restrictions
+
+09:56.560 --> 10:00.840
+ when the athlete does this as part of the national broadcast
+
+10:00.840 --> 10:03.840
+ maybe the teams get a little unhappy
+
+10:03.840 --> 10:05.880
+ about picking that forum to do it, right?
+
+10:05.880 --> 10:08.720
+ But that's more a question of judgment.
+
+10:08.720 --> 10:12.760
+ We have these freedoms and in places
+
+10:12.760 --> 10:14.160
+ that don't have those freedoms,
+
+10:14.160 --> 10:17.080
+ what if our technology can read
+
+10:17.080 --> 10:19.600
+ your underlying affective state?
+
+10:19.600 --> 10:22.440
+ What if our technology can read it even noncontact?
+
+10:22.440 --> 10:24.440
+ What if our technology can read it
+
+10:24.440 --> 10:28.840
+ without your prior consent?
+
+10:28.840 --> 10:31.400
+ And here in the US and my first company,
+
+10:31.400 --> 10:32.960
+ we started Affectiva.
+
+10:32.960 --> 10:35.600
+ We have worked super hard to turn away money
+
+10:35.600 --> 10:38.440
+ and opportunities that try to read people's affect
+
+10:38.440 --> 10:41.360
+ without their prior informed consent.
+
+10:41.360 --> 10:45.160
+ And even the software that is licensible,
+
+10:45.160 --> 10:47.680
+ you have to sign things saying you will only use it
+
+10:47.680 --> 10:51.800
+ in certain ways, which essentially is get people's buy in,
+
+10:51.800 --> 10:52.640
+ right?
+
+10:52.640 --> 10:55.400
+ Don't do this without people agreeing to it.
+
+10:56.840 --> 10:58.640
+ There are other countries where they're not interested
+
+10:58.640 --> 10:59.560
+ in people's buy in.
+
+10:59.560 --> 11:01.440
+ They're just gonna use it.
+
+11:01.440 --> 11:03.080
+ They're gonna inflict it on you.
+
+11:03.080 --> 11:04.480
+ And if you don't like it,
+
+11:04.480 --> 11:08.480
+ you better not scowl in the direction of any sensors.
+
+11:08.480 --> 11:11.440
+ So one, let me just comment on a small tangent.
+
+11:11.440 --> 11:16.000
+ Do you know what the idea of adversarial examples
+
+11:16.000 --> 11:20.880
+ and deep fakes and so on, what you bring up is actually,
+
+11:20.880 --> 11:25.800
+ in one sense, deep fakes provide a comforting protection
+
+11:25.800 --> 11:30.640
+ that you can no longer really trust
+
+11:30.640 --> 11:34.560
+ that the video of your face was legitimate.
+
+11:34.560 --> 11:37.080
+ And therefore you always have an escape clause
+
+11:37.080 --> 11:38.480
+ if a government is trying,
+
+11:38.480 --> 11:43.480
+ if a stable balanced ethical government
+
+11:43.480 --> 11:46.200
+ is trying to accuse you of something,
+
+11:46.200 --> 11:47.040
+ at least you have protection.
+
+11:47.040 --> 11:48.720
+ You can say it was fake news,
+
+11:48.720 --> 11:50.560
+ as is a popular term now.
+
+11:50.560 --> 11:52.320
+ Yeah, that's the general thinking of it.
+
+11:52.320 --> 11:54.320
+ We know how to go into the video
+
+11:54.320 --> 11:58.320
+ and see, for example, your heart rate and respiration
+
+11:58.320 --> 12:02.160
+ and whether or not they've been tampered with.
+
+12:02.160 --> 12:05.480
+ And we also can put fake heart rate and respiration
+
+12:05.480 --> 12:06.320
+ in your video.
+
+12:06.320 --> 12:09.120
+ Now too, we decided we needed to do that
+
+12:09.120 --> 12:12.600
+ after we developed a way to extract it,
+
+12:12.600 --> 12:15.960
+ but we decided we also needed a way to jam it.
+
+12:15.960 --> 12:18.320
+ And so the fact that we took time
+
+12:18.320 --> 12:20.920
+ to do that other step too,
+
+12:20.920 --> 12:22.560
+ that was time that I wasn't spending
+
+12:22.560 --> 12:25.280
+ making the machine more effectively intelligent.
+
+12:25.280 --> 12:28.520
+ And there's a choice in how we spend our time,
+
+12:28.520 --> 12:32.440
+ which is now being swayed a little bit less by this goal
+
+12:32.440 --> 12:34.360
+ and a little bit more like by concern
+
+12:34.360 --> 12:36.600
+ about what's happening in society
+
+12:36.600 --> 12:38.880
+ and what kind of future do we wanna build.
+
+12:38.880 --> 12:41.680
+ And as we step back and say,
+
+12:41.680 --> 12:44.600
+ okay, we don't just build AI to build AI
+
+12:44.600 --> 12:46.520
+ to make Elon Musk more money
+
+12:46.520 --> 12:48.760
+ or to make Amazon Jeff Bezos more money.
+
+12:48.760 --> 12:52.920
+ You could gosh, that's the wrong ethic.
+
+12:52.920 --> 12:54.160
+ Why are we building it?
+
+12:54.160 --> 12:57.240
+ What is the point of building AI?
+
+12:57.240 --> 13:01.560
+ It used to be, it was driven by researchers in academia
+
+13:01.560 --> 13:04.200
+ to get papers published and to make a career for themselves
+
+13:04.200 --> 13:06.080
+ and to do something cool, right?
+
+13:06.080 --> 13:07.680
+ Cause maybe it could be done.
+
+13:07.680 --> 13:12.480
+ Now we realize that this is enabling rich people
+
+13:12.480 --> 13:14.280
+ to get vastly richer.
+
+13:15.560 --> 13:19.800
+ The poor are, the divide is even larger.
+
+13:19.800 --> 13:22.880
+ And is that the kind of future that we want?
+
+13:22.880 --> 13:25.920
+ Maybe we wanna think about, maybe we wanna rethink AI.
+
+13:25.920 --> 13:29.120
+ Maybe we wanna rethink the problems in society
+
+13:29.120 --> 13:32.760
+ that are causing the greatest inequity
+
+13:32.760 --> 13:35.040
+ and rethink how to build AI
+
+13:35.040 --> 13:36.760
+ that's not about a general intelligence,
+
+13:36.760 --> 13:39.320
+ but that's about extending the intelligence
+
+13:39.320 --> 13:41.240
+ and capability of the have nots
+
+13:41.240 --> 13:43.800
+ so that we close these gaps in society.
+
+13:43.800 --> 13:46.640
+ Do you hope that kind of stepping on the break
+
+13:46.640 --> 13:48.000
+ happens organically?
+
+13:48.000 --> 13:51.240
+ Because I think still majority of the force behind AI
+
+13:51.240 --> 13:52.800
+ is the desire to publish papers,
+
+13:52.800 --> 13:55.800
+ is to make money without thinking about the why.
+
+13:55.800 --> 13:57.280
+ Do you hope it happens organically?
+
+13:57.280 --> 13:59.040
+ Is there a room for regulation?
+
+14:01.120 --> 14:02.960
+ Yeah, yeah, yeah, great questions.
+
+14:02.960 --> 14:07.360
+ I prefer the, they talk about the carrot versus the stick.
+
+14:07.360 --> 14:09.160
+ I definitely prefer the carrot to the stick.
+
+14:09.160 --> 14:14.160
+ And in our free world, there's only so much stick, right?
+
+14:14.920 --> 14:17.280
+ You're gonna find a way around it.
+
+14:17.280 --> 14:21.160
+ I generally think less regulation is better.
+
+14:21.160 --> 14:24.440
+ That said, even though my position is classically carrot,
+
+14:24.440 --> 14:26.280
+ no stick, no regulation,
+
+14:26.280 --> 14:29.080
+ I think we do need some regulations in this space.
+
+14:29.080 --> 14:30.720
+ I do think we need regulations
+
+14:30.720 --> 14:33.640
+ around protecting people with their data,
+
+14:33.640 --> 14:38.200
+ that you own your data, not Amazon, not Google.
+
+14:38.200 --> 14:40.800
+ I would like to see people own their own data.
+
+14:40.800 --> 14:42.520
+ I would also like to see the regulations
+
+14:42.520 --> 14:44.520
+ that we have right now around lie detection
+
+14:44.520 --> 14:48.200
+ being extended to emotion recognition in general.
+
+14:48.200 --> 14:51.040
+ That right now you can't use a lie detector on an employee
+
+14:51.040 --> 14:52.760
+ when you're on a candidate,
+
+14:52.760 --> 14:54.720
+ when you're interviewing them for a job.
+
+14:54.720 --> 14:57.800
+ I think similarly, we need to put in place protection
+
+14:57.800 --> 15:00.720
+ around reading people's emotions without their consent
+
+15:00.720 --> 15:02.200
+ and in certain cases,
+
+15:02.200 --> 15:06.160
+ like characterizing them for a job and other opportunities.
+
+15:06.160 --> 15:09.200
+ So I also think that when we're reading emotion
+
+15:09.200 --> 15:11.760
+ that's predictive around mental health,
+
+15:11.760 --> 15:14.240
+ that that should, even though it's not medical data,
+
+15:14.240 --> 15:16.160
+ that that should get the kinds of protections
+
+15:16.160 --> 15:18.520
+ that our medical data gets.
+
+15:18.520 --> 15:20.080
+ What most people don't know yet
+
+15:20.080 --> 15:22.640
+ is right now with your smartphone use,
+
+15:22.640 --> 15:25.280
+ and if you're wearing a sensor
+
+15:25.280 --> 15:27.760
+ and you wanna learn about your stress and your sleep,
+
+15:27.760 --> 15:29.040
+ and your physical activity,
+
+15:29.040 --> 15:30.840
+ and how much you're using your phone
+
+15:30.840 --> 15:32.640
+ and your social interaction,
+
+15:32.640 --> 15:34.960
+ all of that non medical data,
+
+15:34.960 --> 15:37.960
+ when we put it together with machine learning,
+
+15:37.960 --> 15:40.160
+ now called AI, even though the founders of AI
+
+15:40.160 --> 15:41.720
+ wouldn't have called it that,
+
+15:42.960 --> 15:47.960
+ that capability can not only tell that you're calm right now,
+
+15:48.440 --> 15:50.840
+ or that you're getting a little stressed,
+
+15:50.840 --> 15:53.920
+ but it can also predict how you're likely to be tomorrow.
+
+15:53.920 --> 15:55.880
+ If you're likely to be sick or healthy,
+
+15:55.880 --> 15:58.720
+ happy or sad, stressed or calm.
+
+15:58.720 --> 16:00.640
+ Especially when you're tracking data over time.
+
+16:00.640 --> 16:03.760
+ Especially when we're tracking a week of your data or more.
+
+16:03.760 --> 16:06.000
+ Do you have an optimism towards,
+
+16:06.000 --> 16:07.800
+ that a lot of people on our phones
+
+16:07.800 --> 16:10.360
+ are worried about this camera that's looking at us?
+
+16:10.360 --> 16:12.520
+ For the most part, on balance,
+
+16:12.520 --> 16:16.120
+ are you optimistic about the benefits
+
+16:16.120 --> 16:17.480
+ that can be brought from that camera
+
+16:17.480 --> 16:19.640
+ that's looking at billions of us,
+
+16:19.640 --> 16:22.080
+ or should we be more worried?
+
+16:22.080 --> 16:27.080
+ I think we should be a little bit more worried
+
+16:28.840 --> 16:32.480
+ about who's looking at us and listening to us.
+
+16:32.480 --> 16:36.680
+ The device sitting on your countertop in your kitchen,
+
+16:36.680 --> 16:41.680
+ whether it's Alexa or Google Home or Apple, Siri,
+
+16:42.120 --> 16:46.440
+ these devices want to listen,
+
+16:47.480 --> 16:49.640
+ while they say ostensibly to help us.
+
+16:49.640 --> 16:52.080
+ And I think there are great people in these companies
+
+16:52.080 --> 16:54.160
+ who do want to help people.
+
+16:54.160 --> 16:56.160
+ Let me not brand them all bad.
+
+16:56.160 --> 16:59.320
+ I'm a user of products from all of these companies.
+
+16:59.320 --> 17:04.320
+ I'm naming all the A companies, Alphabet, Apple, Amazon.
+
+17:04.360 --> 17:09.120
+ They are awfully big companies, right?
+
+17:09.120 --> 17:11.520
+ They have incredible power.
+
+17:11.520 --> 17:16.520
+ And what if China were to buy them, right?
+
+17:16.520 --> 17:19.320
+ And suddenly, all of that data
+
+17:19.320 --> 17:21.880
+ were not part of free America,
+
+17:21.880 --> 17:23.800
+ but all of that data were part of somebody
+
+17:23.800 --> 17:26.040
+ who just wants to take over the world
+
+17:26.040 --> 17:27.480
+ and you submit to them.
+
+17:27.480 --> 17:31.560
+ And guess what happens if you so much as smirk the wrong way
+
+17:31.560 --> 17:34.000
+ when they say something that you don't like?
+
+17:34.000 --> 17:36.840
+ Well, they have reeducation camps, right?
+
+17:36.840 --> 17:38.360
+ That's a nice word for them.
+
+17:38.360 --> 17:40.800
+ By the way, they have a surplus of organs
+
+17:40.800 --> 17:42.720
+ for people who have surgery these days.
+
+17:42.720 --> 17:44.440
+ They don't have an organ donation problem,
+
+17:44.440 --> 17:45.640
+ because they take your blood.
+
+17:45.640 --> 17:47.480
+ And they know you're a match.
+
+17:47.480 --> 17:51.640
+ And the doctors are on record of taking organs from people
+
+17:51.640 --> 17:54.680
+ who are perfectly healthy and not prisoners.
+
+17:54.680 --> 17:57.960
+ They're just simply not the favored ones of the government.
+
+17:58.960 --> 18:03.960
+ And, you know, that's a pretty freaky evil society.
+
+18:03.960 --> 18:05.920
+ And we can use the word evil there.
+
+18:05.920 --> 18:07.280
+ I was born in the Soviet Union.
+
+18:07.280 --> 18:11.480
+ I can certainly connect to the worry
+
+18:11.480 --> 18:12.520
+ that you're expressing.
+
+18:12.520 --> 18:14.920
+ At the same time, probably both you and I
+
+18:14.920 --> 18:19.000
+ and you're very much so, you know,
+
+18:19.000 --> 18:22.440
+ there's an exciting possibility
+
+18:22.440 --> 18:27.000
+ that you can have a deep connection with the machine.
+
+18:27.000 --> 18:28.000
+ Yeah, yeah.
+
+18:28.000 --> 18:29.000
+ Right, so.
+
+18:29.000 --> 18:35.000
+ Those of us, I've admitted students who say that they,
+
+18:35.000 --> 18:37.400
+ you know, when you list, like, who do you most wish
+
+18:37.400 --> 18:40.640
+ you could have lunch with or dinner with, right?
+
+18:40.640 --> 18:42.680
+ And they'll write, like, I don't like people.
+
+18:42.680 --> 18:44.040
+ I just like computers.
+
+18:44.040 --> 18:45.760
+ And one of them said to me once
+
+18:45.760 --> 18:48.480
+ when I had this party at my house,
+
+18:48.480 --> 18:52.560
+ I want you to know this is my only social event of the year,
+
+18:52.560 --> 18:54.840
+ my one social event of the year.
+
+18:54.840 --> 18:58.080
+ Like, okay, now this is a brilliant machine learning person,
+
+18:58.080 --> 18:59.080
+ right?
+
+18:59.080 --> 19:01.320
+ And we need that kind of brilliance and machine learning.
+
+19:01.320 --> 19:04.040
+ And I love that Computer Science welcomes people
+
+19:04.040 --> 19:07.040
+ who love people and people who are very awkward around people.
+
+19:07.040 --> 19:12.040
+ I love that this is a field that anybody could join.
+
+19:12.040 --> 19:15.040
+ We need all kinds of people.
+
+19:15.040 --> 19:17.040
+ And you don't need to be a social person.
+
+19:17.040 --> 19:19.040
+ I'm not trying to force people who don't like people
+
+19:19.040 --> 19:21.040
+ to suddenly become social.
+
+19:21.040 --> 19:26.040
+ At the same time, if most of the people building the AIs
+
+19:26.040 --> 19:29.040
+ of the future are the kind of people who don't like people,
+
+19:29.040 --> 19:31.040
+ we've got a little bit of a problem.
+
+19:31.040 --> 19:32.040
+ Hold on a second.
+
+19:32.040 --> 19:34.040
+ So let me, let me push back on that.
+
+19:34.040 --> 19:39.040
+ So don't you think a large percentage of the world can,
+
+19:39.040 --> 19:41.040
+ you know, there's loneliness.
+
+19:41.040 --> 19:44.040
+ There is a huge problem with loneliness and it's growing.
+
+19:44.040 --> 19:47.040
+ And so there's a longing for connection.
+
+19:47.040 --> 19:51.040
+ If you're lonely, you're part of a big and growing group.
+
+19:51.040 --> 19:52.040
+ Yes.
+
+19:52.040 --> 19:54.040
+ So we're in it together, I guess.
+
+19:54.040 --> 19:56.040
+ If you're lonely, join the group.
+
+19:56.040 --> 19:57.040
+ You're not alone.
+
+19:57.040 --> 19:58.040
+ You're not alone.
+
+19:58.040 --> 20:00.040
+ That's a good line.
+
+20:00.040 --> 20:04.040
+ But do you think there's a, you talked about some worry,
+
+20:04.040 --> 20:07.040
+ but do you think there's an exciting possibility
+
+20:07.040 --> 20:11.040
+ that something like Alexa, when these kinds of tools
+
+20:11.040 --> 20:16.040
+ can alleviate that loneliness in a way that other humans can't?
+
+20:16.040 --> 20:17.040
+ Yeah.
+
+20:17.040 --> 20:18.040
+ Yeah, definitely.
+
+20:18.040 --> 20:21.040
+ I mean, a great book can kind of alleviate loneliness.
+
+20:21.040 --> 20:22.040
+ Right.
+
+20:22.040 --> 20:23.040
+ Exactly.
+
+20:23.040 --> 20:25.040
+ Because you just get sucked into this amazing story
+
+20:25.040 --> 20:27.040
+ and you can't wait to go spend time with that character.
+
+20:27.040 --> 20:28.040
+ Right.
+
+20:28.040 --> 20:30.040
+ And they're not a human character.
+
+20:30.040 --> 20:32.040
+ There is a human behind it.
+
+20:32.040 --> 20:35.040
+ But yeah, it can be an incredibly delightful way
+
+20:35.040 --> 20:37.040
+ to pass the hours.
+
+20:37.040 --> 20:39.040
+ And it can meet needs.
+
+20:39.040 --> 20:43.040
+ Even, you know, I don't read those trashy romance books,
+
+20:43.040 --> 20:44.040
+ but somebody does, right?
+
+20:44.040 --> 20:46.040
+ And what are they getting from this?
+
+20:46.040 --> 20:50.040
+ Well, probably some of that feeling of being there, right?
+
+20:50.040 --> 20:54.040
+ Being there in that social moment, that romantic moment,
+
+20:54.040 --> 20:56.040
+ or connecting with somebody.
+
+20:56.040 --> 20:59.040
+ I've had a similar experience reading some science fiction books, right?
+
+20:59.040 --> 21:01.040
+ And connecting with the character, Orson Scott Card.
+
+21:01.040 --> 21:05.040
+ And, you know, just amazing writing and Inder's Game
+
+21:05.040 --> 21:07.040
+ and Speaker for the Dead, terrible title.
+
+21:07.040 --> 21:10.040
+ But those kind of books that pull you into a character,
+
+21:10.040 --> 21:13.040
+ and you feel like you're, you feel very social.
+
+21:13.040 --> 21:17.040
+ It's very connected, even though it's not responding to you.
+
+21:17.040 --> 21:19.040
+ And a computer, of course, can respond to you.
+
+21:19.040 --> 21:20.040
+ Right.
+
+21:20.040 --> 21:21.040
+ So it can deepen it, right?
+
+21:21.040 --> 21:25.040
+ You can have a very deep connection,
+
+21:25.040 --> 21:29.040
+ much more than the movie Her, you know, plays up, right?
+
+21:29.040 --> 21:30.040
+ Well, much more.
+
+21:30.040 --> 21:34.040
+ I mean, movie Her is already a pretty deep connection, right?
+
+21:34.040 --> 21:36.040
+ Well, but it's just a movie, right?
+
+21:36.040 --> 21:37.040
+ It's scripted.
+
+21:37.040 --> 21:38.040
+ It's just, you know.
+
+21:38.040 --> 21:42.040
+ But I mean, like, there can be a real interaction
+
+21:42.040 --> 21:46.040
+ where the character can learn and you can learn.
+
+21:46.040 --> 21:49.040
+ You could imagine it not just being you and one character.
+
+21:49.040 --> 21:51.040
+ You can imagine a group of characters.
+
+21:51.040 --> 21:53.040
+ You can imagine a group of people and characters,
+
+21:53.040 --> 21:56.040
+ human and AI connecting.
+
+21:56.040 --> 22:01.040
+ Where maybe a few people can't sort of be friends with everybody,
+
+22:01.040 --> 22:07.040
+ but the few people and their AIs can befriend more people.
+
+22:07.040 --> 22:10.040
+ There can be an extended human intelligence in there
+
+22:10.040 --> 22:14.040
+ where each human can connect with more people that way.
+
+22:14.040 --> 22:18.040
+ But it's still very limited.
+
+22:18.040 --> 22:21.040
+ But there are just, what I mean is there are many more possibilities
+
+22:21.040 --> 22:22.040
+ than what's in that movie.
+
+22:22.040 --> 22:24.040
+ So there's a tension here.
+
+22:24.040 --> 22:28.040
+ So one, you expressed a really serious concern about privacy,
+
+22:28.040 --> 22:31.040
+ about how governments can misuse the information.
+
+22:31.040 --> 22:34.040
+ And there's the possibility of this connection.
+
+22:34.040 --> 22:36.040
+ So let's look at Alexa.
+
+22:36.040 --> 22:38.040
+ So personal assistance.
+
+22:38.040 --> 22:42.040
+ For the most part, as far as I'm aware, they ignore your emotion.
+
+22:42.040 --> 22:47.040
+ They ignore even the context or the existence of you,
+
+22:47.040 --> 22:51.040
+ the intricate, beautiful, complex aspects of who you are,
+
+22:51.040 --> 22:56.040
+ except maybe aspects of your voice that help it recognize
+
+22:56.040 --> 22:58.040
+ speech recognition.
+
+22:58.040 --> 23:03.040
+ Do you think they should move towards trying to understand your emotion?
+
+23:03.040 --> 23:07.040
+ All of these companies are very interested in understanding human emotion.
+
+23:07.040 --> 23:13.040
+ More people are telling Siri every day they want to kill themselves.
+
+23:13.040 --> 23:17.040
+ Apple wants to know the difference between if a person is really suicidal
+
+23:17.040 --> 23:21.040
+ versus if a person is just kind of fooling around with Siri.
+
+23:21.040 --> 23:25.040
+ The words may be the same, the tone of voice,
+
+23:25.040 --> 23:31.040
+ and what surrounds those words is pivotal to understand
+
+23:31.040 --> 23:35.040
+ if they should respond in a very serious way, bring help to that person,
+
+23:35.040 --> 23:40.040
+ or if they should kind of jokingly tease back,
+
+23:40.040 --> 23:45.040
+ you just want to sell me for something else.
+
+23:45.040 --> 23:48.040
+ Like, how do you respond when somebody says that?
+
+23:48.040 --> 23:53.040
+ Well, you do want to err on the side of being careful and taking it seriously.
+
+23:53.040 --> 23:59.040
+ People want to know if the person is happy or stressed.
+
+23:59.040 --> 24:03.040
+ In part, well, so let me give you an altruistic reason
+
+24:03.040 --> 24:08.040
+ and a business profit motivated reason,
+
+24:08.040 --> 24:13.040
+ and there are people in companies that operate on both principles.
+
+24:13.040 --> 24:17.040
+ Altruistic people really care about their customers
+
+24:17.040 --> 24:20.040
+ and really care about helping you feel a little better at the end of the day,
+
+24:20.040 --> 24:24.040
+ and it would just make those people happy if they knew that they made your life better.
+
+24:24.040 --> 24:29.040
+ If you came home stressed and after talking with their product, you felt better.
+
+24:29.040 --> 24:35.040
+ There are other people who maybe have studied the way affect affects decision making
+
+24:35.040 --> 24:39.040
+ and prices people pay, and they know, I don't know if I should tell you,
+
+24:39.040 --> 24:44.040
+ the work of Jen Lerner on heartstrings and purse strings.
+
+24:44.040 --> 24:50.040
+ If we manipulate you into a slightly sadder mood, you'll pay more.
+
+24:50.040 --> 24:54.040
+ You'll pay more to change your situation.
+
+24:54.040 --> 24:58.040
+ You'll pay more for something you don't even need to make yourself feel better.
+
+24:58.040 --> 25:01.040
+ If they sound a little sad, maybe I don't want to cheer them up.
+
+25:01.040 --> 25:05.040
+ Maybe first I want to help them get something,
+
+25:05.040 --> 25:09.040
+ a little shopping therapy that helps them.
+
+25:09.040 --> 25:13.040
+ Which is really difficult for a company that's primarily funded on advertisements,
+
+25:13.040 --> 25:16.040
+ so they're encouraged to get you to...
+
+25:16.040 --> 25:20.040
+ To offer you products or Amazon that's primarily funded on you buying things from their store.
+
+25:20.040 --> 25:25.040
+ Maybe we need regulation in the future to put a little bit of a wall
+
+25:25.040 --> 25:29.040
+ between these agents that have access to our emotion
+
+25:29.040 --> 25:32.040
+ and agents that want to sell us stuff.
+
+25:32.040 --> 25:38.040
+ Maybe there needs to be a little bit more of a firewall in between those.
+
+25:38.040 --> 25:42.040
+ So maybe digging in a little bit on the interaction with Alexa,
+
+25:42.040 --> 25:45.040
+ you mentioned, of course, a really serious concern about,
+
+25:45.040 --> 25:50.040
+ like recognizing emotion if somebody is speaking of suicide or depression and so on,
+
+25:50.040 --> 25:54.040
+ but what about the actual interaction itself?
+
+25:54.040 --> 25:57.040
+ Do you think...
+
+25:57.040 --> 26:01.040
+ You mentioned clippy and being annoying.
+
+26:01.040 --> 26:04.040
+ What is the objective function we're trying to optimize?
+
+26:04.040 --> 26:09.040
+ Is it minimize annoyingness or maximize happiness?
+
+26:09.040 --> 26:12.040
+ Or if we look at human relations,
+
+26:12.040 --> 26:16.040
+ I think that push and pull, the tension, the dance,
+
+26:16.040 --> 26:20.040
+ the annoying, the flaws, that's what makes it fun.
+
+26:20.040 --> 26:23.040
+ So is there a room for...
+
+26:23.040 --> 26:25.040
+ What is the objective function?
+
+26:25.040 --> 26:29.040
+ In times when you want to have a little push and pull, think of kids sparring.
+
+26:29.040 --> 26:34.040
+ You know, I see my sons and one of them wants to provoke the other to be upset.
+
+26:34.040 --> 26:35.040
+ And that's fun.
+
+26:35.040 --> 26:38.040
+ And it's actually healthy to learn where your limits are,
+
+26:38.040 --> 26:40.040
+ to learn how to self regulate.
+
+26:40.040 --> 26:43.040
+ You can imagine a game where it's trying to make you mad
+
+26:43.040 --> 26:45.040
+ and you're trying to show self control.
+
+26:45.040 --> 26:48.040
+ And so if we're doing a AI human interaction
+
+26:48.040 --> 26:51.040
+ that's helping build resilience and self control,
+
+26:51.040 --> 26:54.040
+ whether it's to learn how to not be a bully
+
+26:54.040 --> 26:58.040
+ or how to turn the other cheek or how to deal with an abusive person in your life,
+
+26:58.040 --> 27:03.040
+ then you might need an AI that pushes your buttons, right?
+
+27:03.040 --> 27:07.040
+ But in general, do you want an AI that pushes your buttons?
+
+27:07.040 --> 27:11.040
+ Probably depends on your personality.
+
+27:11.040 --> 27:12.040
+ I don't.
+
+27:12.040 --> 27:17.040
+ I want one that's respectful, that is there to serve me
+
+27:17.040 --> 27:22.040
+ and that is there to extend my ability to do things.
+
+27:22.040 --> 27:24.040
+ I'm not looking for a rival.
+
+27:24.040 --> 27:26.040
+ I'm looking for a helper.
+
+27:26.040 --> 27:29.040
+ And that's the kind of AI I'd put my money on.
+
+27:29.040 --> 27:32.040
+ Your senses for the majority of people in the world,
+
+27:32.040 --> 27:34.040
+ in order to have a rich experience,
+
+27:34.040 --> 27:36.040
+ that's what they're looking for as well.
+
+27:36.040 --> 27:40.040
+ So if you look at the movie Her, spoiler alert,
+
+27:40.040 --> 27:45.040
+ I believe the program, the woman in the movie Her,
+
+27:45.040 --> 27:50.040
+ leaves the person for somebody else.
+
+27:50.040 --> 27:53.040
+ Says they don't want to be dating anymore.
+
+27:53.040 --> 27:56.040
+ Right.
+
+27:56.040 --> 27:59.040
+ Your senses, if Alexis said, you know what?
+
+27:59.040 --> 28:03.040
+ I actually had enough of you for a while,
+
+28:03.040 --> 28:05.040
+ so I'm going to shut myself off.
+
+28:05.040 --> 28:07.040
+ You don't see that as...
+
+28:07.040 --> 28:10.040
+ I'd say you're trash because I'm paid for you, right?
+
+28:10.040 --> 28:14.040
+ We've got to remember,
+
+28:14.040 --> 28:18.040
+ and this is where this blending human AI
+
+28:18.040 --> 28:22.040
+ as if we're equals is really deceptive,
+
+28:22.040 --> 28:26.040
+ because AI is something at the end of the day
+
+28:26.040 --> 28:29.040
+ that my students and I are making in the lab,
+
+28:29.040 --> 28:33.040
+ and we're choosing what it's allowed to say,
+
+28:33.040 --> 28:36.040
+ when it's allowed to speak, what it's allowed to listen to,
+
+28:36.040 --> 28:39.040
+ what it's allowed to act on,
+
+28:39.040 --> 28:43.040
+ given the inputs that we choose to expose it to,
+
+28:43.040 --> 28:45.040
+ what outputs it's allowed to have.
+
+28:45.040 --> 28:49.040
+ It's all something made by a human,
+
+28:49.040 --> 28:52.040
+ and if we want to make something that makes our lives miserable,
+
+28:52.040 --> 28:55.040
+ fine, I wouldn't invest in it as a business,
+
+28:55.040 --> 28:59.040
+ unless it's just there for self regulation training.
+
+28:59.040 --> 29:02.040
+ But I think we need to think about what kind of future we want,
+
+29:02.040 --> 29:05.040
+ and actually your question, I really like the...
+
+29:05.040 --> 29:07.040
+ What is the objective function?
+
+29:07.040 --> 29:10.040
+ Is it to calm people down sometimes?
+
+29:10.040 --> 29:14.040
+ Is it to always make people happy and calm them down?
+
+29:14.040 --> 29:16.040
+ Well, there was a book about that, right?
+
+29:16.040 --> 29:19.040
+ The Brave New World, you know, make everybody happy,
+
+29:19.040 --> 29:22.040
+ take your Soma if you're unhappy, take your happy pill,
+
+29:22.040 --> 29:24.040
+ and if you refuse to take your happy pill,
+
+29:24.040 --> 29:29.040
+ well, we'll threaten you by sending you to Iceland to live there.
+
+29:29.040 --> 29:32.040
+ I lived in Iceland three years. It's a great place.
+
+29:32.040 --> 29:35.040
+ Don't take your Soma, then go to Iceland.
+
+29:35.040 --> 29:37.040
+ A little TV commercial there.
+
+29:37.040 --> 29:40.040
+ I was a child there for a few years. It's a wonderful place.
+
+29:40.040 --> 29:43.040
+ So that part of the book never scared me.
+
+29:43.040 --> 29:48.040
+ But really, do we want AI to manipulate us into submission,
+
+29:48.040 --> 29:49.040
+ into making us happy?
+
+29:49.040 --> 29:56.040
+ Well, if you are a power obsessed, sick dictator, individual
+
+29:56.040 --> 29:59.040
+ who only wants to control other people to get your jollies in life,
+
+29:59.040 --> 30:04.040
+ then yeah, you want to use AI to extend your power in your scale
+
+30:04.040 --> 30:07.040
+ to force people into submission.
+
+30:07.040 --> 30:11.040
+ If you believe that the human race is better off being given freedom
+
+30:11.040 --> 30:15.040
+ and the opportunity to do things that might surprise you,
+
+30:15.040 --> 30:20.040
+ then you want to use AI to extend people's ability to build,
+
+30:20.040 --> 30:23.040
+ you want to build AI that extends human intelligence,
+
+30:23.040 --> 30:27.040
+ that empowers the weak and helps balance the power
+
+30:27.040 --> 30:29.040
+ between the weak and the strong,
+
+30:29.040 --> 30:32.040
+ not that gives more power to the strong.
+
+30:32.040 --> 30:38.040
+ So in this process of empowering people and sensing people
+
+30:38.040 --> 30:43.040
+ and what is your sense on emotion in terms of recognizing emotion?
+
+30:43.040 --> 30:47.040
+ The difference between emotion that is shown and emotion that is felt.
+
+30:47.040 --> 30:54.040
+ So yeah, emotion that is expressed on the surface through your face,
+
+30:54.040 --> 30:57.040
+ your body, and various other things,
+
+30:57.040 --> 31:00.040
+ and what's actually going on deep inside on the biological level,
+
+31:00.040 --> 31:04.040
+ on the neuroscience level, or some kind of cognitive level.
+
+31:04.040 --> 31:06.040
+ Yeah, yeah.
+
+31:06.040 --> 31:08.040
+ So no easy questions here.
+
+31:08.040 --> 31:11.040
+ Yeah, I'm sure there's no definitive answer,
+
+31:11.040 --> 31:16.040
+ but what's your sense, how far can we get by just looking at the face?
+
+31:16.040 --> 31:18.040
+ We're very limited when we just look at the face,
+
+31:18.040 --> 31:22.040
+ but we can get further than most people think we can get.
+
+31:22.040 --> 31:26.040
+ People think, hey, I have a great poker face,
+
+31:26.040 --> 31:28.040
+ therefore all you're ever going to get from me is neutral.
+
+31:28.040 --> 31:30.040
+ Well, that's naive.
+
+31:30.040 --> 31:35.040
+ We can read with the ordinary camera on your laptop or on your phone.
+
+31:35.040 --> 31:39.040
+ We can read from a neutral face if your heart is racing.
+
+31:39.040 --> 31:45.040
+ We can read from a neutral face if your breathing is becoming irregular
+
+31:45.040 --> 31:47.040
+ and showing signs of stress.
+
+31:47.040 --> 31:53.040
+ We can read under some conditions that maybe I won't give you details on,
+
+31:53.040 --> 31:57.040
+ how your heart rate variability power is changing.
+
+31:57.040 --> 32:03.040
+ That could be a sign of stress even when your heart rate is not necessarily accelerating.
+
+32:03.040 --> 32:06.040
+ Sorry, from physiosensors or from the face?
+
+32:06.040 --> 32:09.040
+ From the color changes that you cannot even see,
+
+32:09.040 --> 32:11.040
+ but the camera can see.
+
+32:11.040 --> 32:13.040
+ That's amazing.
+
+32:13.040 --> 32:15.040
+ So you can get a lot of signal.
+
+32:15.040 --> 32:18.040
+ So we get things people can't see using a regular camera,
+
+32:18.040 --> 32:22.040
+ and from that we can tell things about your stress.
+
+32:22.040 --> 32:25.040
+ So if you were just sitting there with a blank face
+
+32:25.040 --> 32:30.040
+ thinking nobody can read my emotion, well, you're wrong.
+
+32:30.040 --> 32:34.040
+ So that's really interesting, but that's from visual information from the face.
+
+32:34.040 --> 32:39.040
+ That's almost like cheating your way to the physiological state of the body
+
+32:39.040 --> 32:43.040
+ by being very clever with what you can do with vision.
+
+32:43.040 --> 32:44.040
+ With signal processing.
+
+32:44.040 --> 32:49.040
+ So that's really impressive, but if you just look at the stuff we humans can see,
+
+32:49.040 --> 32:54.040
+ the smile, the smirks, the subtle, all the facial actions.
+
+32:54.040 --> 32:57.040
+ So then you can hide that on your face for a limited amount of time.
+
+32:57.040 --> 33:01.040
+ Now, if you're just going in for a brief interview and you're hiding it,
+
+33:01.040 --> 33:03.040
+ that's pretty easy for most people.
+
+33:03.040 --> 33:08.040
+ If you are, however, surveilled constantly everywhere you go,
+
+33:08.040 --> 33:13.040
+ then it's going to say, gee, you know, Lex used to smile a lot,
+
+33:13.040 --> 33:15.040
+ and now I'm not seeing so many smiles.
+
+33:15.040 --> 33:22.040
+ And Roz used to, you know, laugh a lot and smile a lot very spontaneously.
+
+33:22.040 --> 33:26.040
+ And now I'm only seeing these not so spontaneous looking smiles
+
+33:26.040 --> 33:28.040
+ and only when she's asked these questions.
+
+33:28.040 --> 33:32.040
+ You know, that's probably not getting enough sleep.
+
+33:32.040 --> 33:35.040
+ We could look at that too.
+
+33:35.040 --> 33:37.040
+ So now I have to be a little careful too.
+
+33:37.040 --> 33:42.040
+ When I say we, you think we can't read your emotion and we can, it's not that binary.
+
+33:42.040 --> 33:48.040
+ What we're reading is more some physiological changes that relate to your activation.
+
+33:48.040 --> 33:52.040
+ Now, that doesn't mean that we know everything about how you feel.
+
+33:52.040 --> 33:54.040
+ In fact, we still know very little about how you feel.
+
+33:54.040 --> 33:56.040
+ Your thoughts are still private.
+
+33:56.040 --> 34:00.040
+ Your nuanced feelings are still completely private.
+
+34:00.040 --> 34:02.040
+ We can't read any of that.
+
+34:02.040 --> 34:06.040
+ So there's some relief that we can't read that.
+
+34:06.040 --> 34:09.040
+ Even brain imaging can't read that.
+
+34:09.040 --> 34:11.040
+ Wearables can't read that.
+
+34:11.040 --> 34:18.040
+ However, as we read your body state changes and we know what's going on in your environment,
+
+34:18.040 --> 34:21.040
+ and we look at patterns of those over time,
+
+34:21.040 --> 34:27.040
+ we can start to make some inferences about what you might be feeling.
+
+34:27.040 --> 34:31.040
+ And that is where it's not just the momentary feeling,
+
+34:31.040 --> 34:34.040
+ but it's more your stance toward things.
+
+34:34.040 --> 34:41.040
+ And that could actually be a little bit more scary with certain kinds of governmental,
+
+34:41.040 --> 34:48.040
+ control free people who want to know more about are you on their team or are you not.
+
+34:48.040 --> 34:51.040
+ And getting that information through over time.
+
+34:51.040 --> 34:54.040
+ So you're saying there's a lot of signal by looking at the change over time.
+
+34:54.040 --> 35:00.040
+ So you've done a lot of exciting work both in computer vision and physiological sounds like wearables.
+
+35:00.040 --> 35:03.040
+ What do you think is the best modality for,
+
+35:03.040 --> 35:08.040
+ what's the best window into the emotional soul?
+
+35:08.040 --> 35:10.040
+ Is it the face? Is it the voice?
+
+35:10.040 --> 35:12.040
+ Depends what you want to know.
+
+35:12.040 --> 35:14.040
+ It depends what you want to know.
+
+35:14.040 --> 35:16.040
+ Everything is informative.
+
+35:16.040 --> 35:18.040
+ Everything we do is informative.
+
+35:18.040 --> 35:20.040
+ So for health and well being and things like that,
+
+35:20.040 --> 35:29.040
+ do you find the wearable, measuring physiological signals is the best for health based stuff?
+
+35:29.040 --> 35:35.040
+ So here I'm going to answer empirically with data and studies we've been doing.
+
+35:35.040 --> 35:37.040
+ We've been doing studies now.
+
+35:37.040 --> 35:40.040
+ These are currently running with lots of different kinds of people,
+
+35:40.040 --> 35:44.040
+ but where we've published data, and I can speak publicly to it,
+
+35:44.040 --> 35:47.040
+ the data are limited right now to New England college students.
+
+35:47.040 --> 35:50.040
+ So that's a small group.
+
+35:50.040 --> 35:52.040
+ Among New England college students,
+
+35:52.040 --> 36:01.040
+ when they are wearing a wearable like the emphatic embrace here that's measuring skin conductance, movement, temperature,
+
+36:01.040 --> 36:10.040
+ and when they are using a smartphone that is collecting their time of day of when they're texting,
+
+36:10.040 --> 36:17.040
+ who they're texting, their movement around it, their GPS, the weather information based upon their location,
+
+36:17.040 --> 36:22.040
+ and when it's using machine learning and putting all of that together and looking not just at right now,
+
+36:22.040 --> 36:28.040
+ but looking at your rhythm of behaviors over about a week.
+
+36:28.040 --> 36:38.040
+ When we look at that, we are very accurate at forecasting tomorrow's stress, mood, and happy sad, mood, and health.
+
+36:38.040 --> 36:43.040
+ And when we look at which pieces of that are most useful,
+
+36:43.040 --> 36:48.040
+ first of all, if you have all the pieces, you get the best results.
+
+36:48.040 --> 36:53.040
+ If you have only the wearable, you get the next best results.
+
+36:53.040 --> 37:00.040
+ And that's still better than 80% accurate at forecasting tomorrow's levels.
+
+37:00.040 --> 37:12.040
+ Isn't that exciting because the wearable stuff with physiological information, it feels like it violates privacy less than the noncontact face based methods.
+
+37:12.040 --> 37:14.040
+ Yeah, it's interesting.
+
+37:14.040 --> 37:18.040
+ I think what people sometimes don't, you know, it's fine.
+
+37:18.040 --> 37:22.040
+ The early days people would say, oh, wearing something or giving blood is invasive, right?
+
+37:22.040 --> 37:26.040
+ Whereas a camera is less invasive because it's not touching you.
+
+37:26.040 --> 37:33.040
+ I think on the contrary, the things that are not touching you are maybe the scariest because you don't know when they're on or off.
+
+37:33.040 --> 37:39.040
+ And you don't know when, and you don't know who's behind it, right?
+
+37:39.040 --> 37:52.040
+ A wearable, depending upon what's happening to the data on it, if it's just stored locally or if it's streaming and what it is being attached to.
+
+37:52.040 --> 37:59.040
+ In a sense, you have the most control over it because it's also very easy to just take it off, right?
+
+37:59.040 --> 38:01.040
+ Now it's not sensing me.
+
+38:01.040 --> 38:07.040
+ So if I'm uncomfortable with what it's sensing, now I'm free, right?
+
+38:07.040 --> 38:13.040
+ If I'm comfortable with what it's sensing, then, and I happen to know everything about this one and what it's doing with it.
+
+38:13.040 --> 38:15.040
+ So I'm quite comfortable with it.
+
+38:15.040 --> 38:20.040
+ Then I'm, you know, I have control, I'm comfortable.
+
+38:20.040 --> 38:26.040
+ Control is one of the biggest factors for an individual in reducing their stress.
+
+38:26.040 --> 38:32.040
+ If I have control over it, if I know others to know about it, then my stress is a lot lower.
+
+38:32.040 --> 38:38.040
+ And I'm making an informed choice about whether to wear it or not, or when to wear it or not.
+
+38:38.040 --> 38:40.040
+ I want to wear it sometimes, maybe not others.
+
+38:40.040 --> 38:43.040
+ Right. So that control, yeah, I'm with you.
+
+38:43.040 --> 38:49.040
+ That control, even if, yeah, the ability to turn it off is a really empowering thing.
+
+38:49.040 --> 38:50.040
+ It's huge.
+
+38:50.040 --> 39:00.040
+ And we need to maybe, you know, if there's regulations, maybe that's number one to protect is people's ability to, is easy to opt out as to opt in.
+
+39:00.040 --> 39:01.040
+ Right.
+
+39:01.040 --> 39:04.040
+ So you've studied a bit of neuroscience as well.
+
+39:04.040 --> 39:16.040
+ How have looking at our own minds, sort of the biological stuff or the neurobiological, the neuroscience look at the signals in our brain,
+
+39:16.040 --> 39:20.040
+ helped you understand the problem and the approach of effective computing.
+
+39:20.040 --> 39:28.040
+ So originally I was a computer architect and I was building hardware and computer designs and I wanted to build ones that work like the brain.
+
+39:28.040 --> 39:33.040
+ So I've been studying the brain as long as I've been studying how to build computers.
+
+39:33.040 --> 39:35.040
+ Have you figured out anything yet?
+
+39:35.040 --> 39:37.040
+ Very little.
+
+39:37.040 --> 39:39.040
+ It's so amazing.
+
+39:39.040 --> 39:46.040
+ You know, they used to think like, oh, if you remove this chunk of the brain and you find this function goes away, well, that's the part of the brain that did it.
+
+39:46.040 --> 39:53.040
+ And then later they realize if you remove this other chunk of the brain, that function comes back and oh no, we really don't understand it.
+
+39:53.040 --> 40:02.040
+ Brains are so interesting and changing all the time and able to change in ways that will probably continue to surprise us.
+
+40:02.040 --> 40:14.040
+ When we were measuring stress, you may know the story where we found an unusual big skin conductance pattern on one wrist and one of our kids with autism.
+
+40:14.040 --> 40:20.040
+ And in trying to figure out how on earth you could be stressed on one wrist and not the other, like how can you get sweaty on one wrist, right?
+
+40:20.040 --> 40:26.040
+ When you when you get stressed with that sympathetic fight or flight response, like you kind of should like sweat more in some places than others,
+
+40:26.040 --> 40:28.040
+ but not more on one wrist than the other.
+
+40:28.040 --> 40:31.040
+ That didn't make any sense.
+
+40:31.040 --> 40:37.040
+ We learned that what had actually happened was a part of his brain had unusual electrical activity.
+
+40:37.040 --> 40:44.040
+ And that caused an unusually large sweat response on one wrist and not the other.
+
+40:44.040 --> 40:49.040
+ And since then we've learned that seizures cause this unusual electrical activity.
+
+40:49.040 --> 40:58.040
+ And depending where the seizure is, if it's in one place and it's staying there, you can have a big electrical response we can pick up with a wearable at one part of the body.
+
+40:58.040 --> 41:07.040
+ You can also have a seizure that spreads over the whole brain generalized grand mal seizure and that response spreads and we can pick it up pretty much anywhere.
+
+41:07.040 --> 41:13.040
+ As we learned this and then later built embrace that's now FDA cleared for seizure detection.
+
+41:13.040 --> 41:23.040
+ We have also built relationships with some of the most amazing doctors in the world who not only help people with unusual brain activity or epilepsy,
+
+41:23.040 --> 41:35.040
+ but some of them are also surgeons and they're going in and they're implanting electrodes, not just to momentarily read the strange patterns of brain activity that we'd like to see return to normal,
+
+41:35.040 --> 41:42.040
+ but also to read out continuously what's happening in some of these deep regions of the brain during most of life when these patients are not seizing.
+
+41:42.040 --> 41:45.040
+ Most of the time they're not seizing. Most of the time they're fine.
+
+41:45.040 --> 41:58.040
+ And so we are now working on mapping those deep brain regions that you can't even usually get with EEG scalp electrodes because the changes deep inside don't reach the surface.
+
+41:58.040 --> 42:04.040
+ But interesting when some of those regions are activated, we see a big skin conductance response.
+
+42:04.040 --> 42:08.040
+ Who would have thunk it, right? Like nothing here, but something here.
+
+42:08.040 --> 42:17.040
+ In fact, right after seizures that we think are the most dangerous ones that precede what's called Sudep, sudden unexpected death and epilepsy.
+
+42:17.040 --> 42:23.040
+ There's a period where the brain waves go flat and it looks like the person's brain has stopped, but it hasn't.
+
+42:23.040 --> 42:32.040
+ The activity has gone deep into a region that can make the cortical activity look flat, like a quick shutdown signal here.
+
+42:32.040 --> 42:38.040
+ It can unfortunately cause breathing to stop if it progresses long enough.
+
+42:38.040 --> 42:43.040
+ Before that happens, we see a big skin conductance response in the data that we have.
+
+42:43.040 --> 42:46.040
+ The longer this flattening, the bigger our response here.
+
+42:46.040 --> 42:52.040
+ So we have been trying to learn, you know, initially like, why are we getting a big response here when there's nothing here?
+
+42:52.040 --> 42:55.040
+ Well, it turns out there's something much deeper.
+
+42:55.040 --> 43:05.040
+ So we can now go inside the brains of some of these individuals, fabulous people who usually aren't seizing and get this data and start to map it.
+
+43:05.040 --> 43:09.040
+ So that's active research that we're doing right now with top medical partners.
+
+43:09.040 --> 43:18.040
+ So this wearable sensor that's looking at skin conductance can capture sort of the ripples of the complexity of what's going on in our brain.
+
+43:18.040 --> 43:27.040
+ So this little device, you have a hope that you can start to get the signal from the interesting things happening in the brain.
+
+43:27.040 --> 43:35.040
+ Yeah, we've already published the strong correlations between the size of this response and the flattening that happens afterwards.
+
+43:35.040 --> 43:42.040
+ And unfortunately also in a real suit up case where the patient died because the, well, we don't know why.
+
+43:42.040 --> 43:48.040
+ We don't know if somebody was there, it would have definitely prevented it, but we know that most suit ups happen when the person's alone.
+
+43:48.040 --> 44:01.040
+ And in this case, a suit up is an acronym, S U D E P, and it stands for the number two cause of years of life lost actually among all neurological disorders.
+
+44:01.040 --> 44:06.040
+ Stroke is number one, suit up is number two, but most people haven't heard of it.
+
+44:06.040 --> 44:11.040
+ Actually, I'll plug my Ted talk, it's on the front page of Ted right now, that talks about this.
+
+44:11.040 --> 44:14.040
+ And we hope to change that.
+
+44:14.040 --> 44:21.040
+ I hope everybody who's heard of SIDS and stroke will now hear of suit up because we think in most cases it's preventable.
+
+44:21.040 --> 44:27.040
+ If people take their meds and aren't alone when they have a seizure, not guaranteed to be preventable.
+
+44:27.040 --> 44:31.040
+ There are some exceptions, but we think most cases probably are.
+
+44:31.040 --> 44:41.040
+ So you have this embrace now in the version two wristband, right, for epilepsy management. That's the one that's FDA approved.
+
+44:41.040 --> 44:42.040
+ Yes.
+
+44:42.040 --> 44:44.040
+ Which is kind of cleared.
+
+44:44.040 --> 44:45.040
+ They say.
+
+44:45.040 --> 44:46.040
+ Sorry.
+
+44:46.040 --> 44:47.040
+ No, it's okay.
+
+44:47.040 --> 44:49.040
+ It essentially means it's approved for marketing.
+
+44:49.040 --> 44:50.040
+ Got it.
+
+44:50.040 --> 44:53.040
+ Just a side note, how difficult is that to do?
+
+44:53.040 --> 44:57.040
+ It's essentially getting FDA approval for computer science technology.
+
+44:57.040 --> 45:04.040
+ It's so agonizing. It's much harder than publishing multiple papers in top medical journals.
+
+45:04.040 --> 45:11.040
+ Yeah, we published peer reviewed, top medical journal, neurology, best results, and that's not good enough for the FDA.
+
+45:11.040 --> 45:19.040
+ Is that system, so if we look at the peer review of medical journals, there's flaws, the strengths, is the FDA approval process?
+
+45:19.040 --> 45:22.040
+ How does it compare to the peer review process?
+
+45:22.040 --> 45:24.040
+ Does it have its strength?
+
+45:24.040 --> 45:26.040
+ I take peer review over FDA any day.
+
+45:26.040 --> 45:28.040
+ Is that a good thing? Is that a good thing for FDA?
+
+45:28.040 --> 45:32.040
+ You're saying, does it stop some amazing technology from getting through?
+
+45:32.040 --> 45:33.040
+ Yeah, it does.
+
+45:33.040 --> 45:37.040
+ The FDA performs a very important good role in keeping people safe.
+
+45:37.040 --> 45:44.040
+ They keep things, they put you through tons of safety testing and that's wonderful and that's great.
+
+45:44.040 --> 45:48.040
+ I'm all in favor of the safety testing.
+
+45:48.040 --> 45:54.040
+ Sometimes they put you through additional testing that they don't have to explain why they put you through it.
+
+45:54.040 --> 46:00.040
+ You don't understand why you're going through it and it doesn't make sense and that's very frustrating.
+
+46:00.040 --> 46:09.040
+ Maybe they have really good reasons and it would do people a service to articulate those reasons.
+
+46:09.040 --> 46:10.040
+ Be more transparent.
+
+46:10.040 --> 46:12.040
+ Be more transparent.
+
+46:12.040 --> 46:27.040
+ As part of Empatica, we have sensors, so what kind of problems can we crack? What kind of things from seizures to autism to, I think I've heard you mentioned depression.
+
+46:27.040 --> 46:35.040
+ What kind of things can we alleviate? Can we detect? What's your hope of how we can make the world a better place with this wearable tech?
+
+46:35.040 --> 46:48.040
+ I would really like to see my fellow brilliant researchers step back and say, what are the really hard problems that we don't know how to solve
+
+46:48.040 --> 46:57.040
+ that come from people maybe we don't even see in our normal life because they're living in the poorer places, they're stuck on the bus,
+
+46:57.040 --> 47:05.040
+ they can't even afford the Uber or the Lyft or the data plan or all these other wonderful things we have that we keep improving on.
+
+47:05.040 --> 47:14.040
+ Meanwhile, there's all these folks left behind in the world and they're struggling with horrible diseases, with depression, with epilepsy, with diabetes,
+
+47:14.040 --> 47:25.040
+ with just awful stuff that maybe a little more time and attention hanging out with them and learning what are their challenges in life, what are their needs?
+
+47:25.040 --> 47:36.040
+ How do we help them have job skills? How do we help them have a hope and a future and a chance to have the great life that so many of us building technology have?
+
+47:36.040 --> 47:43.040
+ And then how would that reshape the kinds of AI that we build? How would that reshape the new apps that we build?
+
+47:43.040 --> 47:57.040
+ Or maybe we need to focus on how to make things more low cost and green instead of $1,000 phones. I mean, come on. Why can't we be thinking more about things that do more with less for these books?
+
+47:57.040 --> 48:09.040
+ Quality of life is not related to the cost of your phone. It's not something that, it's been shown that what, about $75,000 of income and happiness is the same.
+
+48:09.040 --> 48:16.040
+ However, I can tell you, you get a lot of happiness from helping other people. You get a lot more than $75,000 buys.
+
+48:16.040 --> 48:30.040
+ So how do we connect up the people who have real needs with the people who have the ability to build the future and build the kind of future that truly improves the lives of all the people that are currently being left behind?
+
+48:30.040 --> 48:53.040
+ So let me return just briefly on a point, maybe in movie her. So do you think if we look farther into the future, you said so much of the benefit from making our technology more empathetic to us human beings would make them better tools, empower us, make our lives better.
+
+48:53.040 --> 49:08.040
+ If we look farther into the future, do you think we'll ever create an AI system that we can fall in love with and loves us back on a level that is similar to human to human interaction, like in the movie her or beyond?
+
+49:08.040 --> 49:25.040
+ I think we can simulate it in ways that could, you know, sustain engagement for a while. Would it be as good as another person? I don't think so for if you're used to like good people.
+
+49:25.040 --> 49:39.040
+ If you've just grown up with nothing but abuse and you can't stand human beings, can we do something that helps you there that gives you something through a machine? Yeah, but that's pretty low bar, right? If you've only encountered pretty awful people.
+
+49:39.040 --> 50:00.040
+ If you've encountered wonderful, amazing people, we're nowhere near building anything like that. And I'm, I would not bet on building it. I would bet instead on building the kinds of AI that helps all helps kind of raise all boats that helps all people be better people helps all
+
+50:00.040 --> 50:14.040
+ people figure out if they're getting sick tomorrow and helps give them what they need to stay well tomorrow. That's the kind of AI I want to build that improves human lives, not the kind of AI that just walks on the Tonight Show and people go, wow, look how smart that is.
+
+50:14.040 --> 50:18.040
+ You know, really, like, and then it goes back in a box, you know.
+
+50:18.040 --> 50:40.040
+ So on that point, if we continue looking a little bit into the future, do you think an AI that's empathetic and does improve our lives need to have a physical presence, a body, and even let me cautiously say the C word consciousness and even fear of mortality.
+
+50:40.040 --> 50:59.040
+ So some of those human characteristics, do you think it needs to have those aspects? Or can it remain simply machine learning tool that learns from data of behavior that that learns to make us based on previous patterns feel better?
+
+50:59.040 --> 51:02.040
+ Or does it need those elements of consciousness?
+
+51:02.040 --> 51:13.040
+ It depends on your goals. If you're making a movie, it needs a body, it needs a gorgeous body, it needs to act like it has consciousness, it needs to act like it has emotion, right, because that's what sells.
+
+51:13.040 --> 51:17.040
+ That's what's going to get me to show up and enjoy the movie. Okay.
+
+51:17.040 --> 51:34.040
+ In real life, does it need all that? Well, if you've read Orson Scott Card, Ender's Game, Speaker for the Dead, you know, it could just be like a little voice in your earring, right, and you could have an intimate relationship and it could get to know you, and it doesn't need to be a robot.
+
+51:34.040 --> 51:43.040
+ But that doesn't make this compelling of a movie, right? I mean, we already think it's kind of weird when a guy looks like he's talking to himself on the train, you know, even though it's earbuds.
+
+51:43.040 --> 51:57.040
+ So we have these, embodied is more powerful, embodied when you compare interactions with an embodied robot versus a video of a robot versus no robot.
+
+51:57.040 --> 52:08.040
+ The robot is more engaging, the robot gets our attention more, the robot when you walk in your house is more likely to get you to remember to do the things that you asked it to do because it's kind of got a physical presence.
+
+52:08.040 --> 52:21.040
+ You can avoid it if you don't like it, it could see you're avoiding it. There's a lot of power to being embodied. There will be embodied AIs. They have great power and opportunity and potential.
+
+52:21.040 --> 52:32.040
+ There will also be AIs that aren't embodied that just our little software assistants that help us with different things that may get to know things about us.
+
+52:32.040 --> 52:42.040
+ Will they be conscious? There will be attempts to program them to make them appear to be conscious. We can already write programs that make it look like, what do you mean?
+
+52:42.040 --> 52:52.040
+ Of course, I'm aware that you're there, right? I mean, it's trivial to say stuff like that. It's easy to fool people. But does it actually have conscious experience like we do?
+
+52:52.040 --> 53:01.040
+ Nobody has a clue how to do that yet. That seems to be something that is beyond what any of us knows how to build now.
+
+53:01.040 --> 53:11.040
+ Will it have to have that? I think you can get pretty far with a lot of stuff without it. Will we accord it rights?
+
+53:11.040 --> 53:16.040
+ Well, that's more a political game that it is a question of real consciousness.
+
+53:16.040 --> 53:25.040
+ Can you go to jail for turning off Alexa? It's a question for an election maybe a few decades from now.
+
+53:25.040 --> 53:34.040
+ Sophia Robot's already been given rights as a citizen in Saudi Arabia, right? Even before women have full rights.
+
+53:34.040 --> 53:42.040
+ Then the robot was still put back in the box to be shipped to the next place where it would get a paid appearance, right?
+
+53:42.040 --> 53:50.040
+ Yeah, it's dark and almost comedic, if not absurd.
+
+53:50.040 --> 54:03.040
+ So I've heard you speak about your journey in finding faith and how you discovered some wisdoms about life and beyond from reading the Bible.
+
+54:03.040 --> 54:11.040
+ And I've also heard you say that you said scientists too often assume that nothing exists beyond what can be currently measured.
+
+54:11.040 --> 54:28.040
+ Materialism and scientism. In some sense, this assumption enables the near term scientific method, assuming that we can uncover the mysteries of this world by the mechanisms of measurement that we currently have.
+
+54:28.040 --> 54:38.040
+ But we easily forget that we've made this assumption. So what do you think we missed out on by making that assumption?
+
+54:38.040 --> 54:49.040
+ It's fine to limit the scientific method to things we can measure and reason about and reproduce. That's fine.
+
+54:49.040 --> 54:57.040
+ I think we have to recognize that sometimes we scientists also believe in things that happen historically, you know, like I believe the Holocaust happened.
+
+54:57.040 --> 55:11.040
+ I can't prove events from past history scientifically. You prove them with historical evidence, right, with the impact they had on people with eyewitness testimony and things like that.
+
+55:11.040 --> 55:21.040
+ So a good thinker recognizes that science is one of many ways to get knowledge. It's not the only way.
+
+55:21.040 --> 55:31.040
+ And there's been some really bad philosophy and bad thinking recently called scientism where people say science is the only way to get to truth.
+
+55:31.040 --> 55:43.040
+ And it's not. It just isn't. There are other ways that work also, like knowledge of love with someone. You don't prove your love through science, right?
+
+55:43.040 --> 56:01.040
+ So history, philosophy, love, a lot of other things in life show us that there's more ways to gain knowledge and truth if you're willing to believe there is such a thing, and I believe there is, than science.
+
+56:01.040 --> 56:14.040
+ I am a scientist, however, and in my science, I do limit my science to the things that the scientific method can do. But I recognize that it's myopic to say that that's all there is.
+
+56:14.040 --> 56:28.040
+ Right. Just like you listed, there's all the why questions. And really, we know if we're being honest with ourselves, the percent of what we really know is basically zero relative to the full mystery of those.
+
+56:28.040 --> 56:34.040
+ Measure theory, a set of measures, if I have a finite amount of knowledge, which I do.
+
+56:34.040 --> 56:45.040
+ So you said that you believe in truth. So let me ask that old question. What do you think this thing is all about? Life on earth?
+
+56:45.040 --> 56:53.040
+ Life, the universe and everything. I can't put Douglas Adams 42. My favorite number. By the way, that's my street address.
+
+56:53.040 --> 57:00.040
+ My husband and I guessed the exact same number for our house. We got to pick it. And there's a reason we picked 42. Yeah.
+
+57:00.040 --> 57:05.040
+ So is it just 42 or do you have other words that you can put around it?
+
+57:05.040 --> 57:15.040
+ Well, I think there's a grand adventure and I think this life is a part of it. I think there's a lot more to it than meets the eye and the heart and the mind and the soul here.
+
+57:15.040 --> 57:28.040
+ We see, but through a glass dimly in this life, we see only a part of all there is to know. If people haven't read the Bible, they should if they consider themselves educated.
+
+57:28.040 --> 57:36.040
+ And you could read Proverbs and find tremendous wisdom in there that cannot be scientifically proven.
+
+57:36.040 --> 57:43.040
+ But when you read it, there's something in you like a musician knows when the instruments played right and it's beautiful.
+
+57:43.040 --> 57:54.040
+ There's something in you that comes alive and knows that there's a truth there that like your strings are being plucked by the master instead of by me playing when I pluck it.
+
+57:54.040 --> 57:57.040
+ But probably when you play, it sounds spectacular.
+
+57:57.040 --> 58:11.040
+ And when you encounter those truths, there's something in you that sings and knows that there is more than what I can prove mathematically or program a computer to do.
+
+58:11.040 --> 58:19.040
+ Don't get me wrong. The math is gorgeous. The computer programming can be brilliant. It's inspiring, right? We want to do more.
+
+58:19.040 --> 58:26.040
+ None of this squashes my desire to do science or to get knowledge through science. I'm not dissing the science at all.
+
+58:26.040 --> 58:33.040
+ I grow even more in awe of what the science can do because I'm more in awe of all there is we don't know.
+
+58:33.040 --> 58:41.040
+ And really at the heart of science, you have to have a belief that there's truth that there's something greater to be discovered.
+
+58:41.040 --> 58:48.040
+ And some scientists may not want to use the faith word, but it's faith that drives us to do science.
+
+58:48.040 --> 59:00.040
+ It's faith that there is truth that there's something to know that we don't know that it's worth knowing that it's worth working hard and that there is meaning that there is such a thing as meaning,
+
+59:00.040 --> 59:07.040
+ which by the way, science can't prove either. We have to kind of start with some assumptions that there's things like truth and meaning.
+
+59:07.040 --> 59:14.040
+ And these are really questions philosophers own, right? This is their space of philosophers and theologians at some level.
+
+59:14.040 --> 59:24.040
+ So these are things science, you know, if we when people claim that science will tell you all truth, that's there's a name for that.
+
+59:24.040 --> 59:29.040
+ It's it's its own kind of faith. It's scientism and it's very myopic.
+
+59:29.040 --> 59:37.040
+ Yeah, there's a much bigger world out there to be explored in ways that science may not, at least for now, allows to explore.
+
+59:37.040 --> 59:45.040
+ Yeah. And there's meaning and purpose and hope and joy and love and all these awesome things that make it all worthwhile too.
+
+59:45.040 --> 59:49.040
+ I don't think there's a better way to end it, Ross. Thank you so much for talking today.
+
+59:49.040 --> 1:00:11.040
+ Thanks, Lex. What a pleasure. Great questions.
+

vtt/episode_026_large.vtt

@@ -0,0 +1,2534 @@
+WEBVTT
+
+00:00.000 --> 00:02.700
+ The following is a conversation with Sean Carroll.
+
+00:02.700 --> 00:04.900
+ He's a theoretical physicist at Caltech
+
+00:04.900 --> 00:08.780
+ specializing in quantum mechanics, gravity, and cosmology.
+
+00:08.780 --> 00:11.620
+ He's the author of several popular books,
+
+00:11.620 --> 00:15.340
+ one on the arrow of time called From Eternity to Here,
+
+00:15.340 --> 00:17.820
+ one on the Higgs boson called Particle
+
+00:17.820 --> 00:19.140
+ at the End of the Universe,
+
+00:19.140 --> 00:22.540
+ and one on science and philosophy called The Big Picture
+
+00:22.540 --> 00:26.340
+ on the Origins of Life, Meaning, and the Universe Itself.
+
+00:26.340 --> 00:28.700
+ He has an upcoming book on quantum mechanics
+
+00:28.700 --> 00:32.660
+ that you can preorder now called Something Deeply Hidden.
+
+00:32.660 --> 00:36.060
+ He writes one of my favorite blogs on his website,
+
+00:36.060 --> 00:37.980
+ preposterousuniverse.com.
+
+00:37.980 --> 00:40.460
+ I recommend clicking on the Greatest Hits link
+
+00:40.460 --> 00:43.340
+ that lists accessible, interesting posts
+
+00:43.340 --> 00:45.660
+ on the arrow of time, dark matter, dark energy,
+
+00:45.660 --> 00:47.620
+ the Big Bang, general relativity,
+
+00:47.620 --> 00:49.580
+ string theory, quantum mechanics,
+
+00:49.580 --> 00:53.180
+ and the big meta questions about the philosophy of science,
+
+00:53.180 --> 00:57.620
+ God, ethics, politics, academia, and much, much more.
+
+00:57.620 --> 01:00.300
+ Finally, and perhaps most famously,
+
+01:00.300 --> 01:03.660
+ he's the host of a podcast called Mindscape
+
+01:03.660 --> 01:06.940
+ that you should subscribe to and support on Patreon.
+
+01:06.940 --> 01:08.820
+ Along with the Joe Rogan experience,
+
+01:08.820 --> 01:10.500
+ Sam Harris's Making Sense,
+
+01:10.500 --> 01:13.100
+ and Dan Carlin's Hardcore History,
+
+01:13.100 --> 01:15.860
+ Sean's Mindscape podcast is one of my favorite ways
+
+01:15.860 --> 01:18.820
+ to learn new ideas or explore different perspectives
+
+01:18.820 --> 01:22.140
+ and ideas that I thought I understood.
+
+01:22.140 --> 01:24.660
+ It was truly an honor to meet
+
+01:24.660 --> 01:27.240
+ and spend a couple hours with Sean.
+
+01:27.240 --> 01:28.980
+ It's a bit heartbreaking to say
+
+01:28.980 --> 01:30.540
+ that for the first time ever,
+
+01:30.540 --> 01:32.500
+ the audio recorder for this podcast
+
+01:32.500 --> 01:34.940
+ died in the middle of our conversation.
+
+01:34.940 --> 01:36.320
+ There's technical reasons for this,
+
+01:36.320 --> 01:38.380
+ having to do with phantom power
+
+01:38.380 --> 01:41.060
+ that I now understand and will avoid.
+
+01:41.060 --> 01:44.220
+ It took me one hour to notice and fix the problem.
+
+01:44.220 --> 01:48.340
+ So, much like the universe is 68% dark energy,
+
+01:48.340 --> 01:51.340
+ roughly the same amount from this conversation was lost,
+
+01:51.340 --> 01:54.220
+ except in the memories of the two people involved
+
+01:54.220 --> 01:56.320
+ and in my notes.
+
+01:56.320 --> 01:59.940
+ I'm sure we'll talk again and continue this conversation
+
+01:59.940 --> 02:02.420
+ on this podcast or on Sean's.
+
+02:02.420 --> 02:05.300
+ And of course, I look forward to it.
+
+02:05.300 --> 02:07.820
+ This is the Artificial Intelligence podcast.
+
+02:07.820 --> 02:11.060
+ If you enjoy it, subscribe on YouTube, iTunes,
+
+02:11.060 --> 02:12.520
+ support it on Patreon,
+
+02:12.520 --> 02:16.660
+ or simply connect with me on Twitter at Lex Friedman.
+
+02:16.660 --> 02:21.380
+ And now, here's my conversation with Sean Carroll.
+
+02:21.380 --> 02:23.540
+ What do you think is more interesting and impactful,
+
+02:23.540 --> 02:26.860
+ understanding how the universe works at a fundamental level
+
+02:26.860 --> 02:29.180
+ or understanding how the human mind works?
+
+02:29.180 --> 02:31.960
+ You know, of course this is a crazy,
+
+02:31.960 --> 02:33.940
+ meaningless, unanswerable question in some sense,
+
+02:33.940 --> 02:35.140
+ because they're both very interesting
+
+02:35.140 --> 02:37.500
+ and there's no absolute scale of interestingness
+
+02:37.500 --> 02:39.180
+ that we can rate them on.
+
+02:39.180 --> 02:41.140
+ There's a glib answer that says the human brain
+
+02:41.140 --> 02:43.060
+ is part of the universe, right?
+
+02:43.060 --> 02:44.420
+ And therefore, understanding the universe
+
+02:44.420 --> 02:47.020
+ is more fundamental than understanding the human brain.
+
+02:47.020 --> 02:49.580
+ But do you really believe that once we understand
+
+02:49.580 --> 02:51.500
+ the fundamental way the universe works
+
+02:51.500 --> 02:53.740
+ at the particle level, the forces,
+
+02:53.740 --> 02:55.820
+ we would be able to understand how the mind works?
+
+02:55.820 --> 02:56.660
+ No, certainly not.
+
+02:56.660 --> 02:58.740
+ We cannot understand how ice cream works
+
+02:58.740 --> 03:01.060
+ just from understanding how particles work, right?
+
+03:01.060 --> 03:02.740
+ So I'm a big believer in emergence.
+
+03:02.740 --> 03:05.300
+ I'm a big believer that there are different ways
+
+03:05.300 --> 03:06.660
+ of talking about the world
+
+03:07.900 --> 03:11.180
+ beyond just the most fundamental microscopic one.
+
+03:11.180 --> 03:13.860
+ You know, when we talk about tables and chairs
+
+03:13.860 --> 03:15.120
+ and planets and people,
+
+03:15.120 --> 03:17.300
+ we're not talking the language of particle physics
+
+03:17.300 --> 03:18.380
+ and cosmology.
+
+03:18.380 --> 03:20.860
+ So, but understanding the universe,
+
+03:20.860 --> 03:24.060
+ you didn't say just at the most fundamental level, right?
+
+03:24.060 --> 03:26.740
+ So understanding the universe at all levels
+
+03:26.740 --> 03:28.200
+ is part of that.
+
+03:28.200 --> 03:29.940
+ I do think, you know, to be a little bit more fair
+
+03:29.940 --> 03:33.980
+ to the question, there probably are general principles
+
+03:33.980 --> 03:38.500
+ of complexity, biology, information processing,
+
+03:38.500 --> 03:41.820
+ memory, knowledge, creativity
+
+03:41.820 --> 03:45.620
+ that go beyond just the human brain, right?
+
+03:45.620 --> 03:47.800
+ And maybe one could count understanding those
+
+03:47.800 --> 03:49.140
+ as part of understanding the universe.
+
+03:49.140 --> 03:50.480
+ The human brain, as far as we know,
+
+03:50.480 --> 03:54.300
+ is the most complex thing in the universe.
+
+03:54.300 --> 03:57.420
+ So there's, it's certainly absurd to think
+
+03:57.420 --> 03:58.860
+ that by understanding the fundamental laws
+
+03:58.860 --> 04:00.340
+ of particle physics,
+
+04:00.340 --> 04:02.860
+ you get any direct insight on how the brain works.
+
+04:02.860 --> 04:05.940
+ But then there's this step from the fundamentals
+
+04:05.940 --> 04:08.700
+ of particle physics to information processing,
+
+04:08.700 --> 04:10.820
+ which a lot of physicists and philosophers
+
+04:10.820 --> 04:12.460
+ may be a little bit carelessly take
+
+04:12.460 --> 04:14.620
+ when they talk about artificial intelligence.
+
+04:14.620 --> 04:18.020
+ Do you think of the universe
+
+04:18.020 --> 04:21.300
+ as a kind of a computational device?
+
+04:21.300 --> 04:24.140
+ No, to be like, the honest answer there is no.
+
+04:24.140 --> 04:26.300
+ There's a sense in which the universe
+
+04:26.300 --> 04:29.140
+ processes information, clearly.
+
+04:29.140 --> 04:30.700
+ There's a sense in which the universe
+
+04:30.700 --> 04:33.880
+ is like a computer, clearly.
+
+04:33.880 --> 04:36.500
+ But in some sense, I think,
+
+04:36.500 --> 04:38.540
+ I tried to say this once on my blog
+
+04:38.540 --> 04:39.360
+ and no one agreed with me,
+
+04:39.360 --> 04:42.360
+ but the universe is more like a computation
+
+04:42.360 --> 04:45.060
+ than a computer because the universe happens once.
+
+04:45.060 --> 04:46.900
+ A computer is a general purpose machine, right?
+
+04:46.900 --> 04:48.700
+ That you can ask it different questions,
+
+04:48.700 --> 04:50.140
+ even a pocket calculator, right?
+
+04:50.140 --> 04:52.980
+ And it's set up to answer certain kinds of questions.
+
+04:52.980 --> 04:54.340
+ The universe isn't that.
+
+04:54.340 --> 04:57.360
+ So information processing happens in the universe,
+
+04:57.360 --> 04:59.220
+ but it's not what the universe is.
+
+04:59.220 --> 05:01.580
+ And I know your MIT colleague, Seth Lloyd,
+
+05:01.580 --> 05:03.820
+ feels very differently about this, right?
+
+05:03.820 --> 05:07.220
+ Well, you're thinking of the universe as a closed system.
+
+05:07.220 --> 05:08.060
+ I am.
+
+05:08.060 --> 05:11.780
+ So what makes a computer more like a PC,
+
+05:12.980 --> 05:15.500
+ like a computing machine is that there's a human
+
+05:15.500 --> 05:19.100
+ that every once comes up to it and moves the mouse around.
+
+05:19.100 --> 05:19.940
+ So input.
+
+05:19.940 --> 05:20.760
+ Gives it input.
+
+05:20.760 --> 05:21.600
+ Gives it input.
+
+05:23.500 --> 05:26.300
+ And that's why you're saying it's just a computation,
+
+05:26.300 --> 05:29.260
+ a deterministic thing that's just unrolling.
+
+05:29.260 --> 05:32.220
+ But the immense complexity of it
+
+05:32.220 --> 05:34.420
+ is nevertheless like processing.
+
+05:34.420 --> 05:39.420
+ There's a state and then it changes with good rules.
+
+05:40.140 --> 05:41.660
+ And there's a sense for a lot of people
+
+05:41.660 --> 05:44.420
+ that if the brain operates,
+
+05:44.420 --> 05:46.460
+ the human brain operates within that world,
+
+05:46.460 --> 05:49.340
+ then it's simply just a small subset of that.
+
+05:49.340 --> 05:52.500
+ And so there's no reason we can't build
+
+05:52.500 --> 05:55.560
+ arbitrarily great intelligences.
+
+05:55.560 --> 05:56.400
+ Yeah.
+
+05:56.400 --> 05:58.660
+ Do you think of intelligence in this way?
+
+05:58.660 --> 05:59.580
+ Intelligence is tricky.
+
+05:59.580 --> 06:01.660
+ I don't have a definition of it offhand.
+
+06:01.660 --> 06:05.460
+ So I remember this panel discussion that I saw on YouTube.
+
+06:05.460 --> 06:07.620
+ I wasn't there, but Seth Lloyd was on the panel.
+
+06:07.620 --> 06:10.540
+ And so was Martin Rees, the famous astrophysicist.
+
+06:10.540 --> 06:13.780
+ And Seth gave his shtick for why the universe is a computer
+
+06:13.780 --> 06:14.820
+ and explained this.
+
+06:14.820 --> 06:19.140
+ And Martin Rees said, so what is not a computer?
+
+06:19.140 --> 06:22.000
+ And Seth was like, oh, that's a good question.
+
+06:22.000 --> 06:22.840
+ I'm not sure.
+
+06:22.840 --> 06:24.960
+ Because if you have a sufficiently broad definition
+
+06:24.960 --> 06:28.360
+ of what a computer is, then everything is, right?
+
+06:28.360 --> 06:32.140
+ And the simile or the analogy gains force
+
+06:32.140 --> 06:34.380
+ when it excludes some things.
+
+06:34.380 --> 06:36.260
+ You know, is the moon going around the earth
+
+06:36.260 --> 06:38.620
+ performing a computation?
+
+06:38.620 --> 06:41.320
+ I can come up with definitions in which the answer is yes,
+
+06:41.320 --> 06:43.820
+ but it's not a very useful computation.
+
+06:43.820 --> 06:46.140
+ I think that it's absolutely helpful
+
+06:46.140 --> 06:49.620
+ to think about the universe in certain situations,
+
+06:49.620 --> 06:53.020
+ certain contexts, as an information processing device.
+
+06:53.020 --> 06:54.860
+ I'm even guilty of writing a paper
+
+06:54.860 --> 06:56.820
+ called Quantum Circuit Cosmology,
+
+06:56.820 --> 06:59.260
+ where we modeled the whole universe as a quantum circuit.
+
+06:59.260 --> 07:00.100
+ As a circuit.
+
+07:00.100 --> 07:01.340
+ As a circuit, yeah.
+
+07:01.340 --> 07:02.860
+ With qubits kind of thing?
+
+07:02.860 --> 07:05.040
+ With qubits basically, right, yeah.
+
+07:05.040 --> 07:07.440
+ So, and qubits becoming more and more entangled.
+
+07:07.440 --> 07:09.660
+ So do we wanna digress a little bit?
+
+07:09.660 --> 07:10.500
+ Let's do it.
+
+07:10.500 --> 07:11.340
+ It's kind of fun.
+
+07:11.340 --> 07:13.700
+ So here's a mystery about the universe
+
+07:13.700 --> 07:16.880
+ that is so deep and profound that nobody talks about it.
+
+07:16.880 --> 07:19.080
+ Space expands, right?
+
+07:19.080 --> 07:21.940
+ And we talk about, in a certain region of space,
+
+07:21.940 --> 07:23.620
+ a certain number of degrees of freedom,
+
+07:23.620 --> 07:25.540
+ a certain number of ways that the quantum fields
+
+07:25.540 --> 07:28.800
+ and the particles in that region can arrange themselves.
+
+07:28.800 --> 07:32.220
+ That number of degrees of freedom in a region of space
+
+07:32.220 --> 07:33.820
+ is arguably finite.
+
+07:33.820 --> 07:36.660
+ We actually don't know how many there are,
+
+07:36.660 --> 07:37.820
+ but there's a very good argument
+
+07:37.820 --> 07:39.420
+ that says it's a finite number.
+
+07:39.420 --> 07:43.540
+ So as the universe expands and space gets bigger,
+
+07:44.900 --> 07:46.780
+ are there more degrees of freedom?
+
+07:46.780 --> 07:48.540
+ If it's an infinite number, it doesn't really matter.
+
+07:48.540 --> 07:49.980
+ Infinity times two is still infinity.
+
+07:49.980 --> 07:53.480
+ But if it's a finite number, then there's more space,
+
+07:53.480 --> 07:54.420
+ so there's more degrees of freedom.
+
+07:54.420 --> 07:55.740
+ So where did they come from?
+
+07:55.740 --> 07:58.020
+ That would mean the universe is not a closed system.
+
+07:58.020 --> 08:01.500
+ There's more degrees of freedom popping into existence.
+
+08:01.500 --> 08:03.460
+ So what we suggested was
+
+08:03.460 --> 08:05.340
+ that there are more degrees of freedom,
+
+08:05.340 --> 08:07.980
+ and it's not that they're not there to start,
+
+08:07.980 --> 08:10.860
+ but they're not entangled to start.
+
+08:10.860 --> 08:12.820
+ So the universe that you and I know of,
+
+08:12.820 --> 08:15.440
+ the three dimensions around us that we see,
+
+08:15.440 --> 08:18.100
+ we said those are the entangled degrees of freedom
+
+08:18.100 --> 08:19.620
+ making up space time.
+
+08:19.620 --> 08:20.920
+ And as the universe expands,
+
+08:20.920 --> 08:25.180
+ there are a whole bunch of qubits in their zero state
+
+08:25.180 --> 08:28.140
+ that become entangled with the rest of space time
+
+08:28.140 --> 08:31.180
+ through the action of these quantum circuits.
+
+08:31.180 --> 08:35.580
+ So what does it mean that there's now more
+
+08:35.580 --> 08:39.300
+ degrees of freedom as they become more entangled?
+
+08:39.300 --> 08:40.300
+ Yeah, so.
+
+08:40.300 --> 08:41.660
+ As the universe expands.
+
+08:41.660 --> 08:43.300
+ That's right, so there's more and more degrees of freedom
+
+08:43.300 --> 08:46.420
+ that are entangled, that are playing part,
+
+08:46.420 --> 08:47.360
+ playing the role of part
+
+08:47.360 --> 08:49.600
+ of the entangled space time structure.
+
+08:49.600 --> 08:51.980
+ So the basic, the underlying philosophy is
+
+08:51.980 --> 08:54.620
+ that space time itself arises from the entanglement
+
+08:54.620 --> 08:57.560
+ of some fundamental quantum degrees of freedom.
+
+08:57.560 --> 09:00.820
+ Wow, okay, so at which point
+
+09:00.820 --> 09:05.260
+ is most of the entanglement happening?
+
+09:05.260 --> 09:07.460
+ Are we talking about close to the Big Bang?
+
+09:07.460 --> 09:11.820
+ Are we talking about throughout the time of the life?
+
+09:11.820 --> 09:12.660
+ Throughout history, yeah.
+
+09:12.660 --> 09:15.140
+ So the idea is that at the Big Bang,
+
+09:15.140 --> 09:16.780
+ almost all the degrees of freedom
+
+09:16.780 --> 09:19.700
+ that the universe could have were there,
+
+09:19.700 --> 09:22.420
+ but they were unentangled with anything else.
+
+09:22.420 --> 09:23.900
+ And that's a reflection of the fact
+
+09:23.900 --> 09:25.620
+ that the Big Bang had a low entropy.
+
+09:25.620 --> 09:28.020
+ It was a very simple, very small place.
+
+09:28.020 --> 09:31.420
+ And as space expands, more and more degrees of freedom
+
+09:31.420 --> 09:34.300
+ become entangled with the rest of the world.
+
+09:34.300 --> 09:35.940
+ Well, I have to ask John Carroll,
+
+09:35.940 --> 09:37.880
+ what do you think of the thought experiment
+
+09:37.880 --> 09:41.580
+ from Nick Bostrom that we're living in a simulation?
+
+09:41.580 --> 09:44.980
+ So I think, let me contextualize that a little bit more.
+
+09:44.980 --> 09:48.340
+ I think people don't actually take this thought experiments.
+
+09:48.340 --> 09:50.460
+ I think it's quite interesting.
+
+09:50.460 --> 09:52.900
+ It's not very useful, but it's quite interesting.
+
+09:52.900 --> 09:54.500
+ From the perspective of AI,
+
+09:54.500 --> 09:58.020
+ a lot of the learning that can be done usually happens
+
+09:58.020 --> 10:00.580
+ in simulation from artificial examples.
+
+10:00.580 --> 10:03.840
+ And so it's a constructive question to ask,
+
+10:04.900 --> 10:08.240
+ how difficult is our real world to simulate?
+
+10:08.240 --> 10:09.360
+ Right.
+
+10:09.360 --> 10:12.180
+ Which is kind of a dual part of,
+
+10:12.180 --> 10:14.100
+ if we're living in a simulation
+
+10:14.100 --> 10:16.420
+ and somebody built that simulation,
+
+10:16.420 --> 10:18.860
+ if you were to try to do it yourself, how hard would it be?
+
+10:18.860 --> 10:21.100
+ So obviously we could be living in a simulation.
+
+10:21.100 --> 10:23.000
+ If you just want the physical possibility,
+
+10:23.000 --> 10:25.420
+ then I completely agree that it's physically possible.
+
+10:25.420 --> 10:27.380
+ I don't think that we actually are.
+
+10:27.380 --> 10:30.300
+ So take this one piece of data into consideration.
+
+10:30.300 --> 10:33.960
+ You know, we live in a big universe, okay?
+
+10:35.140 --> 10:38.500
+ There's two trillion galaxies in our observable universe
+
+10:38.500 --> 10:41.660
+ with 200 billion stars in each galaxy, et cetera.
+
+10:41.660 --> 10:44.940
+ It would seem to be a waste of resources
+
+10:44.940 --> 10:46.540
+ to have a universe that big going on
+
+10:46.540 --> 10:47.540
+ just to do a simulation.
+
+10:47.540 --> 10:50.140
+ So in other words, I want to be a good Bayesian.
+
+10:50.140 --> 10:52.940
+ I want to ask under this hypothesis,
+
+10:52.940 --> 10:54.960
+ what do I expect to see?
+
+10:54.960 --> 10:56.780
+ So the first thing I would say is I wouldn't expect
+
+10:56.780 --> 11:00.340
+ to see a universe that was that big, okay?
+
+11:00.340 --> 11:02.540
+ The second thing is I wouldn't expect the resolution
+
+11:02.540 --> 11:05.020
+ of the universe to be as good as it is.
+
+11:05.020 --> 11:08.740
+ So it's always possible that if our superhuman simulators
+
+11:08.740 --> 11:09.900
+ only have finite resources,
+
+11:09.900 --> 11:12.420
+ that they don't render the entire universe, right?
+
+11:12.420 --> 11:14.340
+ That the part that is out there,
+
+11:14.340 --> 11:16.300
+ the two trillion galaxies,
+
+11:16.300 --> 11:19.640
+ isn't actually being simulated fully, okay?
+
+11:19.640 --> 11:22.740
+ But then the obvious extrapolation of that
+
+11:22.740 --> 11:25.500
+ is that only I am being simulated fully.
+
+11:25.500 --> 11:29.220
+ Like the rest of you are just non player characters, right?
+
+11:29.220 --> 11:30.520
+ I'm the only thing that is real.
+
+11:30.520 --> 11:32.780
+ The rest of you are just chat bots.
+
+11:32.780 --> 11:34.320
+ Beyond this wall, I see the wall,
+
+11:34.320 --> 11:36.020
+ but there is literally nothing
+
+11:36.020 --> 11:37.360
+ on the other side of the wall.
+
+11:37.360 --> 11:39.300
+ That is sort of the Bayesian prediction.
+
+11:39.300 --> 11:40.180
+ That's what it would be like
+
+11:40.180 --> 11:42.240
+ to do an efficient simulation of me.
+
+11:42.240 --> 11:45.700
+ So like none of that seems quite realistic.
+
+11:45.700 --> 11:50.700
+ I don't see, I hear the argument that it's just possible
+
+11:50.900 --> 11:53.300
+ and easy to simulate lots of things.
+
+11:53.300 --> 11:55.780
+ I don't see any evidence from what we know
+
+11:55.780 --> 11:59.340
+ about our universe that we look like a simulated universe.
+
+11:59.340 --> 12:00.180
+ Now, maybe you can say,
+
+12:00.180 --> 12:01.980
+ well, we don't know what it would look like,
+
+12:01.980 --> 12:04.520
+ but that's just abandoning your Bayesian responsibilities.
+
+12:04.520 --> 12:07.660
+ Like your job is to say under this theory,
+
+12:07.660 --> 12:09.500
+ here's what you would expect to see.
+
+12:09.500 --> 12:11.660
+ Yeah, so certainly if you think about simulation
+
+12:11.660 --> 12:14.340
+ as a thing that's like a video game
+
+12:14.340 --> 12:17.740
+ where only a small subset is being rendered.
+
+12:17.740 --> 12:22.740
+ But say the entire, all the laws of physics,
+
+12:22.740 --> 12:26.540
+ the entire closed system of the quote unquote universe,
+
+12:26.540 --> 12:27.780
+ it had a creator.
+
+12:27.780 --> 12:29.320
+ Yeah, it's always possible.
+
+12:29.320 --> 12:32.280
+ Right, so that's not useful to think about
+
+12:32.280 --> 12:34.020
+ when you're thinking about physics.
+
+12:34.020 --> 12:36.220
+ The way Nick Bostrom phrases it,
+
+12:36.220 --> 12:39.100
+ if it's possible to simulate a universe,
+
+12:39.100 --> 12:40.500
+ eventually we'll do it.
+
+12:40.500 --> 12:41.340
+ Right.
+
+12:42.700 --> 12:44.860
+ You can use that by the way for a lot of things.
+
+12:44.860 --> 12:45.700
+ Well, yeah.
+
+12:45.700 --> 12:48.540
+ But I guess the question is,
+
+12:48.540 --> 12:52.340
+ how hard is it to create a universe?
+
+12:52.340 --> 12:53.820
+ I wrote a little blog post about this
+
+12:53.820 --> 12:55.460
+ and maybe I'm missing something,
+
+12:55.460 --> 12:57.680
+ but there's an argument that says not only
+
+12:57.680 --> 13:00.500
+ that it might be possible to simulate a universe,
+
+13:00.500 --> 13:05.500
+ but probably if you imagine that you actually attribute
+
+13:05.980 --> 13:08.860
+ consciousness and agency to the little things
+
+13:08.860 --> 13:12.020
+ that we're simulating, to our little artificial beings,
+
+13:12.020 --> 13:13.420
+ there's probably a lot more of them
+
+13:13.420 --> 13:15.500
+ than there are ordinary organic beings in the universe
+
+13:15.500 --> 13:17.420
+ or there will be in the future, right?
+
+13:17.420 --> 13:18.500
+ So there's an argument that not only
+
+13:18.500 --> 13:20.760
+ is being a simulation possible,
+
+13:20.760 --> 13:23.560
+ it's probable because in the space
+
+13:23.560 --> 13:24.960
+ of all living consciousnesses,
+
+13:24.960 --> 13:26.620
+ most of them are being simulated, right?
+
+13:26.620 --> 13:28.860
+ Most of them are not at the top level.
+
+13:28.860 --> 13:30.540
+ I think that argument must be wrong
+
+13:30.540 --> 13:33.100
+ because it follows from that argument that,
+
+13:33.100 --> 13:36.920
+ if we're simulated, but we can also simulate other things,
+
+13:36.920 --> 13:38.840
+ well, but if we can simulate other things,
+
+13:38.840 --> 13:41.840
+ they can simulate other things, right?
+
+13:41.840 --> 13:44.320
+ If we give them enough power and resolution
+
+13:44.320 --> 13:45.980
+ and ultimately we'll reach a bottom
+
+13:45.980 --> 13:49.140
+ because the laws of physics in our universe have a bottom,
+
+13:49.140 --> 13:51.000
+ we're made of atoms and so forth,
+
+13:51.000 --> 13:55.100
+ so there will be the cheapest possible simulations.
+
+13:55.100 --> 13:57.700
+ And if you believe the original argument,
+
+13:57.700 --> 13:59.340
+ you should conclude that we should be
+
+13:59.340 --> 14:00.940
+ in the cheapest possible simulation
+
+14:00.940 --> 14:02.660
+ because that's where most people are.
+
+14:02.660 --> 14:03.620
+ But we don't look like that.
+
+14:03.620 --> 14:06.860
+ It doesn't look at all like we're at the edge of resolution,
+
+14:06.860 --> 14:09.540
+ that we're 16 bit things.
+
+14:09.540 --> 14:13.020
+ It seems much easier to make much lower level things
+
+14:13.020 --> 14:13.860
+ than we are.
+
+14:14.980 --> 14:18.220
+ And also, I questioned the whole approach
+
+14:18.220 --> 14:19.460
+ to the anthropic principle
+
+14:19.460 --> 14:22.340
+ that says we are typical observers in the universe.
+
+14:22.340 --> 14:23.660
+ I think that that's not actually,
+
+14:23.660 --> 14:27.340
+ I think that there's a lot of selection that we can do
+
+14:27.340 --> 14:30.180
+ that we're typical within things we already know,
+
+14:30.180 --> 14:32.280
+ but not typical within all of the universe.
+
+14:32.280 --> 14:35.800
+ So do you think there's intelligent life,
+
+14:35.800 --> 14:37.860
+ however you would like to define intelligent life,
+
+14:37.860 --> 14:39.940
+ out there in the universe?
+
+14:39.940 --> 14:44.660
+ My guess is that there is not intelligent life
+
+14:44.660 --> 14:48.820
+ in the observable universe other than us, simply
+
+14:48.820 --> 14:52.540
+ on the basis of the fact that the likely number
+
+14:52.540 --> 14:56.340
+ of other intelligent species in the observable universe,
+
+14:56.340 --> 15:00.320
+ there's two likely numbers, zero or billions.
+
+15:01.500 --> 15:02.580
+ And if there had been billions,
+
+15:02.580 --> 15:04.140
+ you would have noticed already.
+
+15:05.300 --> 15:07.340
+ For there to be literally like a small number,
+
+15:07.340 --> 15:09.380
+ like, you know, Star Trek,
+
+15:09.380 --> 15:13.300
+ there's a dozen intelligent civilizations in our galaxy,
+
+15:13.300 --> 15:17.340
+ but not a billion, that's weird.
+
+15:17.340 --> 15:18.500
+ That's sort of bizarre to me.
+
+15:18.500 --> 15:21.020
+ It's easy for me to imagine that there are zero others
+
+15:21.020 --> 15:22.620
+ because there's just a big bottleneck
+
+15:22.620 --> 15:24.980
+ to making multicellular life
+
+15:24.980 --> 15:27.020
+ or technological life or whatever.
+
+15:27.020 --> 15:28.580
+ It's very hard for me to imagine
+
+15:28.580 --> 15:30.140
+ that there's a whole bunch out there
+
+15:30.140 --> 15:32.300
+ that have somehow remained hidden from us.
+
+15:32.300 --> 15:34.700
+ The question I'd like to ask
+
+15:34.700 --> 15:36.820
+ is what would intelligent life look like?
+
+15:38.140 --> 15:40.500
+ What I mean by that question and where it's going
+
+15:40.500 --> 15:45.500
+ is what if intelligent life is just in some very big ways
+
+15:47.260 --> 15:51.500
+ different than the one that has on Earth?
+
+15:51.500 --> 15:53.900
+ That there's all kinds of intelligent life
+
+15:53.900 --> 15:55.420
+ that operates at different scales
+
+15:55.420 --> 15:57.300
+ of both size and temporal.
+
+15:57.300 --> 15:59.300
+ Right, that's a great possibility
+
+15:59.300 --> 16:00.800
+ because I think we should be humble
+
+16:00.800 --> 16:02.640
+ about what intelligence is, what life is.
+
+16:02.640 --> 16:04.020
+ We don't even agree on what life is,
+
+16:04.020 --> 16:07.020
+ much less what intelligent life is, right?
+
+16:07.020 --> 16:08.980
+ So that's an argument for humility,
+
+16:08.980 --> 16:10.860
+ saying there could be intelligent life
+
+16:10.860 --> 16:13.620
+ of a very different character, right?
+
+16:13.620 --> 16:18.060
+ Like you could imagine the dolphins are intelligent
+
+16:18.060 --> 16:20.500
+ but never invent space travel
+
+16:20.500 --> 16:21.460
+ because they live in the ocean
+
+16:21.460 --> 16:23.220
+ and they don't have thumbs, right?
+
+16:24.180 --> 16:27.860
+ So they never invent technology, they never invent smelting.
+
+16:27.860 --> 16:32.020
+ Maybe the universe is full of intelligent species
+
+16:32.020 --> 16:34.060
+ that just don't make technology, right?
+
+16:34.060 --> 16:36.320
+ That's compatible with the data, I think.
+
+16:36.320 --> 16:39.840
+ And I think maybe what you're pointing at
+
+16:39.840 --> 16:44.440
+ is even more out there versions of intelligence,
+
+16:44.440 --> 16:47.560
+ intelligence in intermolecular clouds
+
+16:47.560 --> 16:49.440
+ or on the surface of a neutron star
+
+16:49.440 --> 16:51.760
+ or in between the galaxies in giant things
+
+16:51.760 --> 16:54.560
+ where the equivalent of a heartbeat is 100 million years.
+
+16:56.440 --> 16:58.080
+ On the one hand, yes,
+
+16:58.080 --> 16:59.860
+ we should be very open minded about those things.
+
+16:59.860 --> 17:04.860
+ On the other hand, all of us share the same laws of physics.
+
+17:04.860 --> 17:08.240
+ There might be something about the laws of physics,
+
+17:08.240 --> 17:09.400
+ even though we don't currently know
+
+17:09.400 --> 17:10.920
+ exactly what that thing would be,
+
+17:10.920 --> 17:15.920
+ that makes meters and years
+
+17:16.160 --> 17:18.920
+ the right length and timescales for intelligent life.
+
+17:19.880 --> 17:22.240
+ Maybe not, but we're made of atoms,
+
+17:22.240 --> 17:23.780
+ atoms have a certain size,
+
+17:23.780 --> 17:27.280
+ we orbit stars or stars have a certain lifetime.
+
+17:27.280 --> 17:30.300
+ It's not impossible to me that there's a sweet spot
+
+17:30.300 --> 17:32.200
+ for intelligent life that we find ourselves in.
+
+17:32.200 --> 17:33.800
+ So I'm open minded either way,
+
+17:33.800 --> 17:35.280
+ I'm open minded either being humble
+
+17:35.280 --> 17:37.080
+ and there's all sorts of different kinds of life
+
+17:37.080 --> 17:39.280
+ or no, there's a reason we just don't know it yet
+
+17:39.280 --> 17:42.080
+ why life like ours is the kind of life that's out there.
+
+17:42.080 --> 17:43.320
+ Yeah, I'm of two minds too,
+
+17:43.320 --> 17:47.200
+ but I often wonder if our brains is just designed
+
+17:47.200 --> 17:52.200
+ to quite obviously to operate and see the world
+
+17:52.720 --> 17:56.360
+ in these timescales and we're almost blind
+
+17:56.360 --> 18:01.200
+ and the tools we've created for detecting things are blind
+
+18:01.200 --> 18:02.760
+ to the kind of observation needed
+
+18:02.760 --> 18:04.920
+ to see intelligent life at other scales.
+
+18:04.920 --> 18:07.040
+ Well, I'm totally open to that,
+
+18:07.040 --> 18:09.240
+ but so here's another argument I would make,
+
+18:09.240 --> 18:11.520
+ we have looked for intelligent life,
+
+18:11.520 --> 18:14.120
+ but we've looked at for it in the dumbest way we can,
+
+18:14.120 --> 18:16.600
+ by turning radio telescopes to the sky.
+
+18:16.600 --> 18:21.040
+ And why in the world would a super advanced civilization
+
+18:21.040 --> 18:24.040
+ randomly beam out radio signals wastefully
+
+18:24.040 --> 18:25.440
+ in all directions into the universe?
+
+18:25.440 --> 18:27.280
+ That just doesn't make any sense,
+
+18:27.280 --> 18:29.100
+ especially because in order to think
+
+18:29.100 --> 18:32.020
+ that you would actually contact another civilization,
+
+18:32.020 --> 18:33.840
+ you would have to do it forever,
+
+18:33.840 --> 18:35.840
+ you have to keep doing it for millions of years,
+
+18:35.840 --> 18:38.280
+ that sounds like a waste of resources.
+
+18:38.280 --> 18:43.120
+ If you thought that there were other solar systems
+
+18:43.120 --> 18:44.520
+ with planets around them,
+
+18:44.520 --> 18:47.000
+ where maybe intelligent life didn't yet exist,
+
+18:47.000 --> 18:48.600
+ but might someday,
+
+18:48.600 --> 18:51.380
+ you wouldn't try to talk to it with radio waves,
+
+18:51.380 --> 18:53.600
+ you would send a spacecraft out there
+
+18:53.600 --> 18:55.560
+ and you would park it around there
+
+18:55.560 --> 18:57.360
+ and it would be like, from our point of view,
+
+18:57.360 --> 19:00.700
+ it'd be like 2001, where there was a monolith.
+
+19:00.700 --> 19:01.540
+ Monolith.
+
+19:01.540 --> 19:02.380
+ There could be an artifact,
+
+19:02.380 --> 19:04.520
+ in fact, the other way works also, right?
+
+19:04.520 --> 19:07.360
+ There could be artifacts in our solar system
+
+19:08.440 --> 19:10.480
+ that have been put there
+
+19:10.480 --> 19:12.280
+ by other technologically advanced civilizations
+
+19:12.280 --> 19:14.640
+ and that's how we will eventually contact them.
+
+19:14.640 --> 19:16.840
+ We just haven't explored the solar system well enough yet
+
+19:16.840 --> 19:17.680
+ to find them.
+
+19:18.580 --> 19:20.000
+ The reason why we don't think about that
+
+19:20.000 --> 19:21.520
+ is because we're young and impatient, right?
+
+19:21.520 --> 19:24.000
+ Like, it would take more than my lifetime
+
+19:24.000 --> 19:26.080
+ to actually send something to another star system
+
+19:26.080 --> 19:27.800
+ and wait for it and then come back.
+
+19:27.800 --> 19:30.800
+ So, but if we start thinking on hundreds of thousands
+
+19:30.800 --> 19:32.720
+ of years or million year time scales,
+
+19:32.720 --> 19:34.600
+ that's clearly the right thing to do.
+
+19:34.600 --> 19:36.800
+ Are you excited by the thing
+
+19:36.800 --> 19:39.360
+ that Elon Musk is doing with SpaceX in general?
+
+19:39.360 --> 19:41.620
+ Space, but the idea of space exploration,
+
+19:41.620 --> 19:45.360
+ even though your, or your species is young and impatient?
+
+19:45.360 --> 19:46.200
+ Yeah.
+
+19:46.200 --> 19:49.200
+ No, I do think that space travel is crucially important,
+
+19:49.200 --> 19:50.800
+ long term.
+
+19:50.800 --> 19:52.500
+ Even to other star systems.
+
+19:52.500 --> 19:57.500
+ And I think that many people overestimate the difficulty
+
+19:57.500 --> 20:00.940
+ because they say, look, if you travel 1% the speed of light
+
+20:00.940 --> 20:02.020
+ to another star system,
+
+20:02.020 --> 20:04.060
+ we'll be dead before we get there, right?
+
+20:04.060 --> 20:06.180
+ And I think that it's much easier.
+
+20:06.180 --> 20:08.120
+ And therefore, when they write their science fiction stories,
+
+20:08.120 --> 20:09.580
+ they imagine we'd go faster than the speed of light
+
+20:09.580 --> 20:11.700
+ because otherwise they're too impatient, right?
+
+20:11.700 --> 20:13.600
+ We're not gonna go faster than the speed of light,
+
+20:13.600 --> 20:16.020
+ but we could easily imagine that the human lifespan
+
+20:16.020 --> 20:18.100
+ gets extended to thousands of years.
+
+20:18.100 --> 20:19.140
+ And once you do that,
+
+20:19.140 --> 20:21.180
+ then the stars are much closer effectively, right?
+
+20:21.180 --> 20:23.260
+ And then what's a hundred year trip, right?
+
+20:23.260 --> 20:25.820
+ So I think that that's gonna be the future,
+
+20:25.820 --> 20:28.700
+ the far future, not my lifetime once again,
+
+20:28.700 --> 20:30.380
+ but baby steps.
+
+20:30.380 --> 20:32.420
+ Unless your lifetime gets extended.
+
+20:32.420 --> 20:34.740
+ Well, it's in a race against time, right?
+
+20:34.740 --> 20:37.340
+ A friend of mine who actually thinks about these things
+
+20:37.340 --> 20:40.460
+ said, you know, you and I are gonna die,
+
+20:40.460 --> 20:43.060
+ but I don't know about our grandchildren.
+
+20:43.060 --> 20:45.940
+ That's, I don't know, predicting the future is hard,
+
+20:45.940 --> 20:47.900
+ but that's at least a plausible scenario.
+
+20:47.900 --> 20:51.820
+ And so, yeah, no, I think that as we discussed earlier,
+
+20:51.820 --> 20:56.780
+ there are threats to the earth, known and unknown, right?
+
+20:56.780 --> 21:01.780
+ Having spread humanity and biology elsewhere
+
+21:02.580 --> 21:04.940
+ is a really important longterm goal.
+
+21:04.940 --> 21:08.900
+ What kind of questions can science not currently answer,
+
+21:08.900 --> 21:09.920
+ but might soon?
+
+21:11.480 --> 21:13.860
+ When you think about the problems and the mysteries
+
+21:13.860 --> 21:17.840
+ before us that may be within reach of science.
+
+21:17.840 --> 21:20.300
+ I think an obvious one is the origin of life.
+
+21:20.300 --> 21:22.780
+ We don't know how that happened.
+
+21:22.780 --> 21:25.300
+ There's a difficulty in knowing how it happened historically
+
+21:25.300 --> 21:27.240
+ actually, you know, literally on earth,
+
+21:27.240 --> 21:30.500
+ but starting life from non life is something
+
+21:30.500 --> 21:32.420
+ I kind of think we're close to, right?
+
+21:32.420 --> 21:33.240
+ We're really.
+
+21:33.240 --> 21:34.080
+ You really think so?
+
+21:34.080 --> 21:36.740
+ Like how difficult is it to start life?
+
+21:36.740 --> 21:39.260
+ Well, I've talked to people,
+
+21:39.260 --> 21:41.780
+ including on the podcast about this.
+
+21:41.780 --> 21:43.340
+ You know, life requires three things.
+
+21:43.340 --> 21:44.220
+ Life as we know it.
+
+21:44.220 --> 21:45.500
+ So there's a difference with life,
+
+21:45.500 --> 21:47.060
+ which who knows what it is,
+
+21:47.060 --> 21:48.140
+ and life as we know it,
+
+21:48.140 --> 21:50.780
+ which we can talk about with some intelligence.
+
+21:50.780 --> 21:53.840
+ So life as we know it requires compartmentalization.
+
+21:53.840 --> 21:56.660
+ You need like a little membrane around your cell.
+
+21:56.660 --> 21:58.980
+ Metabolism, you need to take in food and eat it
+
+21:58.980 --> 22:01.020
+ and let that make you do things.
+
+22:01.020 --> 22:02.620
+ And then replication, okay?
+
+22:02.620 --> 22:04.620
+ So you need to have some information about who you are
+
+22:04.620 --> 22:07.880
+ that you pass down to future generations.
+
+22:07.880 --> 22:11.780
+ In the lab, compartmentalization seems pretty easy.
+
+22:11.780 --> 22:13.780
+ Not hard to make lipid bilayers
+
+22:13.780 --> 22:16.760
+ that come into little cellular walls pretty easily.
+
+22:16.760 --> 22:19.260
+ Metabolism and replication are hard,
+
+22:20.160 --> 22:21.900
+ but replication we're close to.
+
+22:21.900 --> 22:24.960
+ People have made RNA like molecules in the lab
+
+22:24.960 --> 22:28.840
+ that I think the state of the art is,
+
+22:28.840 --> 22:30.660
+ they're not able to make one molecule
+
+22:30.660 --> 22:32.060
+ that reproduces itself,
+
+22:32.060 --> 22:33.600
+ but they're able to make two molecules
+
+22:33.600 --> 22:35.260
+ that reproduce each other.
+
+22:35.260 --> 22:36.100
+ So that's okay.
+
+22:36.100 --> 22:37.100
+ That's pretty close.
+
+22:38.060 --> 22:41.060
+ Metabolism is harder, believe it or not,
+
+22:41.060 --> 22:42.900
+ even though it's sort of the most obvious thing,
+
+22:42.900 --> 22:44.940
+ but you want some sort of controlled metabolism
+
+22:44.940 --> 22:47.500
+ and the actual cellular machinery in our bodies
+
+22:47.500 --> 22:48.660
+ is quite complicated.
+
+22:48.660 --> 22:50.940
+ It's hard to see it just popping into existence
+
+22:50.940 --> 22:51.780
+ all by itself.
+
+22:51.780 --> 22:52.860
+ It probably took a while,
+
+22:53.740 --> 22:56.100
+ but we're making progress.
+
+22:56.100 --> 22:57.240
+ And in fact, I don't think we're spending
+
+22:57.240 --> 22:58.580
+ nearly enough money on it.
+
+22:58.580 --> 23:01.780
+ If I were the NSF, I would flood this area with money
+
+23:01.780 --> 23:05.220
+ because it would change our view of the world
+
+23:05.220 --> 23:06.780
+ if we could actually make life in the lab
+
+23:06.780 --> 23:09.420
+ and understand how it was made originally here on earth.
+
+23:09.420 --> 23:11.160
+ And I'm sure it'd have some ripple effects
+
+23:11.160 --> 23:12.940
+ that help cure disease and so on.
+
+23:12.940 --> 23:14.380
+ I mean, just that understanding.
+
+23:14.380 --> 23:16.700
+ So synthetic biology is a wonderful big frontier
+
+23:16.700 --> 23:17.980
+ where we're making cells.
+
+23:18.940 --> 23:21.100
+ Right now, the best way to do that
+
+23:21.100 --> 23:23.620
+ is to borrow heavily from existing biology, right?
+
+23:23.620 --> 23:25.380
+ Well, Craig Venter several years ago
+
+23:25.380 --> 23:28.220
+ created an artificial cell, but all he did was,
+
+23:28.220 --> 23:29.860
+ not all he did, it was a tremendous accomplishment,
+
+23:29.860 --> 23:33.180
+ but all he did was take out the DNA from a cell
+
+23:33.180 --> 23:37.200
+ and put in entirely new DNA and let it boot up and go.
+
+23:37.200 --> 23:42.200
+ What about the leap to creating intelligent life on earth?
+
+23:43.420 --> 23:44.260
+ Yeah.
+
+23:44.260 --> 23:45.860
+ Again, we define intelligence, of course,
+
+23:45.860 --> 23:49.860
+ but let's just even say Homo sapiens,
+
+23:49.860 --> 23:54.480
+ the modern intelligence in our human brain.
+
+23:55.340 --> 23:58.660
+ Do you have a sense of what's involved in that leap
+
+23:58.660 --> 24:00.420
+ and how big of a leap that is?
+
+24:00.420 --> 24:03.300
+ So AI would count in this, or do you really want life?
+
+24:03.300 --> 24:06.420
+ Do you want really an organism in some sense?
+
+24:06.420 --> 24:07.540
+ AI would count, I think.
+
+24:07.540 --> 24:08.980
+ Okay.
+
+24:08.980 --> 24:11.020
+ Yeah, of course, of course AI would count.
+
+24:11.020 --> 24:13.460
+ Well, let's say artificial consciousness, right?
+
+24:13.460 --> 24:15.500
+ So I do not think we are on the threshold
+
+24:15.500 --> 24:16.760
+ of creating artificial consciousness.
+
+24:16.760 --> 24:18.180
+ I think it's possible.
+
+24:18.180 --> 24:20.300
+ I'm not, again, very educated about how close we are,
+
+24:20.300 --> 24:22.100
+ but my impression is not that we're really close
+
+24:22.100 --> 24:24.820
+ because we understand how little we understand
+
+24:24.820 --> 24:26.460
+ of consciousness and what it is.
+
+24:26.460 --> 24:28.440
+ So if we don't have any idea what it is,
+
+24:28.440 --> 24:29.780
+ it's hard to imagine we're on the threshold
+
+24:29.780 --> 24:31.620
+ of making it ourselves.
+
+24:32.500 --> 24:34.500
+ But it's doable, it's possible.
+
+24:34.500 --> 24:35.960
+ I don't see any obstacles in principle.
+
+24:35.960 --> 24:38.160
+ So yeah, I would hold out some interest
+
+24:38.160 --> 24:40.220
+ in that happening eventually.
+
+24:40.220 --> 24:42.700
+ I think in general, consciousness,
+
+24:42.700 --> 24:44.420
+ I think we would be just surprised
+
+24:44.420 --> 24:49.060
+ how easy consciousness is once we create intelligence.
+
+24:49.060 --> 24:50.540
+ I think consciousness is a thing
+
+24:50.540 --> 24:54.000
+ that's just something we all fake.
+
+24:55.540 --> 24:56.380
+ Well, good.
+
+24:56.380 --> 24:57.680
+ No, actually, I like this idea that in fact,
+
+24:57.680 --> 25:00.500
+ consciousness is way less mysterious than we think
+
+25:00.500 --> 25:02.620
+ because we're all at every time, at every moment,
+
+25:02.620 --> 25:04.500
+ less conscious than we think we are, right?
+
+25:04.500 --> 25:05.460
+ We can fool things.
+
+25:05.460 --> 25:07.780
+ And I think that plus the idea
+
+25:07.780 --> 25:11.180
+ that you not only have artificial intelligent systems,
+
+25:11.180 --> 25:12.980
+ but you put them in a body, right,
+
+25:12.980 --> 25:14.280
+ give them a robot body,
+
+25:15.620 --> 25:18.460
+ that will help the faking a lot.
+
+25:18.460 --> 25:20.980
+ Yeah, I think creating consciousness
+
+25:20.980 --> 25:25.140
+ in artificial consciousness is as simple
+
+25:25.140 --> 25:30.020
+ as asking a Roomba to say, I'm conscious,
+
+25:30.020 --> 25:32.780
+ and refusing to be talked out of it.
+
+25:32.780 --> 25:33.820
+ Could be, it could be.
+
+25:33.820 --> 25:36.740
+ And I mean, I'm almost being silly,
+
+25:36.740 --> 25:38.280
+ but that's what we do.
+
+25:39.660 --> 25:40.940
+ That's what we do with each other.
+
+25:40.940 --> 25:42.020
+ This is the kind of,
+
+25:42.020 --> 25:44.500
+ that consciousness is also a social construct.
+
+25:44.500 --> 25:47.860
+ And a lot of our ideas of intelligence is a social construct.
+
+25:47.860 --> 25:52.820
+ And so reaching that bar involves something that's beyond,
+
+25:52.820 --> 25:54.940
+ that doesn't necessarily involve
+
+25:54.940 --> 25:57.720
+ the fundamental understanding of how you go
+
+25:57.720 --> 26:02.500
+ from electrons to neurons to cognition.
+
+26:02.500 --> 26:05.060
+ No, actually, I think that is an extremely good point.
+
+26:05.060 --> 26:08.660
+ And in fact, what it suggests is,
+
+26:08.660 --> 26:10.540
+ so yeah, you referred to Kate Darling,
+
+26:10.540 --> 26:11.940
+ who I had on the podcast,
+
+26:11.940 --> 26:16.440
+ and who does these experiments with very simple robots,
+
+26:16.440 --> 26:18.060
+ but they look like animals,
+
+26:18.060 --> 26:20.740
+ and they can look like they're experiencing pain,
+
+26:20.740 --> 26:23.380
+ and we human beings react very negatively
+
+26:23.380 --> 26:24.400
+ to these little robots
+
+26:24.400 --> 26:26.300
+ looking like they're experiencing pain.
+
+26:26.300 --> 26:29.980
+ And what you wanna say is, yeah, but they're just robots.
+
+26:29.980 --> 26:31.700
+ It's not really pain, right?
+
+26:31.700 --> 26:33.080
+ It's just some electrons going around.
+
+26:33.080 --> 26:36.300
+ But then you realize, you and I are just electrons
+
+26:36.300 --> 26:38.380
+ going around, and that's what pain is also.
+
+26:38.380 --> 26:43.060
+ And so what I would have an easy time imagining
+
+26:43.060 --> 26:44.740
+ is that there is a spectrum
+
+26:44.740 --> 26:47.420
+ between these simple little robots that Kate works with
+
+26:47.420 --> 26:49.420
+ and a human being,
+
+26:49.420 --> 26:50.940
+ where there are things that sort of
+
+26:50.940 --> 26:52.840
+ by some strict definition,
+
+26:52.840 --> 26:55.460
+ Turing test level thing are not conscious,
+
+26:55.460 --> 26:58.580
+ but nevertheless walk and talk like they're conscious.
+
+26:58.580 --> 27:00.220
+ And it could be that the future is,
+
+27:00.220 --> 27:02.460
+ I mean, Siri is close, right?
+
+27:02.460 --> 27:04.540
+ And so it might be the future
+
+27:04.540 --> 27:07.100
+ has a lot more agents like that.
+
+27:07.100 --> 27:08.860
+ And in fact, rather than someday going,
+
+27:08.860 --> 27:10.700
+ aha, we have consciousness,
+
+27:10.700 --> 27:13.180
+ we'll just creep up on it with more and more
+
+27:13.180 --> 27:15.220
+ accurate reflections of what we expect.
+
+27:15.220 --> 27:18.320
+ And in the future, maybe the present,
+
+27:18.320 --> 27:20.800
+ for example, we haven't met before,
+
+27:20.800 --> 27:25.300
+ and you're basically assuming that I'm human as it's a high
+
+27:25.300 --> 27:28.560
+ probability at this time because the yeah,
+
+27:28.560 --> 27:30.200
+ but in the future,
+
+27:30.200 --> 27:32.000
+ there might be question marks around that, right?
+
+27:32.000 --> 27:33.340
+ Yeah, no, absolutely.
+
+27:33.340 --> 27:35.740
+ Certainly videos are almost to the point
+
+27:35.740 --> 27:36.740
+ where you shouldn't trust them already.
+
+27:36.740 --> 27:39.060
+ Photos you can't trust, right?
+
+27:39.060 --> 27:41.700
+ Videos is easier to trust,
+
+27:41.700 --> 27:44.020
+ but we're getting worse that,
+
+27:44.020 --> 27:46.540
+ we're getting better at faking them, right?
+
+27:46.540 --> 27:48.780
+ Yeah, so physical embodied people,
+
+27:48.780 --> 27:51.020
+ what's so hard about faking that?
+
+27:51.020 --> 27:51.980
+ So this is very depressing,
+
+27:51.980 --> 27:53.420
+ this conversation we're having right now.
+
+27:53.420 --> 27:54.340
+ So I mean,
+
+27:54.340 --> 27:55.180
+ To me, it's exciting.
+
+27:55.180 --> 27:56.300
+ To me, you're doing it.
+
+27:56.300 --> 27:57.780
+ So it's exciting to you,
+
+27:57.780 --> 27:59.060
+ but it's a sobering thought.
+
+27:59.060 --> 28:00.420
+ We're very bad, right?
+
+28:00.420 --> 28:02.820
+ At imagining what the next 50 years are gonna be like
+
+28:02.820 --> 28:04.220
+ when we're in the middle of a phase transition
+
+28:04.220 --> 28:05.260
+ as we are right now.
+
+28:05.260 --> 28:06.740
+ Yeah, and I, in general,
+
+28:06.740 --> 28:09.220
+ I'm not blind to all the threats.
+
+28:09.220 --> 28:14.220
+ I am excited by the power of technology to solve,
+
+28:14.540 --> 28:18.060
+ to protect us against the threats as they evolve.
+
+28:18.060 --> 28:22.340
+ I'm not as much as Steven Pinker optimistic about the world,
+
+28:22.340 --> 28:23.740
+ but in everything I've seen,
+
+28:23.740 --> 28:27.300
+ all of the brilliant people in the world that I've met
+
+28:27.300 --> 28:29.160
+ are good people.
+
+28:29.160 --> 28:30.800
+ So the army of the good
+
+28:30.800 --> 28:33.400
+ in terms of the development of technology is large.
+
+28:33.400 --> 28:36.860
+ Okay, you're way more optimistic than I am.
+
+28:37.820 --> 28:39.060
+ I think that goodness and badness
+
+28:39.060 --> 28:40.900
+ are equally distributed among intelligent
+
+28:40.900 --> 28:42.700
+ and unintelligent people.
+
+28:42.700 --> 28:44.660
+ I don't see much of a correlation there.
+
+28:44.660 --> 28:46.060
+ Interesting.
+
+28:46.060 --> 28:47.300
+ Neither of us have proof.
+
+28:47.300 --> 28:48.420
+ Yeah, exactly.
+
+28:48.420 --> 28:50.660
+ Again, opinions are free, right?
+
+28:50.660 --> 28:52.540
+ Nor definitions of good and evil.
+
+28:52.540 --> 28:57.460
+ We come without definitions or without data opinions.
+
+28:57.460 --> 29:01.980
+ So what kind of questions can science not currently answer
+
+29:01.980 --> 29:04.380
+ and may never be able to answer in your view?
+
+29:04.380 --> 29:06.940
+ Well, the obvious one is what is good and bad?
+
+29:06.940 --> 29:07.860
+ What is right and wrong?
+
+29:07.860 --> 29:09.460
+ I think that there are questions that,
+
+29:09.460 --> 29:11.300
+ science tells us what happens,
+
+29:11.300 --> 29:13.260
+ what the world is and what it does.
+
+29:13.260 --> 29:14.740
+ It doesn't say what the world should do
+
+29:14.740 --> 29:15.580
+ or what we should do,
+
+29:15.580 --> 29:17.800
+ because we're part of the world.
+
+29:17.800 --> 29:19.200
+ But we are part of the world
+
+29:19.200 --> 29:21.460
+ and we have the ability to feel like something's right,
+
+29:21.460 --> 29:22.740
+ something's wrong.
+
+29:22.740 --> 29:25.660
+ And to make a very long story very short,
+
+29:25.660 --> 29:28.000
+ I think that the idea of moral philosophy
+
+29:28.000 --> 29:30.100
+ is systematizing our intuitions
+
+29:30.100 --> 29:31.700
+ of what is right and what is wrong.
+
+29:31.700 --> 29:34.580
+ And science might be able to predict ahead of time
+
+29:34.580 --> 29:36.180
+ what we will do,
+
+29:36.180 --> 29:38.000
+ but it won't ever be able to judge
+
+29:38.000 --> 29:39.600
+ whether we should have done it or not.
+
+29:39.600 --> 29:43.620
+ So, you're kind of unique in terms of scientists.
+
+29:43.620 --> 29:45.520
+ Listen, it doesn't have to do with podcasts,
+
+29:45.520 --> 29:47.660
+ but even just reaching out,
+
+29:47.660 --> 29:49.080
+ I think you referred to as sort of
+
+29:49.080 --> 29:51.300
+ doing interdisciplinary science.
+
+29:51.300 --> 29:54.100
+ So you reach out and talk to people
+
+29:54.100 --> 29:55.980
+ that are outside of your discipline,
+
+29:55.980 --> 30:00.140
+ which I always hope that's what science was for.
+
+30:00.140 --> 30:02.300
+ In fact, I was a little disillusioned
+
+30:02.300 --> 30:06.420
+ when I realized that academia is very siloed.
+
+30:06.420 --> 30:07.260
+ Yeah.
+
+30:07.260 --> 30:09.560
+ And so the question is,
+
+30:10.700 --> 30:13.020
+ how, at your own level,
+
+30:13.020 --> 30:15.380
+ how do you prepare for these conversations?
+
+30:15.380 --> 30:16.900
+ How do you think about these conversations?
+
+30:16.900 --> 30:18.300
+ How do you open your mind enough
+
+30:18.300 --> 30:20.220
+ to have these conversations?
+
+30:20.220 --> 30:21.940
+ And it may be a little bit broader,
+
+30:21.940 --> 30:24.380
+ how can you advise other scientists
+
+30:24.380 --> 30:26.260
+ to have these kinds of conversations?
+
+30:26.260 --> 30:28.180
+ Not at the podcast,
+
+30:28.180 --> 30:29.860
+ the fact that you're doing a podcast is awesome,
+
+30:29.860 --> 30:31.380
+ other people get to hear them,
+
+30:31.380 --> 30:34.700
+ but it's also good to have it without mics in general.
+
+30:34.700 --> 30:37.460
+ It's a good question, but a tough one to answer.
+
+30:37.460 --> 30:40.980
+ I think about a guy I know who's a personal trainer,
+
+30:40.980 --> 30:43.240
+ and he was asked on a podcast,
+
+30:43.240 --> 30:45.700
+ how do we psych ourselves up to do a workout?
+
+30:45.700 --> 30:48.340
+ How do we make that discipline to go and work out?
+
+30:48.340 --> 30:50.300
+ And he's like, why are you asking me?
+
+30:50.300 --> 30:52.340
+ I can't stop working out.
+
+30:52.340 --> 30:54.380
+ I don't need to psych myself up.
+
+30:54.380 --> 30:57.340
+ So, and likewise, he asked me,
+
+30:57.340 --> 30:59.740
+ how do you get to have interdisciplinary conversations
+
+30:59.740 --> 31:00.700
+ on all sorts of different things,
+
+31:00.700 --> 31:01.660
+ all sorts of different people?
+
+31:01.660 --> 31:04.860
+ I'm like, that's what makes me go, right?
+
+31:04.860 --> 31:07.380
+ Like that's, I couldn't stop doing that.
+
+31:07.380 --> 31:09.660
+ I did that long before any of them were recorded.
+
+31:09.660 --> 31:12.380
+ In fact, a lot of the motivation for starting recording it
+
+31:12.380 --> 31:14.420
+ was making sure I would read all these books
+
+31:14.420 --> 31:15.460
+ that I had purchased, right?
+
+31:15.460 --> 31:17.700
+ Like all these books I wanted to read,
+
+31:17.700 --> 31:18.900
+ not enough time to read them.
+
+31:18.900 --> 31:20.700
+ And now if I have the motivation,
+
+31:20.700 --> 31:23.220
+ cause I'm gonna interview Pat Churchland,
+
+31:23.220 --> 31:25.180
+ I'm gonna finally read her book.
+
+31:25.180 --> 31:29.460
+ You know, and it's absolutely true
+
+31:29.460 --> 31:31.700
+ that academia is extraordinarily siloed, right?
+
+31:31.700 --> 31:32.780
+ We don't talk to people.
+
+31:32.780 --> 31:34.260
+ We rarely do.
+
+31:34.260 --> 31:36.460
+ And in fact, when we do, it's punished.
+
+31:36.460 --> 31:38.820
+ You know, like the people who do it successfully
+
+31:38.820 --> 31:41.420
+ generally first became very successful
+
+31:41.420 --> 31:43.100
+ within their little siloed discipline.
+
+31:43.100 --> 31:46.380
+ And only then did they start expanding out.
+
+31:46.380 --> 31:47.660
+ If you're a young person, you know,
+
+31:47.660 --> 31:48.940
+ I have graduate students.
+
+31:48.940 --> 31:52.980
+ I try to be very, very candid with them about this,
+
+31:52.980 --> 31:55.580
+ that it's, you know, most graduate students
+
+31:55.580 --> 31:57.420
+ are to not become faculty members, right?
+
+31:57.420 --> 31:59.020
+ It's a tough road.
+
+31:59.020 --> 32:03.140
+ And so live the life you wanna live,
+
+32:03.140 --> 32:04.620
+ but do it with your eyes open
+
+32:04.620 --> 32:06.900
+ about what it does to your job chances.
+
+32:06.900 --> 32:09.580
+ And the more broad you are
+
+32:09.580 --> 32:12.900
+ and the less time you spend hyper specializing
+
+32:12.900 --> 32:15.780
+ in your field, the lower your job chances are.
+
+32:15.780 --> 32:17.060
+ That's just an academic reality.
+
+32:17.060 --> 32:20.060
+ It's terrible, I don't like it, but it's a reality.
+
+32:20.060 --> 32:22.540
+ And for some people, that's fine.
+
+32:22.540 --> 32:24.660
+ Like there's plenty of people who are wonderful scientists
+
+32:24.660 --> 32:27.140
+ who have zero interest in branching out
+
+32:27.140 --> 32:30.740
+ and talking to things, to anyone outside their field.
+
+32:30.740 --> 32:33.740
+ But it is disillusioning to me.
+
+32:33.740 --> 32:36.180
+ Some of the, you know, romantic notion I had
+
+32:36.180 --> 32:38.220
+ of the intellectual academic life
+
+32:38.220 --> 32:39.940
+ is belied by the reality of it.
+
+32:39.940 --> 32:43.500
+ The idea that we should reach out beyond our discipline
+
+32:43.500 --> 32:48.500
+ and that is a positive good is just so rare
+
+32:48.500 --> 32:53.500
+ in universities that it may as well not exist at all.
+
+32:53.900 --> 32:57.660
+ But that said, even though you're saying you're doing it
+
+32:57.660 --> 33:00.300
+ like the personal trainer, because you just can't help it,
+
+33:00.300 --> 33:02.940
+ you're also an inspiration to others.
+
+33:02.940 --> 33:04.980
+ Like I could speak for myself.
+
+33:05.780 --> 33:09.540
+ You know, I also have a career I'm thinking about, right?
+
+33:09.540 --> 33:12.060
+ And without your podcast,
+
+33:12.060 --> 33:15.060
+ I may have not have been doing this at all, right?
+
+33:15.060 --> 33:19.540
+ So it makes me realize that these kinds of conversations
+
+33:19.540 --> 33:23.340
+ is kind of what science is about in many ways.
+
+33:23.340 --> 33:26.500
+ The reason we write papers, this exchange of ideas,
+
+33:27.460 --> 33:30.540
+ is it's much harder to do interdisciplinary papers,
+
+33:30.540 --> 33:31.380
+ I would say.
+
+33:31.380 --> 33:35.140
+ And conversations are easier.
+
+33:35.140 --> 33:36.820
+ So conversations is the beginning.
+
+33:36.820 --> 33:41.180
+ And in the field of AI, it's obvious
+
+33:41.180 --> 33:45.580
+ that we should think outside of pure computer vision
+
+33:45.580 --> 33:47.540
+ competitions on a particular data sets.
+
+33:47.540 --> 33:49.660
+ We should think about the broader impact
+
+33:49.660 --> 33:53.740
+ of how this can be, you know, reaching out to physics,
+
+33:53.740 --> 33:57.220
+ to psychology, to neuroscience and having these
+
+33:57.220 --> 34:00.580
+ conversations so that you're an inspiration.
+
+34:00.580 --> 34:05.220
+ And so never know how the world changes.
+
+34:05.220 --> 34:08.540
+ I mean, the fact that this stuff is out there
+
+34:08.540 --> 34:12.300
+ and I've a huge number of people come up to me,
+
+34:12.300 --> 34:16.100
+ grad students, really loving the podcast, inspired by it.
+
+34:16.100 --> 34:18.660
+ And they will probably have that,
+
+34:18.660 --> 34:20.740
+ they'll be ripple effects when they become faculty
+
+34:20.740 --> 34:21.580
+ and so on and so on.
+
+34:21.580 --> 34:25.300
+ We can end on a balance between pessimism and optimism.
+
+34:25.300 --> 34:27.780
+ And Sean, thank you so much for talking to me, it was awesome.
+
+34:27.780 --> 34:29.460
+ No, Lex, thank you very much for this conversation.
+
+34:29.460 --> 34:49.460
+ It was great.
+

vtt/episode_026_small.vtt

@@ -0,0 +1,2567 @@
+WEBVTT
+
+00:00.000 --> 00:02.760
+ The following is a conversation with Sean Carroll.
+
+00:02.760 --> 00:04.920
+ He's a theoretical physicist at Caltech,
+
+00:04.920 --> 00:08.800
+ specializing in quantum mechanics, gravity, and cosmology.
+
+00:08.800 --> 00:11.640
+ He's the author of several popular books,
+
+00:11.640 --> 00:15.360
+ one on the Arrow of Time called From Eternity to Hear,
+
+00:15.360 --> 00:17.840
+ one on the Higgs Boson called Particle
+
+00:17.840 --> 00:19.160
+ at the End of the Universe,
+
+00:19.160 --> 00:22.560
+ and one on Science and Philosophy called The Big Picture,
+
+00:22.560 --> 00:26.360
+ on the origins of life, meaning, and the universe itself.
+
+00:26.360 --> 00:28.720
+ He has an upcoming book on quantum mechanics
+
+00:28.720 --> 00:32.760
+ that you can preorder now called Something Deeply Hidden.
+
+00:32.760 --> 00:36.040
+ He writes one of my favorite blogs on his website,
+
+00:36.040 --> 00:37.960
+ preposterousuniverse.com.
+
+00:37.960 --> 00:40.440
+ I recommend clicking on the greatest hits link
+
+00:40.440 --> 00:44.400
+ that lists accessible, interesting posts on the Arrow of Time,
+
+00:44.400 --> 00:47.600
+ dark matter, dark energy, the Big Bang, general relativity,
+
+00:47.600 --> 00:49.560
+ string theory, quantum mechanics,
+
+00:49.560 --> 00:53.160
+ and the big meta questions about the philosophy of science,
+
+00:53.160 --> 00:57.600
+ God, ethics, politics, academia, and much, much more.
+
+00:57.600 --> 01:00.280
+ Finally, and perhaps most famously,
+
+01:00.280 --> 01:03.640
+ he's the host of a podcast called Mindscape
+
+01:03.640 --> 01:06.920
+ that you should subscribe to and support on Patreon.
+
+01:06.920 --> 01:08.800
+ Along with the Joe Rogan experience,
+
+01:08.800 --> 01:10.480
+ Sam Harris is Making Sense,
+
+01:10.480 --> 01:13.080
+ and Dan Carlin's Hardcore History.
+
+01:13.080 --> 01:15.840
+ Sean's Mindscape podcast is one of my favorite ways
+
+01:15.840 --> 01:18.800
+ to learn new ideas or explore different perspectives
+
+01:18.800 --> 01:22.120
+ and ideas that I thought I understood.
+
+01:22.120 --> 01:25.880
+ It was truly an honor to meet and spend a couple hours
+
+01:25.880 --> 01:27.200
+ with Sean.
+
+01:27.200 --> 01:30.480
+ It's a bit heartbreaking to say that for the first time ever,
+
+01:30.480 --> 01:32.760
+ the audio recorder for this podcast died
+
+01:32.760 --> 01:34.880
+ in the middle of our conversation.
+
+01:34.880 --> 01:36.280
+ There are technical reasons for this,
+
+01:36.280 --> 01:38.360
+ having to do with phantom power
+
+01:38.360 --> 01:41.040
+ that I now understand and will avoid.
+
+01:41.040 --> 01:44.200
+ It took me one hour to notice and fix the problem.
+
+01:44.200 --> 01:48.280
+ So, much like the universe's 68% dark energy,
+
+01:48.280 --> 01:51.280
+ roughly the same amount from this conversation was lost,
+
+01:51.280 --> 01:54.160
+ except in the memories of the two people involved
+
+01:54.160 --> 01:56.280
+ and in my notes.
+
+01:56.280 --> 01:59.920
+ I'm sure we'll talk again and continue this conversation
+
+01:59.920 --> 02:02.440
+ on this podcast or on Sean's.
+
+02:02.440 --> 02:05.320
+ And of course, I look forward to it.
+
+02:05.320 --> 02:07.840
+ This is the Artificial Intelligence podcast.
+
+02:07.840 --> 02:09.960
+ If you enjoy it, subscribe on YouTube,
+
+02:09.960 --> 02:12.520
+ iTunes, support on Patreon,
+
+02:12.520 --> 02:16.680
+ or simply connect with me on Twitter at Lex Freedman.
+
+02:16.680 --> 02:21.360
+ And now, here's my conversation with Sean Carroll.
+
+02:21.360 --> 02:23.520
+ What do you think is more interesting and impactful?
+
+02:23.520 --> 02:25.840
+ Understanding how the universe works
+
+02:25.840 --> 02:26.880
+ at a fundamental level
+
+02:26.880 --> 02:29.200
+ or understanding how the human mind works?
+
+02:29.200 --> 02:32.440
+ You know, of course this is a crazy meaningless
+
+02:32.440 --> 02:33.960
+ unanswerable question in some sense,
+
+02:33.960 --> 02:35.160
+ because they're both very interesting
+
+02:35.160 --> 02:37.520
+ and there's no absolute scale of interestingness
+
+02:37.520 --> 02:39.160
+ that we can rate them on.
+
+02:39.160 --> 02:41.160
+ There's a glib answer that says the human brain
+
+02:41.160 --> 02:43.080
+ is part of the universe, right?
+
+02:43.080 --> 02:44.400
+ And therefore, understanding the universe
+
+02:44.400 --> 02:47.000
+ is more fundamental than understanding the human brain.
+
+02:47.000 --> 02:49.600
+ But do you really believe that once we understand
+
+02:49.600 --> 02:51.520
+ the fundamental way the universe works
+
+02:51.520 --> 02:52.680
+ at the particle level,
+
+02:52.680 --> 02:55.800
+ the forces we would be able to understand how the mind works?
+
+02:55.800 --> 02:56.640
+ No, certainly not.
+
+02:56.640 --> 02:58.760
+ We cannot understand how ice cream works
+
+02:58.760 --> 03:01.040
+ just from understanding how particles work, right?
+
+03:01.040 --> 03:02.760
+ So I'm a big believer in emergence.
+
+03:02.760 --> 03:05.320
+ I'm a big believer that there are different ways
+
+03:05.320 --> 03:06.640
+ of talking about the world
+
+03:07.880 --> 03:11.200
+ beyond just the most fundamental microscopic one.
+
+03:11.200 --> 03:13.880
+ You know, when we talk about tables and chairs
+
+03:13.880 --> 03:15.120
+ and planets and people,
+
+03:15.120 --> 03:16.400
+ we're not talking the language
+
+03:16.400 --> 03:18.360
+ of particle physics and cosmology.
+
+03:18.360 --> 03:20.880
+ So, but understanding the universe,
+
+03:20.880 --> 03:24.040
+ you didn't say just at the most fundamental level, right?
+
+03:24.040 --> 03:28.200
+ So understanding the universe at all levels is part of that.
+
+03:28.200 --> 03:29.960
+ I do think, you know, to be a little bit more fair
+
+03:29.960 --> 03:33.960
+ to the question, there probably are general principles
+
+03:33.960 --> 03:38.520
+ of complexity, biology, information processing,
+
+03:38.520 --> 03:41.840
+ memory, knowledge, creativity
+
+03:41.840 --> 03:45.600
+ that go beyond just the human brain, right?
+
+03:45.600 --> 03:47.800
+ And maybe one could count understanding those
+
+03:47.800 --> 03:49.120
+ as part of understanding the universe.
+
+03:49.120 --> 03:53.040
+ The human brain, as far as we know, is the most complex thing
+
+03:53.040 --> 03:54.320
+ in the universe.
+
+03:54.320 --> 03:57.440
+ So there's, it's certainly absurd to think
+
+03:57.440 --> 03:58.880
+ that by understanding the fundamental laws
+
+03:58.880 --> 04:00.400
+ of particle physics,
+
+04:00.400 --> 04:02.880
+ you get any direct insight on how the brain works.
+
+04:02.880 --> 04:04.360
+ But then there's this step
+
+04:04.360 --> 04:06.840
+ from the fundamentals of particle physics
+
+04:06.840 --> 04:08.680
+ to information processing,
+
+04:08.680 --> 04:10.840
+ which a lot of physicists and philosophers
+
+04:10.840 --> 04:12.520
+ may be a little bit carelessly take
+
+04:12.520 --> 04:14.680
+ when they talk about artificial intelligence.
+
+04:14.680 --> 04:18.080
+ Do you think of the universe
+
+04:18.080 --> 04:21.360
+ as a kind of a computational device?
+
+04:21.360 --> 04:22.200
+ No.
+
+04:22.200 --> 04:24.200
+ To be like the honest answer there is no.
+
+04:24.200 --> 04:27.600
+ There's a sense in which the universe processes information
+
+04:27.600 --> 04:29.200
+ clearly.
+
+04:29.200 --> 04:32.720
+ There's a sense in which the universe is like a computer,
+
+04:32.720 --> 04:33.920
+ clearly.
+
+04:33.920 --> 04:36.560
+ But in some sense, I think,
+
+04:36.560 --> 04:38.560
+ I tried to say this once on my blog
+
+04:38.560 --> 04:39.400
+ and no one agreed with me,
+
+04:39.400 --> 04:42.400
+ but the universe is more like a computation
+
+04:42.400 --> 04:45.080
+ than a computer because the universe happens once.
+
+04:45.080 --> 04:46.960
+ A computer is a general purpose machine, right?
+
+04:46.960 --> 04:48.680
+ You can ask it different questions,
+
+04:48.680 --> 04:50.120
+ even a pocket calculator, right?
+
+04:50.120 --> 04:52.960
+ And it's set up to answer certain kinds of questions.
+
+04:52.960 --> 04:54.320
+ The universe isn't that.
+
+04:54.320 --> 04:57.360
+ So information processing happens in the universe,
+
+04:57.360 --> 04:59.120
+ but it's not what the universe is.
+
+04:59.120 --> 05:01.560
+ And I know your MIT colleague, Seth Lloyd,
+
+05:01.560 --> 05:03.840
+ feels very differently about this, right?
+
+05:03.840 --> 05:07.240
+ Well, you're thinking of the universe as a closed system.
+
+05:07.240 --> 05:08.080
+ I am.
+
+05:08.080 --> 05:11.760
+ So what makes a computer more like a PC,
+
+05:13.000 --> 05:14.560
+ like a computing machine,
+
+05:14.560 --> 05:17.720
+ is that there's a human that comes up to it
+
+05:17.720 --> 05:19.120
+ and moves the mouse around,
+
+05:19.120 --> 05:21.680
+ so input gives it input.
+
+05:21.680 --> 05:26.320
+ And that's why you're saying it's just a computation,
+
+05:26.320 --> 05:29.280
+ a deterministic thing that's just unrolling.
+
+05:29.280 --> 05:32.240
+ But the immense complexity of it
+
+05:32.240 --> 05:34.440
+ is nevertheless like processing.
+
+05:34.440 --> 05:39.440
+ There's a state and it changes with rules.
+
+05:40.160 --> 05:41.680
+ And there's a sense for a lot of people
+
+05:41.680 --> 05:45.400
+ that if the brain operates, the human brain operates
+
+05:45.400 --> 05:46.520
+ within that world,
+
+05:46.520 --> 05:49.400
+ then it's simply just a small subset of that.
+
+05:49.400 --> 05:52.560
+ And so there's no reason we can't build
+
+05:52.560 --> 05:55.600
+ arbitrarily great intelligences.
+
+05:55.600 --> 05:56.440
+ Yeah.
+
+05:56.440 --> 05:58.720
+ Do you think of intelligence in this way?
+
+05:58.720 --> 05:59.640
+ Intelligence is tricky.
+
+05:59.640 --> 06:01.720
+ I don't have a definition of it offhand.
+
+06:01.720 --> 06:04.640
+ So I remember this panel discussion
+
+06:04.640 --> 06:06.240
+ that I saw on YouTube, I wasn't there,
+
+06:06.240 --> 06:07.720
+ but Seth Lloyd was on the panel.
+
+06:07.720 --> 06:10.560
+ And so was Martin Rees, the famous astrophysicist.
+
+06:10.560 --> 06:13.800
+ And Seth gave his shtick for why the universe is a computer
+
+06:13.800 --> 06:14.840
+ and explained this.
+
+06:14.840 --> 06:19.360
+ And Martin Rees said, so what is not a computer?
+
+06:19.360 --> 06:22.000
+ And Seth is like, oh, that's a good question.
+
+06:22.000 --> 06:22.840
+ I'm not sure.
+
+06:22.840 --> 06:24.960
+ Because if you have a sufficiently broad definition
+
+06:24.960 --> 06:28.360
+ of what a computer is, then everything is, right?
+
+06:28.360 --> 06:32.160
+ And similarly, or the analogy gains force
+
+06:32.160 --> 06:34.560
+ when it excludes some things.
+
+06:34.560 --> 06:38.640
+ Is the moon going around the earth performing a computation?
+
+06:38.640 --> 06:41.320
+ I can come up with definitions in which the answer is yes,
+
+06:41.320 --> 06:43.840
+ but it's not a very useful computation.
+
+06:43.840 --> 06:46.120
+ I think that it's absolutely helpful
+
+06:46.120 --> 06:49.600
+ to think about the universe in certain situations,
+
+06:49.600 --> 06:53.080
+ certain contexts, as an information processing device.
+
+06:53.080 --> 06:54.840
+ I'm even guilty of writing a paper
+
+06:54.840 --> 06:56.960
+ called Quantum Circuit Cosmology, where
+
+06:56.960 --> 06:59.280
+ we modeled the whole universe as a quantum circuit.
+
+06:59.280 --> 07:00.320
+ As a circuit.
+
+07:00.320 --> 07:01.440
+ As a circuit, yeah.
+
+07:01.440 --> 07:02.840
+ With qubits kind of thing.
+
+07:02.840 --> 07:05.000
+ With qubits, basically, right.
+
+07:05.000 --> 07:07.400
+ So in qubits, becoming more and more entangled.
+
+07:07.400 --> 07:09.640
+ So do we want to digress a little bit?
+
+07:09.640 --> 07:11.000
+ Because this is kind of fun.
+
+07:11.000 --> 07:13.680
+ So here's a mystery about the universe
+
+07:13.680 --> 07:16.840
+ that is so deep and profound that nobody talks about it.
+
+07:16.840 --> 07:19.040
+ Space expands, right?
+
+07:19.040 --> 07:21.880
+ And we talk about, in a certain region of space,
+
+07:21.880 --> 07:23.560
+ a certain number of degrees of freedom,
+
+07:23.560 --> 07:25.480
+ a certain number of ways that the quantum fields
+
+07:25.480 --> 07:28.800
+ and the particles in that region can arrange themselves.
+
+07:28.800 --> 07:32.200
+ That number of degrees of freedom in a region of space
+
+07:32.200 --> 07:33.800
+ is arguably finite.
+
+07:33.800 --> 07:36.640
+ We actually don't know how many there are,
+
+07:36.640 --> 07:39.440
+ but there's a very good argument that says it's a finite number.
+
+07:39.440 --> 07:44.920
+ So as the universe expands and space gets bigger,
+
+07:44.920 --> 07:46.560
+ are there more degrees of freedom?
+
+07:46.560 --> 07:48.520
+ If it's an infinite number, it doesn't really matter.
+
+07:48.520 --> 07:50.000
+ Infinity times 2 is still infinity.
+
+07:50.000 --> 07:53.160
+ But if it's a finite number, then there's more space,
+
+07:53.160 --> 07:54.480
+ so there's more degrees of freedom.
+
+07:54.480 --> 07:55.760
+ So where did they come from?
+
+07:55.760 --> 07:58.000
+ That would mean the universe is not a closed system.
+
+07:58.000 --> 08:01.520
+ There's more degrees of freedom popping into existence.
+
+08:01.520 --> 08:05.320
+ So what we suggested was that there are more degrees of freedom.
+
+08:05.320 --> 08:07.960
+ And it's not that they're not there to start,
+
+08:07.960 --> 08:10.880
+ but they're not entangled to start.
+
+08:10.880 --> 08:12.800
+ So the universe that you and I know of,
+
+08:12.800 --> 08:15.440
+ the three dimensions around us that we see,
+
+08:15.440 --> 08:18.080
+ we said those are the entangled degrees of freedom
+
+08:18.080 --> 08:19.640
+ making up space time.
+
+08:19.640 --> 08:22.640
+ As the universe expands, there are a whole bunch of qubits
+
+08:22.640 --> 08:26.840
+ in their zero state that become entangled
+
+08:26.840 --> 08:28.720
+ with the rest of space time through the action
+
+08:28.720 --> 08:31.200
+ of these quantum circuits.
+
+08:31.200 --> 08:37.080
+ So what does it mean that there's now more degrees of freedom
+
+08:37.080 --> 08:39.280
+ as they become more entangled?
+
+08:39.280 --> 08:40.280
+ Yeah.
+
+08:40.280 --> 08:41.640
+ As the universe expands.
+
+08:41.640 --> 08:41.960
+ That's right.
+
+08:41.960 --> 08:43.280
+ So there's more and more degrees of freedom
+
+08:43.280 --> 08:47.320
+ that are entangled, that are playing the role of part
+
+08:47.320 --> 08:49.600
+ of the entangled space time structure.
+
+08:49.600 --> 08:53.320
+ So the underlying philosophy is that space time itself
+
+08:53.320 --> 08:55.600
+ arises from the entanglement of some fundamental quantum
+
+08:55.600 --> 08:57.680
+ degrees of freedom.
+
+08:57.680 --> 08:58.280
+ Wow.
+
+08:58.280 --> 08:59.780
+ OK.
+
+08:59.780 --> 09:05.200
+ At which point is most of the entanglement happening?
+
+09:05.200 --> 09:07.400
+ Are we talking about close to the Big Bang?
+
+09:07.400 --> 09:11.840
+ Are we talking about throughout the time of the life of the
+
+09:11.840 --> 09:12.340
+ universe?
+
+09:12.340 --> 09:12.840
+ Yeah.
+
+09:12.840 --> 09:15.080
+ So the idea is that at the Big Bang,
+
+09:15.080 --> 09:17.760
+ almost all the degrees of freedom that the universe could
+
+09:17.760 --> 09:22.400
+ have were there, but they were unentangled with anything else.
+
+09:22.400 --> 09:23.840
+ And that's a reflection of the fact
+
+09:23.840 --> 09:25.560
+ that the Big Bang had a low entropy.
+
+09:25.560 --> 09:28.080
+ It was a very simple, very small place.
+
+09:28.080 --> 09:31.360
+ And as space expands, more and more degrees of freedom
+
+09:31.360 --> 09:34.240
+ become entangled with the rest of the world.
+
+09:34.240 --> 09:35.960
+ Well, I have to ask John Carroll,
+
+09:35.960 --> 09:38.160
+ what do you think of the thought experiment from Nick
+
+09:38.160 --> 09:41.560
+ Bostrom that we're living in a simulation?
+
+09:41.560 --> 09:44.880
+ So I think let me contextualize that a little bit more.
+
+09:44.880 --> 09:48.320
+ I think people don't actually take this thought experiment.
+
+09:48.320 --> 09:50.360
+ I think it's quite interesting.
+
+09:50.360 --> 09:52.880
+ It's not very useful, but it's quite interesting.
+
+09:52.880 --> 09:55.440
+ From the perspective of AI, a lot of the learning
+
+09:55.440 --> 09:59.280
+ that can be done usually happens in simulation,
+
+09:59.280 --> 10:01.440
+ artificial examples.
+
+10:01.440 --> 10:03.040
+ And so it's a constructive question
+
+10:03.040 --> 10:09.360
+ to ask how difficult is our real world to simulate,
+
+10:09.360 --> 10:12.400
+ which is kind of a dual part of, if we're
+
+10:12.400 --> 10:16.400
+ living in a simulation and somebody built that simulation,
+
+10:16.400 --> 10:18.840
+ if you were to try to do it yourself, how hard would it be?
+
+10:18.840 --> 10:21.080
+ So obviously, we could be living in a simulation.
+
+10:21.080 --> 10:22.960
+ If you just want the physical possibility,
+
+10:22.960 --> 10:25.360
+ then I completely agree that it's physically possible.
+
+10:25.360 --> 10:27.360
+ I don't think that we actually are.
+
+10:27.360 --> 10:31.880
+ So take this one piece of data into consideration.
+
+10:31.880 --> 10:35.080
+ We live in a big universe.
+
+10:35.080 --> 10:38.480
+ There's two trillion galaxies in our observable universe
+
+10:38.480 --> 10:41.640
+ with 200 billion stars in each galaxy, et cetera.
+
+10:41.640 --> 10:44.920
+ It would seem to be a waste of resources
+
+10:44.920 --> 10:47.600
+ to have a universe that big going on just to do a simulation.
+
+10:47.600 --> 10:50.120
+ So in other words, I want to be a good Bayesian.
+
+10:50.120 --> 10:54.920
+ I want to ask, under this hypothesis, what do I expect to see?
+
+10:54.920 --> 10:56.080
+ So the first thing I would say is I
+
+10:56.080 --> 11:00.280
+ wouldn't expect to see a universe that was that big.
+
+11:00.280 --> 11:02.560
+ The second thing is I wouldn't expect the resolution
+
+11:02.560 --> 11:05.000
+ of the universe to be as good as it is.
+
+11:05.000 --> 11:08.960
+ So it's always possible that if our superhuman simulators only
+
+11:08.960 --> 11:10.840
+ have finite resources that they don't render
+
+11:10.840 --> 11:14.360
+ the entire universe, that the part that is out there,
+
+11:14.360 --> 11:17.040
+ the two trillion galaxies, isn't actually
+
+11:17.040 --> 11:19.600
+ being simulated fully.
+
+11:19.600 --> 11:22.720
+ But then the obvious extrapolation of that
+
+11:22.720 --> 11:25.640
+ is that only I am being simulated fully.
+
+11:25.640 --> 11:29.240
+ The rest of you are just nonplayer characters.
+
+11:29.240 --> 11:30.520
+ I'm the only thing that is real.
+
+11:30.520 --> 11:32.720
+ The rest of you are just chatbots.
+
+11:32.720 --> 11:34.320
+ Beyond this wall, I see the wall,
+
+11:34.320 --> 11:37.360
+ but there is literally nothing on the other side of the wall.
+
+11:37.360 --> 11:39.000
+ That is sort of the Bayesian prediction.
+
+11:39.000 --> 11:40.400
+ That's what it would be like to do
+
+11:40.400 --> 11:42.240
+ an efficient simulation of me.
+
+11:42.240 --> 11:45.760
+ So none of that seems quite realistic.
+
+11:45.760 --> 11:50.880
+ I don't see, I hear the argument that it's just possible
+
+11:50.880 --> 11:53.280
+ and easy to simulate lots of things.
+
+11:53.280 --> 11:57.280
+ I don't see any evidence from what we know about our universe
+
+11:57.280 --> 11:59.280
+ that we look like a simulated universe.
+
+11:59.280 --> 12:01.120
+ Now, maybe you can say, well, we don't know what it would
+
+12:01.120 --> 12:03.000
+ look like, but that's just abandoning
+
+12:03.000 --> 12:04.520
+ your Bayesian responsibilities.
+
+12:04.520 --> 12:07.680
+ Like your job is to say, under this theory,
+
+12:07.680 --> 12:09.480
+ here's what you would expect to see.
+
+12:09.480 --> 12:11.680
+ Yeah, so certainly if you think about a simulation
+
+12:11.680 --> 12:16.680
+ as a thing that's like a video game where only a small subset
+
+12:16.680 --> 12:22.880
+ is being readied, but say all the laws of physics,
+
+12:22.880 --> 12:26.560
+ the entire closed system of the quote unquote universe,
+
+12:26.560 --> 12:27.800
+ it had a creator.
+
+12:27.800 --> 12:29.640
+ Yeah, it's always possible.
+
+12:29.640 --> 12:32.280
+ So that's not useful to think about
+
+12:32.280 --> 12:34.040
+ when you're thinking about physics.
+
+12:34.040 --> 12:38.080
+ The way Nick Bostrom phrases it, if it's possible
+
+12:38.080 --> 12:40.520
+ to simulate a universe, eventually we'll do it.
+
+12:40.520 --> 12:42.720
+ Right.
+
+12:42.720 --> 12:45.560
+ You can use that, by the way, for a lot of things.
+
+12:45.560 --> 12:49.840
+ But I guess the question is, how hard is it
+
+12:49.840 --> 12:52.320
+ to create a universe?
+
+12:52.320 --> 12:53.800
+ I wrote a little blog post about this,
+
+12:53.800 --> 12:55.440
+ and maybe I'm missing something.
+
+12:55.440 --> 12:57.680
+ But there's an argument that says not only
+
+12:57.680 --> 13:00.480
+ that it might be possible to simulate a universe,
+
+13:00.480 --> 13:05.400
+ but probably, if you imagine that you actually
+
+13:05.400 --> 13:07.320
+ attribute consciousness and agency
+
+13:07.320 --> 13:09.920
+ to the little things that we're simulating,
+
+13:09.920 --> 13:12.400
+ to our little artificial beings, there's probably
+
+13:12.400 --> 13:15.000
+ a lot more of them than there are ordinary organic beings
+
+13:15.000 --> 13:17.400
+ in the universe, or there will be in the future.
+
+13:17.400 --> 13:19.600
+ So there's an argument that not only is being a simulation
+
+13:19.600 --> 13:23.520
+ possible, it's probable, because in the space
+
+13:23.520 --> 13:25.480
+ of all living consciousnesses, most of them
+
+13:25.480 --> 13:26.600
+ are being simulated.
+
+13:26.600 --> 13:28.840
+ Most of them are not at the top level.
+
+13:28.840 --> 13:30.520
+ I think that argument must be wrong,
+
+13:30.520 --> 13:34.080
+ because it follows from that argument that if we're simulated,
+
+13:34.080 --> 13:36.880
+ but we can also simulate other things.
+
+13:36.880 --> 13:38.800
+ Well, but if we can simulate other things,
+
+13:38.800 --> 13:41.800
+ they can simulate other things.
+
+13:41.800 --> 13:44.280
+ If we give them enough power and resolution,
+
+13:44.280 --> 13:46.000
+ and ultimately, we'll reach a bottom,
+
+13:46.000 --> 13:47.800
+ because the laws of physics in our universe
+
+13:47.800 --> 13:51.120
+ have a bottom, we're made of atoms and so forth.
+
+13:51.120 --> 13:55.080
+ So there will be the cheapest possible simulations.
+
+13:55.080 --> 13:57.680
+ And if you believe the original argument,
+
+13:57.680 --> 13:59.920
+ you should conclude that we should be in the cheapest
+
+13:59.920 --> 14:02.560
+ possible simulation, because that's where most people are.
+
+14:02.560 --> 14:03.640
+ But we don't look like that.
+
+14:03.640 --> 14:06.880
+ It doesn't look at all like we're at the edge of resolution,
+
+14:06.880 --> 14:09.960
+ that we're 16 bit things.
+
+14:09.960 --> 14:12.840
+ It seems much easier to make much lower level things
+
+14:12.840 --> 14:14.160
+ than we are.
+
+14:14.160 --> 14:18.200
+ So, and also, I question the whole approach
+
+14:18.200 --> 14:19.840
+ to the anthropic principle that says
+
+14:19.840 --> 14:22.320
+ we are typical observers in the universe.
+
+14:22.320 --> 14:23.640
+ I think that that's not actually,
+
+14:23.640 --> 14:27.320
+ I think that there's a lot of selection that we can do
+
+14:27.320 --> 14:30.120
+ that were typical within things we already know,
+
+14:30.120 --> 14:32.240
+ but not typical within all the universe.
+
+14:32.240 --> 14:35.760
+ So do you think there is intelligent life,
+
+14:35.760 --> 14:37.800
+ however you would like to define intelligent life
+
+14:37.800 --> 14:39.920
+ out there in the universe?
+
+14:39.920 --> 14:44.640
+ My guess is that there is not intelligent life
+
+14:44.640 --> 14:46.840
+ in the observable universe other than us.
+
+14:48.320 --> 14:52.480
+ Simply on the basis of the fact that the likely number
+
+14:52.480 --> 14:56.320
+ of other intelligent species in the observable universe,
+
+14:56.320 --> 15:00.280
+ there's two likely numbers, zero or billions.
+
+15:01.480 --> 15:02.560
+ And if there had been billions,
+
+15:02.560 --> 15:04.000
+ you would have noticed already.
+
+15:05.040 --> 15:07.320
+ For there to be literally like a small number,
+
+15:07.320 --> 15:12.320
+ like Star Trek, there's a dozen intelligent civilizations
+
+15:12.440 --> 15:15.040
+ in our galaxy, but not a billion.
+
+15:16.240 --> 15:18.480
+ That's weird, that's sort of bizarre to me.
+
+15:18.480 --> 15:21.000
+ It's easy for me to imagine that there are zero others
+
+15:21.000 --> 15:22.600
+ because there's just a big bottleneck
+
+15:22.600 --> 15:24.960
+ to making multicellular life
+
+15:24.960 --> 15:27.040
+ or technological life or whatever.
+
+15:27.040 --> 15:28.560
+ It's very hard for me to imagine
+
+15:28.560 --> 15:30.160
+ that there's a whole bunch out there
+
+15:30.160 --> 15:32.280
+ that have somehow remained hidden from us.
+
+15:32.280 --> 15:34.880
+ The question I'd like to ask is,
+
+15:34.880 --> 15:37.240
+ what would intelligent life look like?
+
+15:37.240 --> 15:41.120
+ What I mean by that question and where it's going is,
+
+15:41.120 --> 15:45.120
+ what if intelligent life is just fundamentally,
+
+15:45.120 --> 15:49.120
+ in some very big ways, different than the one
+
+15:49.120 --> 15:51.480
+ that has on Earth.
+
+15:51.480 --> 15:53.880
+ That there's all kinds of intelligent life
+
+15:53.880 --> 15:57.560
+ that operates at different scales of both size and temporal.
+
+15:57.560 --> 15:59.280
+ That's a great possibility
+
+15:59.280 --> 16:00.800
+ because I think we should be humble
+
+16:00.800 --> 16:02.640
+ about what intelligence is, what life is.
+
+16:02.640 --> 16:04.040
+ We don't even agree on what life is,
+
+16:04.040 --> 16:06.040
+ much less what intelligent life is, right?
+
+16:06.040 --> 16:08.200
+ So that's an argument for humility,
+
+16:08.200 --> 16:10.080
+ saying there could be intelligent life
+
+16:10.080 --> 16:12.800
+ of a very different character, right?
+
+16:12.800 --> 16:17.240
+ You could imagine that dolphins are intelligent
+
+16:17.240 --> 16:19.760
+ but never invent space travel
+
+16:19.760 --> 16:20.760
+ because they live in the ocean
+
+16:20.760 --> 16:22.760
+ and they don't have thumbs, right?
+
+16:22.760 --> 16:25.840
+ So they never invent technology, they never invent smelting.
+
+16:26.840 --> 16:31.200
+ Maybe the universe is full of intelligent species
+
+16:31.200 --> 16:33.200
+ that just don't make technology, right?
+
+16:33.200 --> 16:35.440
+ That's compatible with the data, I think.
+
+16:35.440 --> 16:38.560
+ And I think maybe what you're pointing at
+
+16:38.560 --> 16:42.560
+ is even more out there versions of intelligence,
+
+16:42.560 --> 16:46.240
+ you know, intelligence in intermolecular clouds
+
+16:46.240 --> 16:48.160
+ or on the surface of a neutron star
+
+16:48.160 --> 16:50.360
+ or in between the galaxies in giant things
+
+16:50.360 --> 16:52.840
+ where the equivalent of a heartbeat is 100 million years.
+
+16:54.840 --> 16:56.760
+ On the one hand, yes,
+
+16:56.760 --> 16:58.560
+ we should be very open minded about those things.
+
+16:58.560 --> 17:03.560
+ On the other hand, we all of us share the same laws of physics.
+
+17:03.560 --> 17:07.040
+ There might be something about the laws of physics
+
+17:07.040 --> 17:08.560
+ even though we don't currently know exactly
+
+17:08.560 --> 17:12.560
+ what that thing would be that makes meters
+
+17:12.560 --> 17:16.560
+ and years the right length and time scales
+
+17:16.560 --> 17:19.560
+ for intelligent life, maybe not.
+
+17:19.560 --> 17:22.560
+ But we're made of atoms, atoms have a certain size,
+
+17:22.560 --> 17:25.560
+ we orbit stars, our stars have a certain lifetime.
+
+17:25.560 --> 17:28.560
+ It's not impossible to me that there's a sweet spot
+
+17:28.560 --> 17:30.560
+ for intelligent life that we find ourselves in.
+
+17:30.560 --> 17:33.560
+ So I'm open minded either way, I'm open minded either being humble
+
+17:33.560 --> 17:35.560
+ and there's all sorts of different kinds of life
+
+17:35.560 --> 17:37.560
+ or no, there's a reason we just don't know it yet
+
+17:37.560 --> 17:40.560
+ why life like ours is the kind of life that's out there.
+
+17:40.560 --> 17:43.560
+ Yeah, I'm of two minds too, but I often wonder
+
+17:43.560 --> 17:48.560
+ if our brains is just designed to, quite obviously,
+
+17:48.560 --> 17:53.560
+ to operate and see the world on these time scales.
+
+17:53.560 --> 17:57.560
+ And we're almost blind and the tools we've created
+
+17:57.560 --> 18:01.560
+ for detecting things are blind to the kind of observation
+
+18:01.560 --> 18:04.560
+ needed to see intelligent life at other scales.
+
+18:04.560 --> 18:06.560
+ Well, I'm totally open to that,
+
+18:06.560 --> 18:08.560
+ but so here's another argument I would make.
+
+18:08.560 --> 18:10.560
+ We have looked for intelligent life,
+
+18:10.560 --> 18:13.560
+ but we've looked at for it in the dumbest way we can
+
+18:13.560 --> 18:15.560
+ by turning radio telescopes to the sky.
+
+18:15.560 --> 18:20.560
+ And why in the world would a super advanced civilization
+
+18:20.560 --> 18:23.560
+ randomly beam out radio signals wastefully
+
+18:23.560 --> 18:25.560
+ in all directions into the universe?
+
+18:25.560 --> 18:28.560
+ It just doesn't make any sense, especially because
+
+18:28.560 --> 18:30.560
+ in order to think that you would actually contact
+
+18:30.560 --> 18:33.560
+ another civilization, you would have to do it forever.
+
+18:33.560 --> 18:35.560
+ You have to keep doing it for millions of years.
+
+18:35.560 --> 18:37.560
+ That sounds like a waste of resources.
+
+18:37.560 --> 18:42.560
+ If you thought that there were other solar systems
+
+18:42.560 --> 18:45.560
+ with planets around them where maybe intelligent life
+
+18:45.560 --> 18:48.560
+ didn't yet exist, but might someday,
+
+18:48.560 --> 18:51.560
+ you wouldn't try to talk to it with radio waves.
+
+18:51.560 --> 18:53.560
+ You would send a spacecraft out there
+
+18:53.560 --> 18:55.560
+ and you would park it around there.
+
+18:55.560 --> 18:57.560
+ And it would be like, from our point of view,
+
+18:57.560 --> 19:00.560
+ it would be like 2001 where there was a monolith.
+
+19:00.560 --> 19:02.560
+ There could be an artifact.
+
+19:02.560 --> 19:04.560
+ In fact, the other way works also, right?
+
+19:04.560 --> 19:07.560
+ There could be artifacts in our solar system
+
+19:07.560 --> 19:11.560
+ that have been put there by other technologically advanced
+
+19:11.560 --> 19:14.560
+ civilizations, and that's how we will eventually contact them.
+
+19:14.560 --> 19:16.560
+ We just haven't explored the solar system well enough yet
+
+19:16.560 --> 19:18.560
+ to find them.
+
+19:18.560 --> 19:20.560
+ The reason why we don't think about that is because
+
+19:20.560 --> 19:21.560
+ we're young and impatient, right?
+
+19:21.560 --> 19:23.560
+ It's like it would take more than my lifetime
+
+19:23.560 --> 19:25.560
+ to actually send something to another star system
+
+19:25.560 --> 19:27.560
+ and wait for it and then come back.
+
+19:27.560 --> 19:30.560
+ But if we start thinking on hundreds of thousands of years
+
+19:30.560 --> 19:32.560
+ or a million year time scales,
+
+19:32.560 --> 19:34.560
+ that's clearly the right thing to do.
+
+19:34.560 --> 19:38.560
+ Are you excited by the thing that Elon Musk is doing with SpaceX
+
+19:38.560 --> 19:41.560
+ in general, but the idea of space exploration,
+
+19:41.560 --> 19:45.560
+ even though you're species is young and impatient?
+
+19:45.560 --> 19:50.560
+ No, I do think that space travel is crucially important, long term.
+
+19:50.560 --> 19:52.560
+ Even to other star systems.
+
+19:52.560 --> 19:57.560
+ And I think that many people overestimate the difficulty
+
+19:57.560 --> 20:00.560
+ because they say, look, if you travel 1% the speed of light
+
+20:00.560 --> 20:03.560
+ to another star system, we'll be dead before we get there, right?
+
+20:03.560 --> 20:05.560
+ And I think that it's much easier.
+
+20:05.560 --> 20:07.560
+ And therefore, when they write their science fiction stories,
+
+20:07.560 --> 20:09.560
+ they imagine we'd go faster than the speed of light
+
+20:09.560 --> 20:11.560
+ because otherwise they're too impatient, right?
+
+20:11.560 --> 20:13.560
+ We're not going to go faster than the speed of light,
+
+20:13.560 --> 20:15.560
+ but we could easily imagine that the human lifespan
+
+20:15.560 --> 20:17.560
+ gets extended to thousands of years.
+
+20:17.560 --> 20:19.560
+ And once you do that, then the stars are much closer.
+
+20:19.560 --> 20:20.560
+ Effectively, right?
+
+20:20.560 --> 20:22.560
+ What's 100 year trip, right?
+
+20:22.560 --> 20:26.560
+ So I think that that's going to be the future, the far future,
+
+20:26.560 --> 20:29.560
+ not my lifetime once again, but baby steps.
+
+20:29.560 --> 20:31.560
+ Unless your lifetime gets extended.
+
+20:31.560 --> 20:33.560
+ Well, it's in a race against time, right?
+
+20:33.560 --> 20:37.560
+ A friend of mine who actually thinks about these things said,
+
+20:37.560 --> 20:39.560
+ you know, you and I are going to die,
+
+20:39.560 --> 20:42.560
+ but I don't know about our grandchildren.
+
+20:42.560 --> 20:45.560
+ I don't know, predicting the future is hard,
+
+20:45.560 --> 20:47.560
+ but that's the least plausible scenario.
+
+20:47.560 --> 20:51.560
+ And so, yeah, no, I think that as we discussed earlier,
+
+20:51.560 --> 20:56.560
+ there are threats to the earth, known and unknown, right?
+
+20:56.560 --> 21:02.560
+ Having spread humanity and biology elsewhere
+
+21:02.560 --> 21:04.560
+ is a really important longterm goal.
+
+21:04.560 --> 21:08.560
+ What kind of questions can science not currently answer,
+
+21:08.560 --> 21:11.560
+ but might soon?
+
+21:11.560 --> 21:14.560
+ When you think about the problems and the mysteries before us,
+
+21:14.560 --> 21:17.560
+ that may be within reach of science.
+
+21:17.560 --> 21:19.560
+ I think an obvious one is the origin of life.
+
+21:19.560 --> 21:21.560
+ We don't know how that happened.
+
+21:21.560 --> 21:24.560
+ There's a difficulty in knowing how it happened historically,
+
+21:24.560 --> 21:26.560
+ actually, you know, literally on earth,
+
+21:26.560 --> 21:29.560
+ but starting life from nonlife
+
+21:29.560 --> 21:32.560
+ is something I kind of think we're close to, right?
+
+21:32.560 --> 21:33.560
+ You really think so?
+
+21:33.560 --> 21:35.560
+ Like, how difficult is it to start life?
+
+21:35.560 --> 21:36.560
+ I do.
+
+21:36.560 --> 21:40.560
+ Well, I've talked to people, including on the podcast, about this.
+
+21:40.560 --> 21:42.560
+ You know, life requires three things.
+
+21:42.560 --> 21:44.560
+ Life as we know it.
+
+21:44.560 --> 21:46.560
+ There's a difference between life, who knows what it is,
+
+21:46.560 --> 21:47.560
+ and life as we know it,
+
+21:47.560 --> 21:50.560
+ which we can talk about with some intelligence.
+
+21:50.560 --> 21:53.560
+ Life as we know it requires compartmentalization.
+
+21:53.560 --> 21:56.560
+ You need a little membrane around your cell.
+
+21:56.560 --> 21:58.560
+ Metabolism, you need to take in food and eat it
+
+21:58.560 --> 22:00.560
+ and let that make you do things.
+
+22:00.560 --> 22:02.560
+ And then replication.
+
+22:02.560 --> 22:04.560
+ You need to have some information about who you are,
+
+22:04.560 --> 22:07.560
+ that you pass down to future generations.
+
+22:07.560 --> 22:11.560
+ In the lab, compartmentalization seems pretty easy,
+
+22:11.560 --> 22:13.560
+ not hard to make lipid bilayers
+
+22:13.560 --> 22:16.560
+ that come into little cellular walls pretty easily.
+
+22:16.560 --> 22:19.560
+ Metabolism and replication are hard,
+
+22:19.560 --> 22:21.560
+ but replication we're close to.
+
+22:21.560 --> 22:25.560
+ People have made RNA like molecules in the lab that...
+
+22:25.560 --> 22:28.560
+ I think the state of the art is
+
+22:28.560 --> 22:31.560
+ they're not able to make one molecule that reproduces itself,
+
+22:31.560 --> 22:34.560
+ but they're able to make two molecules that reproduce each other.
+
+22:34.560 --> 22:37.560
+ So that's okay. That's pretty close.
+
+22:37.560 --> 22:40.560
+ Metabolism is harder, believe it or not,
+
+22:40.560 --> 22:42.560
+ even though it's sort of the most obvious thing,
+
+22:42.560 --> 22:44.560
+ but you want some sort of controlled metabolism
+
+22:44.560 --> 22:48.560
+ and the actual cellular machinery in our bodies is quite complicated.
+
+22:48.560 --> 22:51.560
+ It's hard to see it just popping into existence all by itself.
+
+22:51.560 --> 22:53.560
+ It probably took a while.
+
+22:53.560 --> 22:55.560
+ But we're making progress.
+
+22:55.560 --> 22:58.560
+ In fact, I don't think we're spending nearly enough money on it.
+
+22:58.560 --> 23:01.560
+ If I were the NSF, I would flood this area with money
+
+23:01.560 --> 23:04.560
+ because it would change our view of the world
+
+23:04.560 --> 23:06.560
+ if we could actually make life in the lab
+
+23:06.560 --> 23:09.560
+ and understand how it was made originally here on Earth.
+
+23:09.560 --> 23:11.560
+ I'm sure it would have some ripple effects
+
+23:11.560 --> 23:13.560
+ that help cure diseases and so on.
+
+23:13.560 --> 23:15.560
+ That's right.
+
+23:15.560 --> 23:18.560
+ Synthetic biology is a wonderful big frontier where we're making cells.
+
+23:18.560 --> 23:21.560
+ Right now, the best way to do that
+
+23:21.560 --> 23:23.560
+ is to borrow heavily from existing biology.
+
+23:23.560 --> 23:26.560
+ Craig Ventner several years ago created an artificial cell,
+
+23:26.560 --> 23:28.560
+ but all he did was...
+
+23:28.560 --> 23:30.560
+ not all he did, it was a tremendous accomplishment,
+
+23:30.560 --> 23:33.560
+ but all he did was take out the DNA from a cell
+
+23:33.560 --> 23:36.560
+ and put in entirely new DNA and let it boot up and go.
+
+23:36.560 --> 23:43.560
+ What about the leap to creating intelligent life on Earth?
+
+23:43.560 --> 23:45.560
+ However, again, we define intelligence, of course,
+
+23:45.560 --> 23:49.560
+ but let's just even say homo sapiens,
+
+23:49.560 --> 23:54.560
+ the modern intelligence in our human brain.
+
+23:54.560 --> 23:58.560
+ Do you have a sense of what's involved in that leap
+
+23:58.560 --> 24:00.560
+ and how big of a leap that is?
+
+24:00.560 --> 24:02.560
+ So AI would count in this?
+
+24:02.560 --> 24:04.560
+ Or do you really want life?
+
+24:04.560 --> 24:06.560
+ AI would count in some sense.
+
+24:06.560 --> 24:08.560
+ AI would count, I think.
+
+24:08.560 --> 24:10.560
+ Of course, AI would count.
+
+24:10.560 --> 24:12.560
+ Well, let's say artificial consciousness.
+
+24:12.560 --> 24:14.560
+ I do not think we are on the threshold
+
+24:14.560 --> 24:16.560
+ of creating artificial consciousness.
+
+24:16.560 --> 24:18.560
+ I think it's possible.
+
+24:18.560 --> 24:20.560
+ I'm not, again, very educated about how close we are,
+
+24:20.560 --> 24:22.560
+ but my impression is not that we're really close
+
+24:22.560 --> 24:24.560
+ because we understand how little we understand
+
+24:24.560 --> 24:26.560
+ of consciousness and what it is.
+
+24:26.560 --> 24:28.560
+ So if we don't have any idea what it is,
+
+24:28.560 --> 24:30.560
+ it's hard to imagine we're on the threshold
+
+24:30.560 --> 24:32.560
+ of making it ourselves.
+
+24:32.560 --> 24:34.560
+ But it's doable, it's possible.
+
+24:34.560 --> 24:36.560
+ I don't see any obstacles in principle,
+
+24:36.560 --> 24:38.560
+ so yeah, I would hold out some interest
+
+24:38.560 --> 24:40.560
+ in that happening eventually.
+
+24:40.560 --> 24:42.560
+ I think in general, consciousness,
+
+24:42.560 --> 24:44.560
+ I think it would be just surprised
+
+24:44.560 --> 24:46.560
+ how easy consciousness is
+
+24:46.560 --> 24:48.560
+ once we create intelligence.
+
+24:48.560 --> 24:50.560
+ I think consciousness is a thing
+
+24:50.560 --> 24:54.560
+ that's just something we all fake.
+
+24:54.560 --> 24:56.560
+ Well, good.
+
+24:56.560 --> 24:58.560
+ No, actually, I like this idea that, in fact,
+
+24:58.560 --> 25:00.560
+ consciousness is way less mysterious than we think
+
+25:00.560 --> 25:02.560
+ because we're all at every time,
+
+25:02.560 --> 25:04.560
+ at every moment, less conscious than we think we are.
+
+25:04.560 --> 25:06.560
+ We can fool things.
+
+25:06.560 --> 25:08.560
+ And I think that plus the idea that you
+
+25:08.560 --> 25:10.560
+ not only have artificial intelligence systems,
+
+25:10.560 --> 25:12.560
+ but you put them in a body,
+
+25:12.560 --> 25:14.560
+ give them a robot body,
+
+25:14.560 --> 25:18.560
+ that will help the faking a lot.
+
+25:18.560 --> 25:20.560
+ Yeah, I think creating consciousness
+
+25:20.560 --> 25:22.560
+ in artificial consciousness
+
+25:22.560 --> 25:24.560
+ is as simple
+
+25:24.560 --> 25:26.560
+ as asking a Roomba
+
+25:26.560 --> 25:28.560
+ to say, I'm conscious
+
+25:28.560 --> 25:32.560
+ and refusing to be talked out of it.
+
+25:32.560 --> 25:34.560
+ It could be.
+
+25:34.560 --> 25:36.560
+ I mean, I'm almost being silly,
+
+25:36.560 --> 25:38.560
+ but that's what we do.
+
+25:38.560 --> 25:40.560
+ That's what we do with each other.
+
+25:40.560 --> 25:44.560
+ The consciousness is also a social construct,
+
+25:44.560 --> 25:46.560
+ and a lot of our ideas of intelligence
+
+25:46.560 --> 25:48.560
+ is a social construct,
+
+25:48.560 --> 25:50.560
+ and so reaching that bar involves
+
+25:50.560 --> 25:52.560
+ something that's beyond,
+
+25:52.560 --> 25:54.560
+ that doesn't necessarily involve
+
+25:54.560 --> 25:56.560
+ the fundamental understanding
+
+25:56.560 --> 25:58.560
+ of how you go from
+
+25:58.560 --> 26:00.560
+ electrons to neurons
+
+26:00.560 --> 26:02.560
+ to cognition.
+
+26:02.560 --> 26:04.560
+ No, actually, I think that is an extremely good point,
+
+26:04.560 --> 26:06.560
+ and in fact,
+
+26:06.560 --> 26:08.560
+ what it suggests is,
+
+26:08.560 --> 26:10.560
+ so yeah, you referred to Kate Darling,
+
+26:10.560 --> 26:12.560
+ who I had on the podcast,
+
+26:12.560 --> 26:14.560
+ and who does these experiments with
+
+26:14.560 --> 26:16.560
+ very simple robots,
+
+26:16.560 --> 26:18.560
+ but they look like animals,
+
+26:18.560 --> 26:20.560
+ and they can look like they're experiencing pain,
+
+26:20.560 --> 26:22.560
+ and we human beings react
+
+26:22.560 --> 26:24.560
+ very negatively to these little robots
+
+26:24.560 --> 26:26.560
+ looking like they're experiencing pain,
+
+26:26.560 --> 26:28.560
+ and what you want to say is,
+
+26:28.560 --> 26:30.560
+ yeah, but they're just robots.
+
+26:30.560 --> 26:32.560
+ It's not really pain.
+
+26:32.560 --> 26:34.560
+ It's just some electrons going around,
+
+26:34.560 --> 26:36.560
+ but then you realize you and I
+
+26:36.560 --> 26:38.560
+ are just electrons going around,
+
+26:38.560 --> 26:40.560
+ and that's what pain is also.
+
+26:40.560 --> 26:42.560
+ What I would have an easy time imagining
+
+26:42.560 --> 26:44.560
+ is that there is a spectrum
+
+26:44.560 --> 26:46.560
+ between these simple little robots
+
+26:46.560 --> 26:48.560
+ that Kate works with
+
+26:48.560 --> 26:50.560
+ and a human being,
+
+26:50.560 --> 26:52.560
+ where there are things that,
+
+26:52.560 --> 26:54.560
+ like a human touring test level thing
+
+26:54.560 --> 26:56.560
+ are not conscious,
+
+26:56.560 --> 26:58.560
+ but nevertheless walk and talk
+
+26:58.560 --> 27:00.560
+ like they're conscious,
+
+27:00.560 --> 27:02.560
+ and it could be that the future is,
+
+27:02.560 --> 27:04.560
+ I mean, Siri is close, right?
+
+27:04.560 --> 27:06.560
+ And so it might be the future
+
+27:06.560 --> 27:08.560
+ has a lot more agents like that,
+
+27:08.560 --> 27:10.560
+ and in fact, rather than someday going,
+
+27:10.560 --> 27:12.560
+ aha, we have consciousness,
+
+27:12.560 --> 27:14.560
+ we'll just creep up on it
+
+27:14.560 --> 27:16.560
+ with more and more accurate reflections
+
+27:16.560 --> 27:18.560
+ of what we expect.
+
+27:18.560 --> 27:20.560
+ And in the future, maybe the present,
+
+27:20.560 --> 27:22.560
+ and you're basically assuming
+
+27:22.560 --> 27:24.560
+ that I'm human.
+
+27:24.560 --> 27:26.560
+ I get a high probability.
+
+27:26.560 --> 27:28.560
+ At this time, because the,
+
+27:28.560 --> 27:30.560
+ but in the future,
+
+27:30.560 --> 27:32.560
+ there might be question marks around that, right?
+
+27:32.560 --> 27:34.560
+ Yeah, no, absolutely.
+
+27:34.560 --> 27:36.560
+ Certainly videos are almost to the point
+
+27:36.560 --> 27:38.560
+ where you shouldn't trust them already.
+
+27:38.560 --> 27:40.560
+ Photos you can't trust, right?
+
+27:40.560 --> 27:42.560
+ Videos is easier to trust,
+
+27:42.560 --> 27:44.560
+ but we're getting worse.
+
+27:44.560 --> 27:46.560
+ We're getting better at faking them, right?
+
+27:46.560 --> 27:48.560
+ Yeah, so physical, embodied people,
+
+27:48.560 --> 27:50.560
+ what's so hard about faking that?
+
+27:50.560 --> 27:52.560
+ This is very depressing,
+
+27:52.560 --> 27:54.560
+ this conversation we're having right now.
+
+27:54.560 --> 27:56.560
+ To me, it's exciting.
+
+27:56.560 --> 27:58.560
+ You're doing it, so it's exciting to you,
+
+27:58.560 --> 28:00.560
+ but it's a sobering thought.
+
+28:00.560 --> 28:02.560
+ We're very bad at imagining
+
+28:02.560 --> 28:04.560
+ what the next 50 years are going to be like
+
+28:04.560 --> 28:06.560
+ when we're in the middle of a phase transition
+
+28:06.560 --> 28:08.560
+ as we are right now.
+
+28:08.560 --> 28:10.560
+ Yeah, and in general,
+
+28:10.560 --> 28:12.560
+ I'm not blind to all the threats.
+
+28:12.560 --> 28:14.560
+ I am excited by the power of technology
+
+28:14.560 --> 28:16.560
+ to solve,
+
+28:16.560 --> 28:18.560
+ as they evolve.
+
+28:18.560 --> 28:20.560
+ I'm not as much as Steven Pinker
+
+28:20.560 --> 28:22.560
+ optimistic about the world,
+
+28:22.560 --> 28:24.560
+ but in everything I've seen,
+
+28:24.560 --> 28:26.560
+ all the brilliant people in the world
+
+28:26.560 --> 28:28.560
+ that I've met are good people.
+
+28:28.560 --> 28:30.560
+ So the army of the good
+
+28:30.560 --> 28:32.560
+ in terms of the development of technology is large.
+
+28:32.560 --> 28:34.560
+ Okay, you're way more
+
+28:34.560 --> 28:36.560
+ optimistic than I am.
+
+28:36.560 --> 28:38.560
+ I think that goodness and badness
+
+28:38.560 --> 28:40.560
+ are equally distributed among intelligent
+
+28:40.560 --> 28:42.560
+ and unintelligent people.
+
+28:42.560 --> 28:44.560
+ I don't see much of a correlation there.
+
+28:44.560 --> 28:46.560
+ Interesting.
+
+28:46.560 --> 28:48.560
+ Neither of us have proof.
+
+28:48.560 --> 28:50.560
+ Yeah, exactly. Again, opinions are free, right?
+
+28:50.560 --> 28:52.560
+ Nor definitions of good and evil.
+
+28:52.560 --> 28:54.560
+ Without definitions
+
+28:54.560 --> 28:56.560
+ or without data
+
+28:56.560 --> 28:58.560
+ opinions.
+
+28:58.560 --> 29:00.560
+ So what kind of questions can science not
+
+29:00.560 --> 29:02.560
+ currently answer
+
+29:02.560 --> 29:04.560
+ and may never be able to answer in your view?
+
+29:04.560 --> 29:06.560
+ Well, the obvious one is what is good and bad.
+
+29:06.560 --> 29:08.560
+ What is right and wrong?
+
+29:08.560 --> 29:10.560
+ I think that there are questions that science tells us
+
+29:10.560 --> 29:12.560
+ what happens, what the world is,
+
+29:12.560 --> 29:14.560
+ doesn't say what the world should do
+
+29:14.560 --> 29:16.560
+ or what we should do because we're part of the world.
+
+29:16.560 --> 29:18.560
+ But we are part of the world
+
+29:18.560 --> 29:20.560
+ and we have the ability to feel like
+
+29:20.560 --> 29:22.560
+ something's right, something's wrong.
+
+29:22.560 --> 29:24.560
+ And to make a very long story
+
+29:24.560 --> 29:26.560
+ very short, I think that the idea
+
+29:26.560 --> 29:28.560
+ of moral philosophy is
+
+29:28.560 --> 29:30.560
+ systematizing our intuitions of what is right
+
+29:30.560 --> 29:32.560
+ and what is wrong.
+
+29:32.560 --> 29:34.560
+ And science might be able to predict ahead of time
+
+29:34.560 --> 29:36.560
+ what we will do,
+
+29:36.560 --> 29:38.560
+ but it won't ever be able to judge
+
+29:38.560 --> 29:40.560
+ whether we should have done it or not.
+
+29:40.560 --> 29:42.560
+ You know, you're kind of unique in terms of scientists.
+
+29:42.560 --> 29:44.560
+ It doesn't
+
+29:44.560 --> 29:46.560
+ have to do with podcasts, but
+
+29:46.560 --> 29:48.560
+ even just reaching out, I think you refer to
+
+29:48.560 --> 29:50.560
+ as sort of doing interdisciplinary science.
+
+29:50.560 --> 29:52.560
+ So you reach out
+
+29:52.560 --> 29:54.560
+ and talk to people
+
+29:54.560 --> 29:56.560
+ that are outside of your discipline,
+
+29:56.560 --> 29:58.560
+ which I always
+
+29:58.560 --> 30:00.560
+ hope that's what science was for.
+
+30:00.560 --> 30:02.560
+ In fact, I was a little disillusioned
+
+30:02.560 --> 30:04.560
+ when I realized that academia
+
+30:04.560 --> 30:06.560
+ is very siloed.
+
+30:06.560 --> 30:08.560
+ Yeah.
+
+30:08.560 --> 30:10.560
+ The question is
+
+30:10.560 --> 30:12.560
+ how,
+
+30:12.560 --> 30:14.560
+ at your own level, how do you prepare for these conversations?
+
+30:14.560 --> 30:16.560
+ How do you think about these conversations?
+
+30:16.560 --> 30:18.560
+ How do you open your mind enough
+
+30:18.560 --> 30:20.560
+ to have these conversations?
+
+30:20.560 --> 30:22.560
+ And it may be a little bit broader.
+
+30:22.560 --> 30:24.560
+ How can you advise other scientists
+
+30:24.560 --> 30:26.560
+ to have these kinds of conversations?
+
+30:26.560 --> 30:28.560
+ Not at the podcast.
+
+30:28.560 --> 30:30.560
+ The fact that you're doing a podcast is awesome.
+
+30:30.560 --> 30:32.560
+ Other people get to hear them.
+
+30:32.560 --> 30:34.560
+ But it's also good to have it without mics in general.
+
+30:34.560 --> 30:36.560
+ It's a good question, but a tough one
+
+30:36.560 --> 30:38.560
+ to answer. I think about
+
+30:38.560 --> 30:40.560
+ a guy I know is a personal trainer
+
+30:40.560 --> 30:42.560
+ and he was asked on a podcast
+
+30:42.560 --> 30:44.560
+ how do we psych ourselves up
+
+30:44.560 --> 30:46.560
+ to do a workout? How do we make
+
+30:46.560 --> 30:48.560
+ that discipline to go and work out?
+
+30:48.560 --> 30:50.560
+ And he's like, why are you asking me?
+
+30:50.560 --> 30:52.560
+ I can't stop working out.
+
+30:52.560 --> 30:54.560
+ I don't need to psych myself up.
+
+30:54.560 --> 30:56.560
+ Likewise, you asked me
+
+30:56.560 --> 30:58.560
+ how do you get to have
+
+30:58.560 --> 31:00.560
+ interdisciplinary conversations and all sorts of different things
+
+31:00.560 --> 31:02.560
+ with all sorts of different people?
+
+31:02.560 --> 31:04.560
+ That's what makes me go.
+
+31:04.560 --> 31:06.560
+ I couldn't stop
+
+31:06.560 --> 31:08.560
+ doing that. I did that long before
+
+31:08.560 --> 31:10.560
+ any of them were recorded. In fact,
+
+31:10.560 --> 31:12.560
+ a lot of the motivation for starting recording it
+
+31:12.560 --> 31:14.560
+ was making sure I would read all these books
+
+31:14.560 --> 31:16.560
+ that I had purchased. All these books
+
+31:16.560 --> 31:18.560
+ I wanted to read. Not enough time to read them.
+
+31:18.560 --> 31:20.560
+ And now, if I have the motivation
+
+31:20.560 --> 31:22.560
+ because I'm going to interview Pat
+
+31:22.560 --> 31:24.560
+ Churchland, I'm going to finally read her
+
+31:24.560 --> 31:26.560
+ book.
+
+31:26.560 --> 31:28.560
+ And
+
+31:28.560 --> 31:30.560
+ it's absolutely true that academia is
+
+31:30.560 --> 31:32.560
+ extraordinarily siloed. We don't talk to people.
+
+31:32.560 --> 31:34.560
+ We rarely do.
+
+31:34.560 --> 31:36.560
+ And in fact, when we do, it's punished.
+
+31:36.560 --> 31:38.560
+ The people who do it successfully
+
+31:38.560 --> 31:40.560
+ generally first became
+
+31:40.560 --> 31:42.560
+ very successful within their little siloed discipline.
+
+31:42.560 --> 31:44.560
+ And only then
+
+31:44.560 --> 31:46.560
+ did they start expanding out.
+
+31:46.560 --> 31:48.560
+ If you're a young person, I have graduate students
+
+31:48.560 --> 31:50.560
+ and I try to be very, very
+
+31:50.560 --> 31:52.560
+ candid with them about this.
+
+31:52.560 --> 31:54.560
+ That it's
+
+31:54.560 --> 31:56.560
+ most graduate students do not become faculty members.
+
+31:56.560 --> 31:58.560
+ It's a tough road.
+
+31:58.560 --> 32:00.560
+ And so
+
+32:00.560 --> 32:02.560
+ you live the life you want to live
+
+32:02.560 --> 32:04.560
+ but do it with your eyes open
+
+32:04.560 --> 32:06.560
+ about what it does to your job chances.
+
+32:06.560 --> 32:08.560
+ And the more
+
+32:08.560 --> 32:10.560
+ broad you are and the less
+
+32:10.560 --> 32:12.560
+ time you spend hyper
+
+32:12.560 --> 32:14.560
+ specializing in your field, the lower
+
+32:14.560 --> 32:16.560
+ your job chances are. That's just an academic
+
+32:16.560 --> 32:18.560
+ reality. It's terrible. I don't like it.
+
+32:18.560 --> 32:20.560
+ But it's a reality.
+
+32:20.560 --> 32:22.560
+ And for some people
+
+32:22.560 --> 32:24.560
+ that's fine. Like there's plenty of people
+
+32:24.560 --> 32:26.560
+ who are wonderful scientists who have zero
+
+32:26.560 --> 32:28.560
+ interest in branching out and talking to
+
+32:28.560 --> 32:30.560
+ things to anyone outside their field.
+
+32:30.560 --> 32:32.560
+ But
+
+32:32.560 --> 32:34.560
+ it is disillusioning to me
+
+32:34.560 --> 32:36.560
+ some of the romantic notion
+
+32:36.560 --> 32:38.560
+ I had of the intellectual academic life
+
+32:38.560 --> 32:40.560
+ is belied by the reality
+
+32:40.560 --> 32:42.560
+ of it. The idea that we should
+
+32:42.560 --> 32:44.560
+ reach out beyond our discipline
+
+32:44.560 --> 32:46.560
+ and that is a positive good
+
+32:46.560 --> 32:48.560
+ is just so
+
+32:48.560 --> 32:50.560
+ rare in
+
+32:50.560 --> 32:52.560
+ universities that it may as well
+
+32:52.560 --> 32:54.560
+ not exist at all. But
+
+32:54.560 --> 32:56.560
+ that said, even though you're saying
+
+32:56.560 --> 32:58.560
+ you're doing it like the personal trainer
+
+32:58.560 --> 33:00.560
+ because you just can't help it, you're also
+
+33:00.560 --> 33:02.560
+ an inspiration to others.
+
+33:02.560 --> 33:04.560
+ Like I could speak for myself.
+
+33:04.560 --> 33:06.560
+ You know,
+
+33:06.560 --> 33:08.560
+ I also have a career I'm thinking about
+
+33:08.560 --> 33:10.560
+ right. And without
+
+33:10.560 --> 33:12.560
+ your podcast, I may have
+
+33:12.560 --> 33:14.560
+ not have been doing this at all.
+
+33:14.560 --> 33:16.560
+ Right. So it
+
+33:16.560 --> 33:18.560
+ makes me realize that these kinds
+
+33:18.560 --> 33:20.560
+ of conversations is kind of what science is about.
+
+33:20.560 --> 33:22.560
+ In many
+
+33:22.560 --> 33:24.560
+ ways. The reason we write papers
+
+33:24.560 --> 33:26.560
+ this exchange of ideas
+
+33:26.560 --> 33:28.560
+ is much harder to do
+
+33:28.560 --> 33:30.560
+ into the disciplinary papers, I would say.
+
+33:30.560 --> 33:32.560
+ Yeah. Right.
+
+33:32.560 --> 33:34.560
+ And conversations are easier.
+
+33:34.560 --> 33:36.560
+ So conversations is the beginning
+
+33:36.560 --> 33:38.560
+ and in the field of AI
+
+33:38.560 --> 33:40.560
+ that it's
+
+33:40.560 --> 33:42.560
+ obvious that we should think outside
+
+33:42.560 --> 33:44.560
+ of pure
+
+33:44.560 --> 33:46.560
+ computer vision competitions and in particular
+
+33:46.560 --> 33:48.560
+ data sets. We should think about the broader
+
+33:48.560 --> 33:50.560
+ impact of how this can be
+
+33:50.560 --> 33:52.560
+ you know, reaching
+
+33:52.560 --> 33:54.560
+ out to physics, to psychology
+
+33:54.560 --> 33:56.560
+ to neuroscience
+
+33:56.560 --> 33:58.560
+ and having these conversations.
+
+33:58.560 --> 34:00.560
+ So you're an inspiration
+
+34:00.560 --> 34:02.560
+ and so. Well, thank you very much.
+
+34:02.560 --> 34:04.560
+ Never know how the world
+
+34:04.560 --> 34:06.560
+ changes. I mean
+
+34:06.560 --> 34:08.560
+ the fact that this stuff is out there
+
+34:08.560 --> 34:10.560
+ and I've
+
+34:10.560 --> 34:12.560
+ a huge number of people come up to me
+
+34:12.560 --> 34:14.560
+ grad students really loving the
+
+34:14.560 --> 34:16.560
+ podcast inspired by it and
+
+34:16.560 --> 34:18.560
+ they will probably have that
+
+34:18.560 --> 34:20.560
+ there'll be ripple effects when they become faculty
+
+34:20.560 --> 34:22.560
+ and so on. So we can end
+
+34:22.560 --> 34:24.560
+ on a balance between pessimism
+
+34:24.560 --> 34:26.560
+ and optimism and Sean, thank you so much
+
+34:26.560 --> 34:28.560
+ for talking. It was awesome. No, Lex, thank you very
+
+34:28.560 --> 34:52.560
+ much for this conversation. It was great.
+

vtt/episode_028_large.vtt

@@ -0,0 +1,2705 @@
+WEBVTT
+
+00:00.000 --> 00:03.120
+ The following is a conversation with Chris Sampson.
+
+00:03.120 --> 00:06.000
+ He was a CTO of the Google self driving car team,
+
+00:06.000 --> 00:08.880
+ a key engineer and leader behind the Carnegie Mellon
+
+00:08.880 --> 00:12.000
+ University autonomous vehicle entries in the DARPA Grand
+
+00:12.000 --> 00:16.160
+ Challenges and the winner of the DARPA Urban Challenge.
+
+00:16.160 --> 00:20.100
+ Today, he's the CEO of Aurora Innovation, an autonomous
+
+00:20.100 --> 00:21.360
+ vehicle software company.
+
+00:21.360 --> 00:23.600
+ He started with Sterling Anderson,
+
+00:23.600 --> 00:25.960
+ who was the former director of Tesla Autopilot,
+
+00:25.960 --> 00:30.120
+ and drew back now, Uber's former autonomy and perception lead.
+
+00:30.120 --> 00:32.880
+ Chris is one of the top roboticists and autonomous
+
+00:32.880 --> 00:36.320
+ vehicle experts in the world, and a longtime voice
+
+00:36.320 --> 00:38.840
+ of reason in a space that is shrouded
+
+00:38.840 --> 00:41.320
+ in both mystery and hype.
+
+00:41.320 --> 00:43.600
+ He both acknowledges the incredible challenges
+
+00:43.600 --> 00:46.480
+ involved in solving the problem of autonomous driving
+
+00:46.480 --> 00:49.760
+ and is working hard to solve it.
+
+00:49.760 --> 00:52.400
+ This is the Artificial Intelligence podcast.
+
+00:52.400 --> 00:54.720
+ If you enjoy it, subscribe on YouTube,
+
+00:54.720 --> 00:57.920
+ give it five stars on iTunes, support it on Patreon,
+
+00:57.920 --> 00:59.720
+ or simply connect with me on Twitter
+
+00:59.720 --> 01:03.240
+ at Lex Friedman, spelled F R I D M A N.
+
+01:03.240 --> 01:09.120
+ And now, here's my conversation with Chris Sampson.
+
+01:09.120 --> 01:11.960
+ You were part of both the DARPA Grand Challenge
+
+01:11.960 --> 01:13.880
+ and the DARPA Urban Challenge teams
+
+01:13.880 --> 01:17.040
+ at CMU with Red Whitaker.
+
+01:17.040 --> 01:19.720
+ What technical or philosophical things
+
+01:19.720 --> 01:22.240
+ have you learned from these races?
+
+01:22.240 --> 01:26.600
+ I think the high order bit was that it could be done.
+
+01:26.600 --> 01:30.200
+ I think that was the thing that was
+
+01:30.200 --> 01:34.880
+ incredible about the first of the Grand Challenges,
+
+01:34.880 --> 01:38.160
+ that I remember I was a grad student at Carnegie Mellon,
+
+01:38.160 --> 01:45.360
+ and there was kind of this dichotomy of it
+
+01:45.360 --> 01:46.720
+ seemed really hard, so that would
+
+01:46.720 --> 01:48.800
+ be cool and interesting.
+
+01:48.800 --> 01:52.800
+ But at the time, we were the only robotics institute around,
+
+01:52.800 --> 01:55.560
+ and so if we went into it and fell on our faces,
+
+01:55.560 --> 01:58.360
+ that would be embarrassing.
+
+01:58.360 --> 02:01.120
+ So I think just having the will to go do it,
+
+02:01.120 --> 02:02.880
+ to try to do this thing that at the time
+
+02:02.880 --> 02:05.000
+ was marked as darn near impossible,
+
+02:05.000 --> 02:06.960
+ and then after a couple of tries,
+
+02:06.960 --> 02:08.420
+ be able to actually make it happen,
+
+02:08.420 --> 02:12.320
+ I think that was really exciting.
+
+02:12.320 --> 02:15.040
+ But at which point did you believe it was possible?
+
+02:15.040 --> 02:16.960
+ Did you from the very beginning?
+
+02:16.960 --> 02:18.000
+ Did you personally?
+
+02:18.000 --> 02:19.800
+ Because you're one of the lead engineers.
+
+02:19.800 --> 02:21.800
+ You actually had to do a lot of the work.
+
+02:21.800 --> 02:23.880
+ Yeah, I was the technical director there,
+
+02:23.880 --> 02:26.120
+ and did a lot of the work, along with a bunch
+
+02:26.120 --> 02:28.420
+ of other really good people.
+
+02:28.420 --> 02:29.760
+ Did I believe it could be done?
+
+02:29.760 --> 02:31.080
+ Yeah, of course.
+
+02:31.080 --> 02:32.760
+ Why would you go do something you thought
+
+02:32.760 --> 02:34.800
+ was completely impossible?
+
+02:34.800 --> 02:36.260
+ We thought it was going to be hard.
+
+02:36.260 --> 02:37.800
+ We didn't know how we were going to be able to do it.
+
+02:37.800 --> 02:42.880
+ We didn't know if we'd be able to do it the first time.
+
+02:42.880 --> 02:45.960
+ Turns out we couldn't.
+
+02:45.960 --> 02:48.400
+ That, yeah, I guess you have to.
+
+02:48.400 --> 02:52.960
+ I think there's a certain benefit to naivete, right?
+
+02:52.960 --> 02:55.440
+ That if you don't know how hard something really is,
+
+02:55.440 --> 02:59.600
+ you try different things, and it gives you an opportunity
+
+02:59.600 --> 03:04.120
+ that others who are wiser maybe don't have.
+
+03:04.120 --> 03:05.720
+ What were the biggest pain points?
+
+03:05.720 --> 03:08.880
+ Mechanical, sensors, hardware, software,
+
+03:08.880 --> 03:11.800
+ algorithms for mapping, localization,
+
+03:11.800 --> 03:13.680
+ just general perception, control?
+
+03:13.680 --> 03:15.320
+ Like hardware, software, first of all?
+
+03:15.320 --> 03:20.120
+ I think that's the joy of this field, is that it's all hard
+
+03:20.120 --> 03:25.360
+ and that you have to be good at each part of it.
+
+03:25.360 --> 03:32.360
+ So for the urban challenges, if I look back at it from today,
+
+03:32.360 --> 03:38.960
+ it should be easy today, that it was a static world.
+
+03:38.960 --> 03:40.800
+ There weren't other actors moving through it,
+
+03:40.800 --> 03:42.480
+ is what that means.
+
+03:42.480 --> 03:47.080
+ It was out in the desert, so you get really good GPS.
+
+03:47.080 --> 03:51.400
+ So that went, and we could map it roughly.
+
+03:51.400 --> 03:55.160
+ And so in retrospect now, it's within the realm of things
+
+03:55.160 --> 03:57.840
+ we could do back then.
+
+03:57.840 --> 03:59.720
+ Just actually getting the vehicle and the,
+
+03:59.720 --> 04:00.680
+ there's a bunch of engineering work
+
+04:00.680 --> 04:04.760
+ to get the vehicle so that we could control it and drive it.
+
+04:04.760 --> 04:09.600
+ That's still a pain today, but it was even more so back then.
+
+04:09.600 --> 04:14.280
+ And then the uncertainty of exactly what they wanted us to do
+
+04:14.280 --> 04:17.040
+ was part of the challenge as well.
+
+04:17.040 --> 04:19.440
+ Right, you didn't actually know the track heading in here.
+
+04:19.440 --> 04:21.480
+ You knew approximately, but you didn't actually
+
+04:21.480 --> 04:23.520
+ know the route that was going to be taken.
+
+04:23.520 --> 04:24.920
+ That's right, we didn't know the route.
+
+04:24.920 --> 04:28.600
+ We didn't even really, the way the rules had been described,
+
+04:28.600 --> 04:29.800
+ you had to kind of guess.
+
+04:29.800 --> 04:33.360
+ So if you think back to that challenge,
+
+04:33.360 --> 04:36.960
+ the idea was that the government would give us,
+
+04:36.960 --> 04:40.320
+ the DARPA would give us a set of waypoints
+
+04:40.320 --> 04:43.520
+ and kind of the width that you had to stay within
+
+04:43.520 --> 04:46.800
+ between the line that went between each of those waypoints.
+
+04:46.800 --> 04:49.280
+ And so the most devious thing they could have done
+
+04:49.280 --> 04:53.280
+ is set a kilometer wide corridor across a field
+
+04:53.280 --> 04:58.280
+ of scrub brush and rocks and said, go figure it out.
+
+04:58.520 --> 05:01.920
+ Fortunately, it really, it turned into basically driving
+
+05:01.920 --> 05:05.000
+ along a set of trails, which is much more relevant
+
+05:05.000 --> 05:07.920
+ to the application they were looking for.
+
+05:08.760 --> 05:12.080
+ But no, it was a hell of a thing back in the day.
+
+05:12.080 --> 05:16.640
+ So the legend, Red, was kind of leading that effort
+
+05:16.640 --> 05:19.120
+ in terms of just broadly speaking.
+
+05:19.120 --> 05:22.040
+ So you're a leader now.
+
+05:22.040 --> 05:25.000
+ What have you learned from Red about leadership?
+
+05:25.000 --> 05:26.200
+ I think there's a couple things.
+
+05:26.200 --> 05:31.080
+ One is go and try those really hard things.
+
+05:31.080 --> 05:34.480
+ That's where there is an incredible opportunity.
+
+05:34.480 --> 05:36.560
+ I think the other big one, though,
+
+05:36.560 --> 05:40.680
+ is to see people for who they can be, not who they are.
+
+05:41.720 --> 05:43.720
+ It's one of the things that I actually,
+
+05:43.720 --> 05:46.080
+ one of the deepest lessons I learned from Red
+
+05:46.080 --> 05:50.200
+ was that he would look at undergraduates
+
+05:50.200 --> 05:55.200
+ or graduate students and empower them to be leaders,
+
+05:56.120 --> 06:00.320
+ to have responsibility, to do great things
+
+06:00.320 --> 06:04.480
+ that I think another person might look at them
+
+06:04.480 --> 06:06.600
+ and think, oh, well, that's just an undergraduate student.
+
+06:06.600 --> 06:07.720
+ What could they know?
+
+06:08.680 --> 06:12.720
+ And so I think that kind of trust but verify,
+
+06:12.720 --> 06:14.480
+ have confidence in what people can become,
+
+06:14.480 --> 06:16.680
+ I think is a really powerful thing.
+
+06:16.680 --> 06:20.440
+ So through that, let's just fast forward through the history.
+
+06:20.440 --> 06:24.160
+ Can you maybe talk through the technical evolution
+
+06:24.160 --> 06:26.200
+ of autonomous vehicle systems
+
+06:26.200 --> 06:29.960
+ from the first two Grand Challenges to the Urban Challenge
+
+06:29.960 --> 06:33.560
+ to today, are there major shifts in your mind
+
+06:33.560 --> 06:37.240
+ or is it the same kind of technology just made more robust?
+
+06:37.240 --> 06:39.840
+ I think there's been some big, big steps.
+
+06:40.880 --> 06:43.720
+ So for the Grand Challenge,
+
+06:43.720 --> 06:48.720
+ the real technology that unlocked that was HD mapping.
+
+06:51.400 --> 06:54.200
+ Prior to that, a lot of the off road robotics work
+
+06:55.160 --> 06:58.480
+ had been done without any real prior model
+
+06:58.480 --> 07:01.400
+ of what the vehicle was going to encounter.
+
+07:01.400 --> 07:04.880
+ And so that innovation that the fact that we could get
+
+07:05.960 --> 07:10.960
+ decimeter resolution models was really a big deal.
+
+07:13.440 --> 07:18.200
+ And that allowed us to kind of bound the complexity
+
+07:18.200 --> 07:19.680
+ of the driving problem the vehicle had
+
+07:19.680 --> 07:21.040
+ and allowed it to operate at speed
+
+07:21.040 --> 07:23.800
+ because we could assume things about the environment
+
+07:23.800 --> 07:25.360
+ that it was going to encounter.
+
+07:25.360 --> 07:29.720
+ So that was the big step there.
+
+07:31.280 --> 07:35.280
+ For the Urban Challenge,
+
+07:37.240 --> 07:39.280
+ one of the big technological innovations there
+
+07:39.280 --> 07:41.040
+ was the multi beam LIDAR
+
+07:41.960 --> 07:45.760
+ and being able to generate high resolution,
+
+07:45.760 --> 07:48.680
+ mid to long range 3D models of the world
+
+07:48.680 --> 07:53.680
+ and use that for understanding the world around the vehicle.
+
+07:53.680 --> 07:56.600
+ And that was really kind of a game changing technology.
+
+07:58.600 --> 08:00.000
+ In parallel with that,
+
+08:00.000 --> 08:04.360
+ we saw a bunch of other technologies
+
+08:04.360 --> 08:06.120
+ that had been kind of converging
+
+08:06.120 --> 08:08.440
+ half their day in the sun.
+
+08:08.440 --> 08:12.560
+ So Bayesian estimation had been,
+
+08:12.560 --> 08:17.560
+ SLAM had been a big field in robotics.
+
+08:17.840 --> 08:20.760
+ You would go to a conference a couple of years before that
+
+08:20.760 --> 08:24.880
+ and every paper would effectively have SLAM somewhere in it.
+
+08:24.880 --> 08:29.320
+ And so seeing that the Bayesian estimation techniques
+
+08:30.720 --> 08:33.400
+ play out on a very visible stage,
+
+08:33.400 --> 08:36.520
+ I thought that was pretty exciting to see.
+
+08:38.080 --> 08:41.560
+ And mostly SLAM was done based on LIDAR at that time.
+
+08:41.560 --> 08:44.560
+ Yeah, and in fact, we weren't really doing SLAM per se
+
+08:45.600 --> 08:47.480
+ in real time because we had a model ahead of time,
+
+08:47.480 --> 08:51.040
+ we had a roadmap, but we were doing localization.
+
+08:51.040 --> 08:53.560
+ And we were using the LIDAR or the cameras
+
+08:53.560 --> 08:55.400
+ depending on who exactly was doing it
+
+08:55.400 --> 08:57.560
+ to localize to a model of the world.
+
+08:57.560 --> 09:00.160
+ And I thought that was a big step
+
+09:00.160 --> 09:05.160
+ from kind of naively trusting GPS, INS before that.
+
+09:06.640 --> 09:09.840
+ And again, lots of work had been going on in this field.
+
+09:09.840 --> 09:13.040
+ Certainly this was not doing anything
+
+09:13.040 --> 09:16.840
+ particularly innovative in SLAM or in localization,
+
+09:16.840 --> 09:20.200
+ but it was seeing that technology necessary
+
+09:20.200 --> 09:21.800
+ in a real application on a big stage,
+
+09:21.800 --> 09:23.080
+ I thought was very cool.
+
+09:23.080 --> 09:24.000
+ So for the urban challenge,
+
+09:24.000 --> 09:28.600
+ those are already maps constructed offline in general.
+
+09:28.600 --> 09:30.920
+ And did people do that individually,
+
+09:30.920 --> 09:33.600
+ did individual teams do it individually
+
+09:33.600 --> 09:36.440
+ so they had their own different approaches there
+
+09:36.440 --> 09:41.440
+ or did everybody kind of share that information
+
+09:41.720 --> 09:42.880
+ at least intuitively?
+
+09:42.880 --> 09:47.880
+ So DARPA gave all the teams a model of the world, a map.
+
+09:49.640 --> 09:53.240
+ And then one of the things that we had to figure out
+
+09:53.240 --> 09:56.080
+ back then was, and it's still one of these things
+
+09:56.080 --> 09:57.280
+ that trips people up today
+
+09:57.280 --> 10:00.280
+ is actually the coordinate system.
+
+10:00.280 --> 10:03.080
+ So you get a latitude longitude
+
+10:03.080 --> 10:05.040
+ and to so many decimal places,
+
+10:05.040 --> 10:07.360
+ you don't really care about kind of the ellipsoid
+
+10:07.360 --> 10:09.560
+ of the earth that's being used.
+
+10:09.560 --> 10:12.240
+ But when you want to get to 10 centimeter
+
+10:12.240 --> 10:14.400
+ or centimeter resolution,
+
+10:14.400 --> 10:18.520
+ you care whether the coordinate system is NADS 83
+
+10:18.520 --> 10:23.520
+ or WGS 84 or these are different ways to describe
+
+10:24.200 --> 10:26.760
+ both the kind of non sphericalness of the earth,
+
+10:26.760 --> 10:31.080
+ but also kind of the, I think,
+
+10:31.080 --> 10:32.080
+ I can't remember which one,
+
+10:32.080 --> 10:33.600
+ the tectonic shifts that are happening
+
+10:33.600 --> 10:37.000
+ and how to transform the global datum as a function of that.
+
+10:37.000 --> 10:41.020
+ So getting a map and then actually matching it to reality
+
+10:41.020 --> 10:42.880
+ to centimeter resolution, that was kind of interesting
+
+10:42.880 --> 10:44.040
+ and fun back then.
+
+10:44.040 --> 10:46.760
+ So how much work was the perception doing there?
+
+10:46.760 --> 10:51.760
+ So how much were you relying on localization based on maps
+
+10:52.480 --> 10:55.760
+ without using perception to register to the maps?
+
+10:55.760 --> 10:58.000
+ And I guess the question is how advanced
+
+10:58.000 --> 10:59.800
+ was perception at that point?
+
+10:59.800 --> 11:01.960
+ It's certainly behind where we are today, right?
+
+11:01.960 --> 11:05.840
+ We're more than a decade since the urban challenge.
+
+11:05.840 --> 11:08.640
+ But the core of it was there.
+
+11:08.640 --> 11:13.120
+ That we were tracking vehicles.
+
+11:13.120 --> 11:15.640
+ We had to do that at 100 plus meter range
+
+11:15.640 --> 11:18.320
+ because we had to merge with other traffic.
+
+11:18.320 --> 11:21.240
+ We were using, again, Bayesian estimates
+
+11:21.240 --> 11:23.860
+ for state of these vehicles.
+
+11:23.860 --> 11:25.580
+ We had to deal with a bunch of the problems
+
+11:25.580 --> 11:26.920
+ that you think of today,
+
+11:26.920 --> 11:29.820
+ of predicting where that vehicle's going to be
+
+11:29.820 --> 11:31.060
+ a few seconds into the future.
+
+11:31.060 --> 11:32.380
+ We had to deal with the fact
+
+11:32.380 --> 11:35.320
+ that there were multiple hypotheses for that
+
+11:35.320 --> 11:37.660
+ because a vehicle at an intersection might be going right
+
+11:37.660 --> 11:38.780
+ or it might be going straight
+
+11:38.780 --> 11:40.620
+ or it might be making a left turn.
+
+11:41.500 --> 11:44.120
+ And we had to deal with the challenge of the fact
+
+11:44.120 --> 11:47.600
+ that our behavior was going to impact the behavior
+
+11:47.600 --> 11:48.960
+ of that other operator.
+
+11:48.960 --> 11:53.480
+ And we did a lot of that in relatively naive ways,
+
+11:53.480 --> 11:54.820
+ but it kind of worked.
+
+11:54.820 --> 11:57.080
+ Still had to have some kind of solution.
+
+11:57.080 --> 11:59.960
+ And so where does that, 10 years later,
+
+11:59.960 --> 12:01.520
+ where does that take us today
+
+12:01.520 --> 12:04.260
+ from that artificial city construction
+
+12:04.260 --> 12:07.000
+ to real cities to the urban environment?
+
+12:07.000 --> 12:09.160
+ Yeah, I think the biggest thing
+
+12:09.160 --> 12:14.160
+ is that the actors are truly unpredictable.
+
+12:15.720 --> 12:18.800
+ That most of the time, the drivers on the road,
+
+12:18.800 --> 12:23.800
+ the other road users are out there behaving well,
+
+12:24.080 --> 12:25.880
+ but every once in a while they're not.
+
+12:27.080 --> 12:32.080
+ The variety of other vehicles is, you have all of them.
+
+12:32.080 --> 12:35.840
+ In terms of behavior, in terms of perception, or both?
+
+12:35.840 --> 12:36.680
+ Both.
+
+12:38.740 --> 12:40.520
+ Back then we didn't have to deal with cyclists,
+
+12:40.520 --> 12:42.800
+ we didn't have to deal with pedestrians,
+
+12:42.800 --> 12:44.800
+ didn't have to deal with traffic lights.
+
+12:46.260 --> 12:49.400
+ The scale over which that you have to operate is now
+
+12:49.400 --> 12:51.120
+ is much larger than the air base
+
+12:51.120 --> 12:52.720
+ that we were thinking about back then.
+
+12:52.720 --> 12:55.420
+ So what, easy question,
+
+12:56.280 --> 12:59.720
+ what do you think is the hardest part about driving?
+
+12:59.720 --> 13:00.560
+ Easy question.
+
+13:00.560 --> 13:02.560
+ Yeah, no, I'm joking.
+
+13:02.560 --> 13:07.440
+ I'm sure nothing really jumps out at you as one thing,
+
+13:07.440 --> 13:12.440
+ but in the jump from the urban challenge to the real world,
+
+13:12.920 --> 13:15.320
+ is there something that's a particular,
+
+13:15.320 --> 13:18.480
+ you foresee as very serious, difficult challenge?
+
+13:18.480 --> 13:21.080
+ I think the most fundamental difference
+
+13:21.080 --> 13:25.340
+ is that we're doing it for real.
+
+13:26.760 --> 13:28.960
+ That in that environment,
+
+13:28.960 --> 13:31.880
+ it was both a limited complexity environment
+
+13:31.880 --> 13:33.240
+ because certain actors weren't there,
+
+13:33.240 --> 13:35.380
+ because the roads were maintained,
+
+13:35.380 --> 13:37.360
+ there were barriers keeping people separate
+
+13:37.360 --> 13:39.400
+ from robots at the time,
+
+13:40.840 --> 13:43.300
+ and it only had to work for 60 miles.
+
+13:43.300 --> 13:46.160
+ Which, looking at it from 2006,
+
+13:46.160 --> 13:48.960
+ it had to work for 60 miles, right?
+
+13:48.960 --> 13:50.940
+ Looking at it from now,
+
+13:51.880 --> 13:53.720
+ we want things that will go and drive
+
+13:53.720 --> 13:57.160
+ for half a million miles,
+
+13:57.160 --> 14:00.020
+ and it's just a different game.
+
+14:00.940 --> 14:03.480
+ So how important,
+
+14:03.480 --> 14:06.080
+ you said LiDAR came into the game early on,
+
+14:06.080 --> 14:07.880
+ and it's really the primary driver
+
+14:07.880 --> 14:10.240
+ of autonomous vehicles today as a sensor.
+
+14:10.240 --> 14:11.920
+ So how important is the role of LiDAR
+
+14:11.920 --> 14:14.800
+ in the sensor suite in the near term?
+
+14:14.800 --> 14:16.740
+ So I think it's essential.
+
+14:17.920 --> 14:20.480
+ I believe, but I also believe that cameras are essential,
+
+14:20.480 --> 14:22.120
+ and I believe the radar is essential.
+
+14:22.120 --> 14:26.280
+ I think that you really need to use
+
+14:26.280 --> 14:28.720
+ the composition of data from these different sensors
+
+14:28.720 --> 14:32.640
+ if you want the thing to really be robust.
+
+14:32.640 --> 14:34.360
+ The question I wanna ask,
+
+14:34.360 --> 14:35.600
+ let's see if we can untangle it,
+
+14:35.600 --> 14:39.320
+ is what are your thoughts on the Elon Musk
+
+14:39.320 --> 14:42.340
+ provocative statement that LiDAR is a crutch,
+
+14:42.340 --> 14:47.340
+ that it's a kind of, I guess, growing pains,
+
+14:47.760 --> 14:49.920
+ and that much of the perception task
+
+14:49.920 --> 14:52.120
+ can be done with cameras?
+
+14:52.120 --> 14:55.440
+ So I think it is undeniable
+
+14:55.440 --> 14:59.360
+ that people walk around without lasers in their foreheads,
+
+14:59.360 --> 15:01.880
+ and they can get into vehicles and drive them,
+
+15:01.880 --> 15:05.600
+ and so there's an existence proof
+
+15:05.600 --> 15:09.600
+ that you can drive using passive vision.
+
+15:10.880 --> 15:12.720
+ No doubt, can't argue with that.
+
+15:12.720 --> 15:14.680
+ In terms of sensors, yeah, so there's proof.
+
+15:14.680 --> 15:16.000
+ Yeah, in terms of sensors, right?
+
+15:16.000 --> 15:20.200
+ So there's an example that we all go do it,
+
+15:20.200 --> 15:21.380
+ many of us every day.
+
+15:21.380 --> 15:26.380
+ In terms of LiDAR being a crutch, sure.
+
+15:28.180 --> 15:33.100
+ But in the same way that the combustion engine
+
+15:33.100 --> 15:35.260
+ was a crutch on the path to an electric vehicle,
+
+15:35.260 --> 15:39.300
+ in the same way that any technology ultimately gets
+
+15:40.840 --> 15:44.380
+ replaced by some superior technology in the future,
+
+15:44.380 --> 15:47.740
+ and really the way that I look at this
+
+15:47.740 --> 15:51.460
+ is that the way we get around on the ground,
+
+15:51.460 --> 15:53.920
+ the way that we use transportation is broken,
+
+15:55.280 --> 15:59.740
+ and that we have this, I think the number I saw this morning,
+
+15:59.740 --> 16:04.060
+ 37,000 Americans killed last year on our roads,
+
+16:04.060 --> 16:05.380
+ and that's just not acceptable.
+
+16:05.380 --> 16:09.460
+ And so any technology that we can bring to bear
+
+16:09.460 --> 16:12.860
+ that accelerates this self driving technology
+
+16:12.860 --> 16:14.640
+ coming to market and saving lives
+
+16:14.640 --> 16:17.320
+ is technology we should be using.
+
+16:18.280 --> 16:20.840
+ And it feels just arbitrary to say,
+
+16:20.840 --> 16:25.840
+ well, I'm not okay with using lasers
+
+16:26.240 --> 16:27.820
+ because that's whatever,
+
+16:27.820 --> 16:30.720
+ but I am okay with using an eight megapixel camera
+
+16:30.720 --> 16:32.880
+ or a 16 megapixel camera.
+
+16:32.880 --> 16:34.640
+ These are just bits of technology,
+
+16:34.640 --> 16:36.360
+ and we should be taking the best technology
+
+16:36.360 --> 16:41.360
+ from the tool bin that allows us to go and solve a problem.
+
+16:41.360 --> 16:45.160
+ The question I often talk to, well, obviously you do as well,
+
+16:45.160 --> 16:48.280
+ to sort of automotive companies,
+
+16:48.280 --> 16:51.360
+ and if there's one word that comes up more often
+
+16:51.360 --> 16:55.280
+ than anything, it's cost, and trying to drive costs down.
+
+16:55.280 --> 17:00.280
+ So while it's true that it's a tragic number, the 37,000,
+
+17:01.400 --> 17:04.880
+ the question is, and I'm not the one asking this question
+
+17:04.880 --> 17:05.820
+ because I hate this question,
+
+17:05.820 --> 17:09.960
+ but we want to find the cheapest sensor suite
+
+17:09.960 --> 17:13.280
+ that creates a safe vehicle.
+
+17:13.280 --> 17:18.220
+ So in that uncomfortable trade off,
+
+17:18.220 --> 17:23.220
+ do you foresee LiDAR coming down in cost in the future,
+
+17:23.680 --> 17:26.680
+ or do you see a day where level four autonomy
+
+17:26.680 --> 17:29.880
+ is possible without LiDAR?
+
+17:29.880 --> 17:32.880
+ I see both of those, but it's really a matter of time.
+
+17:32.880 --> 17:36.040
+ And I think really, maybe I would talk to the question
+
+17:36.040 --> 17:37.840
+ you asked about the cheapest sensor.
+
+17:37.840 --> 17:40.360
+ I don't think that's actually what you want.
+
+17:40.360 --> 17:45.360
+ What you want is a sensor suite that is economically viable.
+
+17:45.680 --> 17:49.440
+ And then after that, everything is about margin
+
+17:49.440 --> 17:52.120
+ and driving costs out of the system.
+
+17:52.120 --> 17:55.360
+ What you also want is a sensor suite that works.
+
+17:55.360 --> 17:58.200
+ And so it's great to tell a story about
+
+17:59.600 --> 18:03.260
+ how it would be better to have a self driving system
+
+18:03.260 --> 18:08.040
+ with a $50 sensor instead of a $500 sensor.
+
+18:08.040 --> 18:10.520
+ But if the $500 sensor makes it work
+
+18:10.520 --> 18:14.760
+ and the $50 sensor doesn't work, who cares?
+
+18:15.680 --> 18:20.020
+ So long as you can actually have an economic opportunity,
+
+18:20.020 --> 18:21.520
+ there's an economic opportunity there.
+
+18:21.520 --> 18:23.760
+ And the economic opportunity is important
+
+18:23.760 --> 18:27.760
+ because that's how you actually have a sustainable business
+
+18:27.760 --> 18:31.120
+ and that's how you can actually see this come to scale
+
+18:31.120 --> 18:32.400
+ and be out in the world.
+
+18:32.400 --> 18:34.780
+ And so when I look at LiDAR,
+
+18:35.960 --> 18:38.880
+ I see a technology that has no underlying
+
+18:38.880 --> 18:42.420
+ fundamentally expense to it, fundamental expense to it.
+
+18:42.420 --> 18:46.080
+ It's going to be more expensive than an imager
+
+18:46.080 --> 18:50.360
+ because CMOS processes or FAP processes
+
+18:51.360 --> 18:55.080
+ are dramatically more scalable than mechanical processes.
+
+18:56.200 --> 18:58.320
+ But we still should be able to drive costs down
+
+18:58.320 --> 19:00.120
+ substantially on that side.
+
+19:00.120 --> 19:04.840
+ And then I also do think that with the right business model
+
+19:05.880 --> 19:07.560
+ you can absorb more,
+
+19:07.560 --> 19:09.480
+ certainly more cost on the bill of materials.
+
+19:09.480 --> 19:12.600
+ Yeah, if the sensor suite works, extra value is provided,
+
+19:12.600 --> 19:15.480
+ thereby you don't need to drive costs down to zero.
+
+19:15.480 --> 19:17.100
+ It's the basic economics.
+
+19:17.100 --> 19:18.820
+ You've talked about your intuition
+
+19:18.820 --> 19:22.200
+ that level two autonomy is problematic
+
+19:22.200 --> 19:25.920
+ because of the human factor of vigilance,
+
+19:25.920 --> 19:28.040
+ decrement, complacency, over trust and so on,
+
+19:28.040 --> 19:29.600
+ just us being human.
+
+19:29.600 --> 19:31.120
+ We over trust the system,
+
+19:31.120 --> 19:34.240
+ we start doing even more so partaking
+
+19:34.240 --> 19:37.180
+ in the secondary activities like smartphones and so on.
+
+19:38.680 --> 19:43.000
+ Have your views evolved on this point in either direction?
+
+19:43.000 --> 19:44.800
+ Can you speak to it?
+
+19:44.800 --> 19:47.480
+ So, and I want to be really careful
+
+19:47.480 --> 19:50.380
+ because sometimes this gets twisted in a way
+
+19:50.380 --> 19:53.040
+ that I certainly didn't intend.
+
+19:53.040 --> 19:58.040
+ So active safety systems are a really important technology
+
+19:58.040 --> 20:00.680
+ that we should be pursuing and integrating into vehicles.
+
+20:02.080 --> 20:04.280
+ And there's an opportunity in the near term
+
+20:04.280 --> 20:06.520
+ to reduce accidents, reduce fatalities,
+
+20:06.520 --> 20:10.320
+ and we should be pushing on that.
+
+20:11.960 --> 20:14.680
+ Level two systems are systems
+
+20:14.680 --> 20:18.080
+ where the vehicle is controlling two axes.
+
+20:18.080 --> 20:21.720
+ So braking and throttle slash steering.
+
+20:23.480 --> 20:25.680
+ And I think there are variants of level two systems
+
+20:25.680 --> 20:27.280
+ that are supporting the driver.
+
+20:27.280 --> 20:31.080
+ That absolutely we should encourage to be out there.
+
+20:31.080 --> 20:32.880
+ Where I think there's a real challenge
+
+20:32.880 --> 20:37.640
+ is in the human factors part around this
+
+20:37.640 --> 20:41.240
+ and the misconception from the public
+
+20:41.240 --> 20:43.600
+ around the capability set that that enables
+
+20:43.600 --> 20:45.640
+ and the trust that they should have in it.
+
+20:46.640 --> 20:50.000
+ And that is where I kind of,
+
+20:50.000 --> 20:52.920
+ I'm actually incrementally more concerned
+
+20:52.920 --> 20:54.440
+ around level three systems
+
+20:54.440 --> 20:58.440
+ and how exactly a level two system is marketed and delivered
+
+20:58.440 --> 21:01.840
+ and how much effort people have put into those human factors.
+
+21:01.840 --> 21:05.640
+ So I still believe several things around this.
+
+21:05.640 --> 21:09.440
+ One is people will overtrust the technology.
+
+21:09.440 --> 21:11.440
+ We've seen over the last few weeks
+
+21:11.440 --> 21:14.040
+ a spate of people sleeping in their Tesla.
+
+21:14.920 --> 21:19.920
+ I watched an episode last night of Trevor Noah
+
+21:19.920 --> 21:23.920
+ talking about this and him,
+
+21:23.920 --> 21:26.720
+ this is a smart guy who has a lot of resources
+
+21:26.720 --> 21:30.720
+ at his disposal describing a Tesla as a self driving car
+
+21:30.720 --> 21:33.480
+ and that why shouldn't people be sleeping in their Tesla?
+
+21:33.480 --> 21:36.560
+ And it's like, well, because it's not a self driving car
+
+21:36.560 --> 21:38.840
+ and it is not intended to be
+
+21:38.840 --> 21:43.840
+ and these people will almost certainly die at some point
+
+21:46.400 --> 21:48.040
+ or hurt other people.
+
+21:48.040 --> 21:50.080
+ And so we need to really be thoughtful
+
+21:50.080 --> 21:51.840
+ about how that technology is described
+
+21:51.840 --> 21:53.280
+ and brought to market.
+
+21:54.240 --> 21:59.240
+ I also think that because of the economic challenges
+
+21:59.240 --> 22:01.240
+ we were just talking about,
+
+22:01.240 --> 22:05.160
+ that these level two driver assistance systems,
+
+22:05.160 --> 22:07.280
+ that technology path will diverge
+
+22:07.280 --> 22:10.200
+ from the technology path that we need to be on
+
+22:10.200 --> 22:14.080
+ to actually deliver truly self driving vehicles,
+
+22:14.080 --> 22:16.920
+ ones where you can get in it and drive it.
+
+22:16.920 --> 22:20.800
+ Can get in it and sleep and have the equivalent
+
+22:20.800 --> 22:24.680
+ or better safety than a human driver behind the wheel.
+
+22:24.680 --> 22:27.520
+ Because again, the economics are very different
+
+22:28.480 --> 22:30.880
+ in those two worlds and so that leads
+
+22:30.880 --> 22:32.800
+ to divergent technology.
+
+22:32.800 --> 22:34.680
+ So you just don't see the economics
+
+22:34.680 --> 22:38.560
+ of gradually increasing from level two
+
+22:38.560 --> 22:41.600
+ and doing so quickly enough
+
+22:41.600 --> 22:44.480
+ to where it doesn't cause safety, critical safety concerns.
+
+22:44.480 --> 22:47.680
+ You believe that it needs to diverge at this point
+
+22:48.680 --> 22:50.800
+ into basically different routes.
+
+22:50.800 --> 22:55.560
+ And really that comes back to what are those L2
+
+22:55.560 --> 22:57.080
+ and L1 systems doing?
+
+22:57.080 --> 22:59.840
+ And they are driver assistance functions
+
+22:59.840 --> 23:04.400
+ where the people that are marketing that responsibly
+
+23:04.400 --> 23:08.000
+ are being very clear and putting human factors in place
+
+23:08.000 --> 23:12.440
+ such that the driver is actually responsible for the vehicle
+
+23:12.440 --> 23:15.160
+ and that the technology is there to support the driver.
+
+23:15.160 --> 23:19.880
+ And the safety cases that are built around those
+
+23:19.880 --> 23:24.040
+ are dependent on that driver attention and attentiveness.
+
+23:24.040 --> 23:28.000
+ And at that point, you can kind of give up
+
+23:29.160 --> 23:31.240
+ to some degree for economic reasons,
+
+23:31.240 --> 23:33.480
+ you can give up on say false negatives.
+
+23:34.800 --> 23:36.200
+ And the way to think about this
+
+23:36.200 --> 23:39.320
+ is for a four collision mitigation braking system,
+
+23:39.320 --> 23:43.960
+ if it half the times the driver missed a vehicle
+
+23:43.960 --> 23:46.080
+ in front of it, it hit the brakes
+
+23:46.080 --> 23:47.680
+ and brought the vehicle to a stop,
+
+23:47.680 --> 23:51.640
+ that would be an incredible, incredible advance
+
+23:51.640 --> 23:53.040
+ in safety on our roads, right?
+
+23:53.040 --> 23:55.000
+ That would be equivalent to seat belts.
+
+23:55.000 --> 23:56.600
+ But it would mean that if that vehicle
+
+23:56.600 --> 23:59.440
+ wasn't being monitored, it would hit one out of two cars.
+
+24:00.600 --> 24:05.120
+ And so economically, that's a perfectly good solution
+
+24:05.120 --> 24:06.280
+ for a driver assistance system.
+
+24:06.280 --> 24:07.240
+ What you should do at that point,
+
+24:07.240 --> 24:09.240
+ if you can get it to work 50% of the time,
+
+24:09.240 --> 24:10.520
+ is drive the cost out of that
+
+24:10.520 --> 24:13.320
+ so you can get it on as many vehicles as possible.
+
+24:13.320 --> 24:14.760
+ But driving the cost out of it
+
+24:14.760 --> 24:18.800
+ doesn't drive up performance on the false negative case.
+
+24:18.800 --> 24:21.440
+ And so you'll continue to not have a technology
+
+24:21.440 --> 24:25.680
+ that could really be available for a self driven vehicle.
+
+24:25.680 --> 24:28.440
+ So clearly the communication,
+
+24:28.440 --> 24:31.600
+ and this probably applies to all four vehicles as well,
+
+24:31.600 --> 24:34.440
+ the marketing and communication
+
+24:34.440 --> 24:37.040
+ of what the technology is actually capable of,
+
+24:37.040 --> 24:38.400
+ how hard it is, how easy it is,
+
+24:38.400 --> 24:41.000
+ all that kind of stuff is highly problematic.
+
+24:41.000 --> 24:45.640
+ So say everybody in the world was perfectly communicated
+
+24:45.640 --> 24:48.400
+ and were made to be completely aware
+
+24:48.400 --> 24:50.000
+ of every single technology out there,
+
+24:50.000 --> 24:52.840
+ what it's able to do.
+
+24:52.840 --> 24:54.120
+ What's your intuition?
+
+24:54.120 --> 24:56.880
+ And now we're maybe getting into philosophical ground.
+
+24:56.880 --> 25:00.000
+ Is it possible to have a level two vehicle
+
+25:00.000 --> 25:03.280
+ where we don't over trust it?
+
+25:04.680 --> 25:05.800
+ I don't think so.
+
+25:05.800 --> 25:10.800
+ If people truly understood the risks and internalized it,
+
+25:11.160 --> 25:14.320
+ then sure, you could do that safely.
+
+25:14.320 --> 25:16.160
+ But that's a world that doesn't exist.
+
+25:16.160 --> 25:17.520
+ The people are going to,
+
+25:18.720 --> 25:20.760
+ if the facts are put in front of them,
+
+25:20.760 --> 25:24.440
+ they're gonna then combine that with their experience.
+
+25:24.440 --> 25:28.360
+ And let's say they're using an L2 system
+
+25:28.360 --> 25:30.800
+ and they go up and down the 101 every day
+
+25:30.800 --> 25:32.720
+ and they do that for a month.
+
+25:32.720 --> 25:36.200
+ And it just worked every day for a month.
+
+25:36.200 --> 25:39.000
+ Like that's pretty compelling at that point,
+
+25:39.000 --> 25:41.800
+ just even if you know the statistics,
+
+25:41.800 --> 25:43.400
+ you're like, well, I don't know,
+
+25:43.400 --> 25:44.760
+ maybe there's something funny about those.
+
+25:44.760 --> 25:46.920
+ Maybe they're driving in difficult places.
+
+25:46.920 --> 25:49.840
+ Like I've seen it with my own eyes, it works.
+
+25:49.840 --> 25:52.400
+ And the problem is that that sample size that they have,
+
+25:52.400 --> 25:53.880
+ so it's 30 miles up and down,
+
+25:53.880 --> 25:56.360
+ so 60 miles times 30 days,
+
+25:56.360 --> 25:58.720
+ so 60, 180, 1,800 miles.
+
+25:58.720 --> 26:03.280
+ Like that's a drop in the bucket
+
+26:03.280 --> 26:07.640
+ compared to the, what, 85 million miles between fatalities.
+
+26:07.640 --> 26:11.400
+ And so they don't really have a true estimate
+
+26:11.400 --> 26:14.440
+ based on their personal experience of the real risks,
+
+26:14.440 --> 26:15.640
+ but they're gonna trust it anyway,
+
+26:15.640 --> 26:16.480
+ because it's hard not to.
+
+26:16.480 --> 26:18.640
+ It worked for a month, what's gonna change?
+
+26:18.640 --> 26:21.640
+ So even if you start a perfect understanding of the system,
+
+26:21.640 --> 26:24.160
+ your own experience will make it drift.
+
+26:24.160 --> 26:25.920
+ I mean, that's a big concern.
+
+26:25.920 --> 26:28.160
+ Over a year, over two years even,
+
+26:28.160 --> 26:29.440
+ it doesn't have to be months.
+
+26:29.440 --> 26:32.920
+ And I think that as this technology moves
+
+26:32.920 --> 26:37.760
+ from what I would say is kind of the more technology savvy
+
+26:37.760 --> 26:40.880
+ ownership group to the mass market,
+
+26:42.640 --> 26:44.600
+ you may be able to have some of those folks
+
+26:44.600 --> 26:46.280
+ who are really familiar with technology,
+
+26:46.280 --> 26:48.840
+ they may be able to internalize it better.
+
+26:48.840 --> 26:50.800
+ And your kind of immunization
+
+26:50.800 --> 26:53.360
+ against this kind of false risk assessment
+
+26:53.360 --> 26:54.280
+ might last longer,
+
+26:54.280 --> 26:58.680
+ but as folks who aren't as savvy about that
+
+26:58.680 --> 27:00.880
+ read the material and they compare that
+
+27:00.880 --> 27:02.160
+ to their personal experience,
+
+27:02.160 --> 27:07.160
+ I think there it's going to move more quickly.
+
+27:08.160 --> 27:11.280
+ So your work, the program that you've created at Google
+
+27:11.280 --> 27:16.280
+ and now at Aurora is focused more on the second path
+
+27:16.600 --> 27:18.480
+ of creating full autonomy.
+
+27:18.480 --> 27:20.880
+ So it's such a fascinating,
+
+27:20.880 --> 27:24.560
+ I think it's one of the most interesting AI problems
+
+27:24.560 --> 27:25.600
+ of the century, right?
+
+27:25.600 --> 27:28.280
+ It's, I just talked to a lot of people,
+
+27:28.280 --> 27:29.440
+ just regular people, I don't know,
+
+27:29.440 --> 27:31.720
+ my mom, about autonomous vehicles,
+
+27:31.720 --> 27:34.520
+ and you begin to grapple with ideas
+
+27:34.520 --> 27:38.080
+ of giving your life control over to a machine.
+
+27:38.080 --> 27:40.040
+ It's philosophically interesting,
+
+27:40.040 --> 27:41.760
+ it's practically interesting.
+
+27:41.760 --> 27:43.720
+ So let's talk about safety.
+
+27:43.720 --> 27:46.240
+ How do you think we demonstrate,
+
+27:46.240 --> 27:47.880
+ you've spoken about metrics in the past,
+
+27:47.880 --> 27:51.880
+ how do you think we demonstrate to the world
+
+27:51.880 --> 27:56.160
+ that an autonomous vehicle, an Aurora system is safe?
+
+27:56.160 --> 27:57.320
+ This is one where it's difficult
+
+27:57.320 --> 27:59.280
+ because there isn't a soundbite answer.
+
+27:59.280 --> 28:04.280
+ That we have to show a combination of work
+
+28:05.960 --> 28:08.360
+ that was done diligently and thoughtfully,
+
+28:08.360 --> 28:10.840
+ and this is where something like a functional safety process
+
+28:10.840 --> 28:11.680
+ is part of that.
+
+28:11.680 --> 28:14.360
+ It's like here's the way we did the work,
+
+28:15.280 --> 28:17.160
+ that means that we were very thorough.
+
+28:17.160 --> 28:20.040
+ So if you believe that what we said
+
+28:20.040 --> 28:21.440
+ about this is the way we did it,
+
+28:21.440 --> 28:22.720
+ then you can have some confidence
+
+28:22.720 --> 28:25.200
+ that we were thorough in the engineering work
+
+28:25.200 --> 28:26.920
+ we put into the system.
+
+28:26.920 --> 28:28.920
+ And then on top of that,
+
+28:28.920 --> 28:32.000
+ to kind of demonstrate that we weren't just thorough,
+
+28:32.000 --> 28:33.800
+ we were actually good at what we did,
+
+28:35.280 --> 28:38.200
+ there'll be a kind of a collection of evidence
+
+28:38.200 --> 28:40.440
+ in terms of demonstrating that the capabilities
+
+28:40.440 --> 28:42.920
+ worked the way we thought they did,
+
+28:42.920 --> 28:45.320
+ statistically and to whatever degree
+
+28:45.320 --> 28:47.280
+ we can demonstrate that,
+
+28:48.160 --> 28:50.320
+ both in some combination of simulations,
+
+28:50.320 --> 28:53.080
+ some combination of unit testing
+
+28:53.080 --> 28:54.640
+ and decomposition testing,
+
+28:54.640 --> 28:57.000
+ and then some part of it will be on road data.
+
+28:58.160 --> 29:02.680
+ And I think the way we'll ultimately
+
+29:02.680 --> 29:04.000
+ convey this to the public
+
+29:04.000 --> 29:06.760
+ is there'll be clearly some conversation
+
+29:06.760 --> 29:08.200
+ with the public about it,
+
+29:08.200 --> 29:12.040
+ but we'll kind of invoke the kind of the trusted nodes
+
+29:12.040 --> 29:13.880
+ and that we'll spend more time
+
+29:13.880 --> 29:17.280
+ being able to go into more depth with folks like NHTSA
+
+29:17.280 --> 29:19.720
+ and other federal and state regulatory bodies
+
+29:19.720 --> 29:22.080
+ and kind of given that they are
+
+29:22.080 --> 29:25.200
+ operating in the public interest and they're trusted,
+
+29:26.240 --> 29:28.640
+ that if we can show enough work to them
+
+29:28.640 --> 29:30.000
+ that they're convinced,
+
+29:30.000 --> 29:33.800
+ then I think we're in a pretty good place.
+
+29:33.800 --> 29:35.000
+ That means you work with people
+
+29:35.000 --> 29:36.920
+ that are essentially experts at safety
+
+29:36.920 --> 29:39.000
+ to try to discuss and show.
+
+29:39.000 --> 29:41.720
+ Do you think, the answer's probably no,
+
+29:41.720 --> 29:42.920
+ but just in case,
+
+29:42.920 --> 29:44.360
+ do you think there exists a metric?
+
+29:44.360 --> 29:46.320
+ So currently people have been using
+
+29:46.320 --> 29:48.200
+ number of disengagements.
+
+29:48.200 --> 29:50.120
+ And it quickly turns into a marketing scheme
+
+29:50.120 --> 29:54.280
+ to sort of you alter the experiments you run to adjust.
+
+29:54.280 --> 29:56.280
+ I think you've spoken that you don't like.
+
+29:56.280 --> 29:57.120
+ Don't love it.
+
+29:57.120 --> 29:59.680
+ No, in fact, I was on the record telling DMV
+
+29:59.680 --> 30:01.960
+ that I thought this was not a great metric.
+
+30:01.960 --> 30:05.280
+ Do you think it's possible to create a metric,
+
+30:05.280 --> 30:09.440
+ a number that could demonstrate safety
+
+30:09.440 --> 30:12.320
+ outside of fatalities?
+
+30:12.320 --> 30:13.440
+ So I do.
+
+30:13.440 --> 30:16.560
+ And I think that it won't be just one number.
+
+30:17.600 --> 30:21.280
+ So as we are internally grappling with this,
+
+30:21.280 --> 30:23.560
+ and at some point we'll be able to talk
+
+30:23.560 --> 30:25.040
+ more publicly about it,
+
+30:25.040 --> 30:28.520
+ is how do we think about human performance
+
+30:28.520 --> 30:29.840
+ in different tasks,
+
+30:29.840 --> 30:32.160
+ say detecting traffic lights
+
+30:32.160 --> 30:36.200
+ or safely making a left turn across traffic?
+
+30:37.680 --> 30:40.080
+ And what do we think the failure rates are
+
+30:40.080 --> 30:42.520
+ for those different capabilities for people?
+
+30:42.520 --> 30:44.760
+ And then demonstrating to ourselves
+
+30:44.760 --> 30:48.480
+ and then ultimately folks in the regulatory role
+
+30:48.480 --> 30:50.760
+ and then ultimately the public
+
+30:50.760 --> 30:52.400
+ that we have confidence that our system
+
+30:52.400 --> 30:54.760
+ will work better than that.
+
+30:54.760 --> 30:57.040
+ And so these individual metrics
+
+30:57.040 --> 31:00.720
+ will kind of tell a compelling story ultimately.
+
+31:01.760 --> 31:03.920
+ I do think at the end of the day
+
+31:03.920 --> 31:06.640
+ what we care about in terms of safety
+
+31:06.640 --> 31:11.320
+ is life saved and injuries reduced.
+
+31:12.160 --> 31:15.280
+ And then ultimately kind of casualty dollars
+
+31:16.440 --> 31:19.360
+ that people aren't having to pay to get their car fixed.
+
+31:19.360 --> 31:22.680
+ And I do think that in aviation
+
+31:22.680 --> 31:25.880
+ they look at a kind of an event pyramid
+
+31:25.880 --> 31:28.600
+ where a crash is at the top of that
+
+31:28.600 --> 31:30.440
+ and that's the worst event obviously
+
+31:30.440 --> 31:34.240
+ and then there's injuries and near miss events and whatnot
+
+31:34.240 --> 31:37.320
+ and violation of operating procedures
+
+31:37.320 --> 31:40.160
+ and you kind of build a statistical model
+
+31:40.160 --> 31:44.440
+ of the relevance of the low severity things
+
+31:44.440 --> 31:45.280
+ or the high severity things.
+
+31:45.280 --> 31:46.120
+ And I think that's something
+
+31:46.120 --> 31:48.200
+ where we'll be able to look at as well
+
+31:48.200 --> 31:51.840
+ because an event per 85 million miles
+
+31:51.840 --> 31:54.440
+ is statistically a difficult thing
+
+31:54.440 --> 31:56.800
+ even at the scale of the U.S.
+
+31:56.800 --> 31:59.360
+ to kind of compare directly.
+
+31:59.360 --> 32:02.240
+ And that event fatality that's connected
+
+32:02.240 --> 32:07.240
+ to an autonomous vehicle is significantly
+
+32:07.440 --> 32:09.160
+ at least currently magnified
+
+32:09.160 --> 32:12.320
+ in the amount of attention it gets.
+
+32:12.320 --> 32:15.080
+ So that speaks to public perception.
+
+32:15.080 --> 32:16.720
+ I think the most popular topic
+
+32:16.720 --> 32:19.480
+ about autonomous vehicles in the public
+
+32:19.480 --> 32:23.080
+ is the trolley problem formulation, right?
+
+32:23.080 --> 32:27.000
+ Which has, let's not get into that too much
+
+32:27.000 --> 32:29.600
+ but is misguided in many ways.
+
+32:29.600 --> 32:32.320
+ But it speaks to the fact that people are grappling
+
+32:32.320 --> 32:36.160
+ with this idea of giving control over to a machine.
+
+32:36.160 --> 32:41.160
+ So how do you win the hearts and minds of the people
+
+32:41.560 --> 32:44.600
+ that autonomy is something that could be a part
+
+32:44.600 --> 32:45.520
+ of their lives?
+
+32:45.520 --> 32:47.640
+ I think you let them experience it, right?
+
+32:47.640 --> 32:50.440
+ I think it's right.
+
+32:50.440 --> 32:52.800
+ I think people should be skeptical.
+
+32:52.800 --> 32:55.680
+ I think people should ask questions.
+
+32:55.680 --> 32:57.000
+ I think they should doubt
+
+32:57.000 --> 33:00.120
+ because this is something new and different.
+
+33:00.120 --> 33:01.880
+ They haven't touched it yet.
+
+33:01.880 --> 33:03.640
+ And I think that's perfectly reasonable.
+
+33:03.640 --> 33:07.320
+ And, but at the same time,
+
+33:07.320 --> 33:09.320
+ it's clear there's an opportunity to make the road safer.
+
+33:09.320 --> 33:12.440
+ It's clear that we can improve access to mobility.
+
+33:12.440 --> 33:14.960
+ It's clear that we can reduce the cost of mobility.
+
+33:16.640 --> 33:19.480
+ And that once people try that
+
+33:19.480 --> 33:22.720
+ and understand that it's safe
+
+33:22.720 --> 33:24.440
+ and are able to use in their daily lives,
+
+33:24.440 --> 33:25.280
+ I think it's one of these things
+
+33:25.280 --> 33:28.040
+ that will just be obvious.
+
+33:28.040 --> 33:32.240
+ And I've seen this practically in demonstrations
+
+33:32.240 --> 33:35.560
+ that I've given where I've had people come in
+
+33:35.560 --> 33:38.840
+ and they're very skeptical.
+
+33:38.840 --> 33:40.440
+ Again, in a vehicle, my favorite one
+
+33:40.440 --> 33:42.560
+ is taking somebody out on the freeway
+
+33:42.560 --> 33:46.000
+ and we're on the 101 driving at 65 miles an hour.
+
+33:46.000 --> 33:48.400
+ And after 10 minutes, they kind of turn and ask,
+
+33:48.400 --> 33:49.480
+ is that all it does?
+
+33:49.480 --> 33:52.080
+ And you're like, it's a self driving car.
+
+33:52.080 --> 33:54.840
+ I'm not sure exactly what you thought it would do, right?
+
+33:54.840 --> 33:57.920
+ But it becomes mundane,
+
+33:58.840 --> 34:01.480
+ which is exactly what you want a technology
+
+34:01.480 --> 34:02.720
+ like this to be, right?
+
+34:02.720 --> 34:07.280
+ We don't really, when I turn the light switch on in here,
+
+34:07.280 --> 34:12.000
+ I don't think about the complexity of those electrons
+
+34:12.000 --> 34:14.200
+ being pushed down a wire from wherever it was
+
+34:14.200 --> 34:15.240
+ and being generated.
+
+34:15.240 --> 34:19.080
+ It's like, I just get annoyed if it doesn't work, right?
+
+34:19.080 --> 34:21.400
+ And what I value is the fact
+
+34:21.400 --> 34:23.080
+ that I can do other things in this space.
+
+34:23.080 --> 34:24.560
+ I can see my colleagues.
+
+34:24.560 --> 34:26.160
+ I can read stuff on a paper.
+
+34:26.160 --> 34:29.200
+ I can not be afraid of the dark.
+
+34:30.360 --> 34:33.320
+ And I think that's what we want this technology to be like
+
+34:33.320 --> 34:34.640
+ is it's in the background
+
+34:34.640 --> 34:37.120
+ and people get to have those life experiences
+
+34:37.120 --> 34:38.440
+ and do so safely.
+
+34:38.440 --> 34:42.160
+ So putting this technology in the hands of people
+
+34:42.160 --> 34:46.320
+ speaks to scale of deployment, right?
+
+34:46.320 --> 34:50.880
+ So what do you think the dreaded question about the future
+
+34:50.880 --> 34:53.560
+ because nobody can predict the future,
+
+34:53.560 --> 34:57.240
+ but just maybe speak poetically
+
+34:57.240 --> 35:00.880
+ about when do you think we'll see a large scale deployment
+
+35:00.880 --> 35:05.880
+ of autonomous vehicles, 10,000, those kinds of numbers?
+
+35:06.680 --> 35:08.240
+ We'll see that within 10 years.
+
+35:09.240 --> 35:10.240
+ I'm pretty confident.
+
+35:14.040 --> 35:16.040
+ What's an impressive scale?
+
+35:16.040 --> 35:19.200
+ What moment, so you've done the DARPA challenge
+
+35:19.200 --> 35:20.440
+ where there's one vehicle.
+
+35:20.440 --> 35:23.960
+ At which moment does it become, wow, this is serious scale?
+
+35:23.960 --> 35:26.520
+ So I think the moment it gets serious
+
+35:26.520 --> 35:31.520
+ is when we really do have a driverless vehicle
+
+35:32.240 --> 35:34.120
+ operating on public roads
+
+35:35.000 --> 35:37.960
+ and that we can do that kind of continuously.
+
+35:37.960 --> 35:38.880
+ Without a safety driver.
+
+35:38.880 --> 35:40.440
+ Without a safety driver in the vehicle.
+
+35:40.440 --> 35:41.560
+ I think at that moment,
+
+35:41.560 --> 35:44.400
+ we've kind of crossed the zero to one threshold.
+
+35:45.920 --> 35:50.200
+ And then it is about how do we continue to scale that?
+
+35:50.200 --> 35:53.960
+ How do we build the right business models?
+
+35:53.960 --> 35:56.320
+ How do we build the right customer experience around it
+
+35:56.320 --> 35:59.960
+ so that it is actually a useful product out in the world?
+
+36:00.960 --> 36:03.600
+ And I think that is really,
+
+36:03.600 --> 36:05.920
+ at that point it moves from
+
+36:05.920 --> 36:09.200
+ what is this kind of mixed science engineering project
+
+36:09.200 --> 36:12.360
+ into engineering and commercialization
+
+36:12.360 --> 36:15.840
+ and really starting to deliver on the value
+
+36:15.840 --> 36:20.680
+ that we all see here and actually making that real in the world.
+
+36:20.680 --> 36:22.240
+ What do you think that deployment looks like?
+
+36:22.240 --> 36:26.440
+ Where do we first see the inkling of no safety driver,
+
+36:26.440 --> 36:28.600
+ one or two cars here and there?
+
+36:28.600 --> 36:29.800
+ Is it on the highway?
+
+36:29.800 --> 36:33.160
+ Is it in specific routes in the urban environment?
+
+36:33.160 --> 36:36.920
+ I think it's gonna be urban, suburban type environments.
+
+36:37.880 --> 36:41.560
+ Yeah, with Aurora, when we thought about how to tackle this,
+
+36:41.560 --> 36:45.040
+ it was kind of in vogue to think about trucking
+
+36:46.040 --> 36:47.800
+ as opposed to urban driving.
+
+36:47.800 --> 36:51.280
+ And again, the human intuition around this
+
+36:51.280 --> 36:55.400
+ is that freeways are easier to drive on
+
+36:57.080 --> 36:59.280
+ because everybody's kind of going in the same direction
+
+36:59.280 --> 37:01.560
+ and lanes are a little wider, et cetera.
+
+37:01.560 --> 37:03.320
+ And I think that that intuition is pretty good,
+
+37:03.320 --> 37:06.040
+ except we don't really care about most of the time.
+
+37:06.040 --> 37:08.400
+ We care about all of the time.
+
+37:08.400 --> 37:10.880
+ And when you're driving on a freeway with a truck,
+
+37:10.880 --> 37:13.440
+ say 70 miles an hour,
+
+37:14.600 --> 37:16.240
+ and you've got 70,000 pound load with you,
+
+37:16.240 --> 37:18.880
+ that's just an incredible amount of kinetic energy.
+
+37:18.880 --> 37:21.440
+ And so when that goes wrong, it goes really wrong.
+
+37:22.640 --> 37:27.640
+ And those challenges that you see occur more rarely,
+
+37:27.800 --> 37:31.120
+ so you don't get to learn as quickly.
+
+37:31.120 --> 37:34.720
+ And they're incrementally more difficult than urban driving,
+
+37:34.720 --> 37:37.440
+ but they're not easier than urban driving.
+
+37:37.440 --> 37:41.640
+ And so I think this happens in moderate speed
+
+37:41.640 --> 37:45.280
+ urban environments because if two vehicles crash
+
+37:45.280 --> 37:48.120
+ at 25 miles per hour, it's not good,
+
+37:48.120 --> 37:50.120
+ but probably everybody walks away.
+
+37:51.080 --> 37:53.720
+ And those events where there's the possibility
+
+37:53.720 --> 37:55.800
+ for that occurring happen frequently.
+
+37:55.800 --> 37:58.000
+ So we get to learn more rapidly.
+
+37:58.000 --> 38:01.360
+ We get to do that with lower risk for everyone.
+
+38:02.520 --> 38:04.360
+ And then we can deliver value to people
+
+38:04.360 --> 38:05.880
+ that need to get from one place to another.
+
+38:05.880 --> 38:08.160
+ And once we've got that solved,
+
+38:08.160 --> 38:11.320
+ then the freeway driving part of this just falls out.
+
+38:11.320 --> 38:13.080
+ But we're able to learn more safely,
+
+38:13.080 --> 38:15.200
+ more quickly in the urban environment.
+
+38:15.200 --> 38:18.760
+ So 10 years and then scale 20, 30 year,
+
+38:18.760 --> 38:22.040
+ who knows if a sufficiently compelling experience
+
+38:22.040 --> 38:24.400
+ is created, it could be faster and slower.
+
+38:24.400 --> 38:27.160
+ Do you think there could be breakthroughs
+
+38:27.160 --> 38:29.920
+ and what kind of breakthroughs might there be
+
+38:29.920 --> 38:32.400
+ that completely change that timeline?
+
+38:32.400 --> 38:35.360
+ Again, not only am I asking you to predict the future,
+
+38:35.360 --> 38:37.360
+ I'm asking you to predict breakthroughs
+
+38:37.360 --> 38:38.360
+ that haven't happened yet.
+
+38:38.360 --> 38:41.440
+ So what's the, I think another way to ask that
+
+38:41.440 --> 38:44.320
+ would be if I could wave a magic wand,
+
+38:44.320 --> 38:46.720
+ what part of the system would I make work today
+
+38:46.720 --> 38:49.480
+ to accelerate it as quickly as possible?
+
+38:52.120 --> 38:54.200
+ Don't say infrastructure, please don't say infrastructure.
+
+38:54.200 --> 38:56.320
+ No, it's definitely not infrastructure.
+
+38:56.320 --> 39:00.600
+ It's really that perception forecasting capability.
+
+39:00.600 --> 39:04.840
+ So if tomorrow you could give me a perfect model
+
+39:04.840 --> 39:06.960
+ of what's happened, what is happening
+
+39:06.960 --> 39:09.200
+ and what will happen for the next five seconds
+
+39:10.360 --> 39:13.040
+ around a vehicle on the roadway,
+
+39:13.040 --> 39:15.360
+ that would accelerate things pretty dramatically.
+
+39:15.360 --> 39:17.600
+ Are you, in terms of staying up at night,
+
+39:17.600 --> 39:21.760
+ are you mostly bothered by cars, pedestrians or cyclists?
+
+39:21.760 --> 39:25.960
+ So I worry most about the vulnerable road users
+
+39:25.960 --> 39:28.480
+ about the combination of cyclists and cars, right?
+
+39:28.480 --> 39:31.960
+ Or cyclists and pedestrians because they're not in armor.
+
+39:31.960 --> 39:36.480
+ The cars, they're bigger, they've got protection
+
+39:36.480 --> 39:39.440
+ for the people and so the ultimate risk is lower there.
+
+39:41.080 --> 39:43.240
+ Whereas a pedestrian or a cyclist,
+
+39:43.240 --> 39:46.480
+ they're out on the road and they don't have any protection
+
+39:46.480 --> 39:49.720
+ and so we need to pay extra attention to that.
+
+39:49.720 --> 39:54.120
+ Do you think about a very difficult technical challenge
+
+39:55.720 --> 39:58.520
+ of the fact that pedestrians,
+
+39:58.520 --> 40:00.240
+ if you try to protect pedestrians
+
+40:00.240 --> 40:04.560
+ by being careful and slow, they'll take advantage of that.
+
+40:04.560 --> 40:09.040
+ So the game theoretic dance, does that worry you
+
+40:09.040 --> 40:12.480
+ of how, from a technical perspective, how we solve that?
+
+40:12.480 --> 40:14.560
+ Because as humans, the way we solve that
+
+40:14.560 --> 40:17.240
+ is kind of nudge our way through the pedestrians
+
+40:17.240 --> 40:20.000
+ which doesn't feel, from a technical perspective,
+
+40:20.000 --> 40:22.300
+ as a appropriate algorithm.
+
+40:23.200 --> 40:25.920
+ But do you think about how we solve that problem?
+
+40:25.920 --> 40:30.920
+ Yeah, I think there's two different concepts there.
+
+40:31.360 --> 40:35.820
+ So one is, am I worried that because these vehicles
+
+40:35.820 --> 40:37.600
+ are self driving, people will kind of step in the road
+
+40:37.600 --> 40:38.640
+ and take advantage of them?
+
+40:38.640 --> 40:43.640
+ And I've heard this and I don't really believe it
+
+40:43.760 --> 40:45.960
+ because if I'm driving down the road
+
+40:45.960 --> 40:48.400
+ and somebody steps in front of me, I'm going to stop.
+
+40:50.600 --> 40:53.660
+ Even if I'm annoyed, I'm not gonna just drive
+
+40:53.660 --> 40:56.400
+ through a person stood in the road.
+
+40:56.400 --> 41:00.400
+ And so I think today people can take advantage of this
+
+41:00.400 --> 41:02.560
+ and you do see some people do it.
+
+41:02.560 --> 41:04.180
+ I guess there's an incremental risk
+
+41:04.180 --> 41:05.880
+ because maybe they have lower confidence
+
+41:05.880 --> 41:07.720
+ that I'm gonna see them than they might have
+
+41:07.720 --> 41:10.400
+ for an automated vehicle and so maybe that shifts
+
+41:10.400 --> 41:12.040
+ it a little bit.
+
+41:12.040 --> 41:14.360
+ But I think people don't wanna get hit by cars.
+
+41:14.360 --> 41:17.080
+ And so I think that I'm not that worried
+
+41:17.080 --> 41:18.760
+ about people walking out of the 101
+
+41:18.760 --> 41:23.760
+ and creating chaos more than they would today.
+
+41:24.400 --> 41:27.040
+ Regarding kind of the nudging through a big stream
+
+41:27.040 --> 41:30.040
+ of pedestrians leaving a concert or something,
+
+41:30.040 --> 41:33.520
+ I think that is further down the technology pipeline.
+
+41:33.520 --> 41:36.960
+ I think that you're right, that's tricky.
+
+41:36.960 --> 41:38.620
+ I don't think it's necessarily,
+
+41:40.360 --> 41:43.600
+ I think the algorithm people use for this is pretty simple.
+
+41:43.600 --> 41:44.800
+ It's kind of just move forward slowly
+
+41:44.800 --> 41:46.800
+ and if somebody's really close then stop.
+
+41:46.800 --> 41:50.880
+ And I think that that probably can be replicated
+
+41:50.880 --> 41:54.040
+ pretty easily and particularly given that
+
+41:54.040 --> 41:55.720
+ you don't do this at 30 miles an hour,
+
+41:55.720 --> 41:59.080
+ you do it at one, that even in those situations
+
+41:59.080 --> 42:01.200
+ the risk is relatively minimal.
+
+42:01.200 --> 42:03.640
+ But it's not something we're thinking about
+
+42:03.640 --> 42:04.560
+ in any serious way.
+
+42:04.560 --> 42:07.920
+ And probably that's less an algorithm problem
+
+42:07.920 --> 42:10.160
+ and more creating a human experience.
+
+42:10.160 --> 42:14.300
+ So the HCI people that create a visual display
+
+42:14.300 --> 42:16.260
+ that you're pleasantly as a pedestrian
+
+42:16.260 --> 42:20.760
+ nudged out of the way, that's an experience problem,
+
+42:20.760 --> 42:22.000
+ not an algorithm problem.
+
+42:22.880 --> 42:25.480
+ Who's the main competitor to Aurora today?
+
+42:25.480 --> 42:28.640
+ And how do you outcompete them in the long run?
+
+42:28.640 --> 42:31.200
+ So we really focus a lot on what we're doing here.
+
+42:31.200 --> 42:34.480
+ I think that, I've said this a few times,
+
+42:34.480 --> 42:37.960
+ that this is a huge difficult problem
+
+42:37.960 --> 42:40.320
+ and it's great that a bunch of companies are tackling it
+
+42:40.320 --> 42:42.320
+ because I think it's so important for society
+
+42:42.320 --> 42:43.800
+ that somebody gets there.
+
+42:43.800 --> 42:48.800
+ So we don't spend a whole lot of time
+
+42:49.120 --> 42:51.600
+ thinking tactically about who's out there
+
+42:51.600 --> 42:55.240
+ and how do we beat that person individually.
+
+42:55.240 --> 42:58.720
+ What are we trying to do to go faster ultimately?
+
+42:59.760 --> 43:02.640
+ Well part of it is the leadership team we have
+
+43:02.640 --> 43:04.200
+ has got pretty tremendous experience.
+
+43:04.200 --> 43:06.440
+ And so we kind of understand the landscape
+
+43:06.440 --> 43:09.160
+ and understand where the cul de sacs are to some degree
+
+43:09.160 --> 43:10.980
+ and we try and avoid those.
+
+43:10.980 --> 43:14.260
+ I think there's a part of it,
+
+43:14.260 --> 43:16.260
+ just this great team we've built.
+
+43:16.260 --> 43:19.080
+ People, this is a technology and a company
+
+43:19.080 --> 43:22.320
+ that people believe in the mission of
+
+43:22.320 --> 43:23.740
+ and so it allows us to attract
+
+43:23.740 --> 43:25.740
+ just awesome people to go work.
+
+43:26.800 --> 43:29.320
+ We've got a culture I think that people appreciate
+
+43:29.320 --> 43:30.460
+ that allows them to focus,
+
+43:30.460 --> 43:33.120
+ allows them to really spend time solving problems.
+
+43:33.120 --> 43:35.900
+ And I think that keeps them energized.
+
+43:35.900 --> 43:38.940
+ And then we've invested hard,
+
+43:38.940 --> 43:43.500
+ invested heavily in the infrastructure
+
+43:43.500 --> 43:46.540
+ and architectures that we think will ultimately accelerate us.
+
+43:46.540 --> 43:50.660
+ So because of the folks we're able to bring in early on,
+
+43:50.660 --> 43:53.540
+ because of the great investors we have,
+
+43:53.540 --> 43:56.780
+ we don't spend all of our time doing demos
+
+43:56.780 --> 43:58.660
+ and kind of leaping from one demo to the next.
+
+43:58.660 --> 44:02.820
+ We've been given the freedom to invest in
+
+44:03.940 --> 44:05.500
+ infrastructure to do machine learning,
+
+44:05.500 --> 44:08.600
+ infrastructure to pull data from our on road testing,
+
+44:08.600 --> 44:11.500
+ infrastructure to use that to accelerate engineering.
+
+44:11.500 --> 44:14.480
+ And I think that early investment
+
+44:14.480 --> 44:17.340
+ and continuing investment in those kind of tools
+
+44:17.340 --> 44:19.780
+ will ultimately allow us to accelerate
+
+44:19.780 --> 44:21.940
+ and do something pretty incredible.
+
+44:21.940 --> 44:23.420
+ Chris, beautifully put.
+
+44:23.420 --> 44:24.660
+ It's a good place to end.
+
+44:24.660 --> 44:26.500
+ Thank you so much for talking today.
+
+44:26.500 --> 44:47.940
+ Thank you very much. Really enjoyed it.
+

vtt/episode_028_small.vtt

@@ -0,0 +1,2630 @@
+WEBVTT
+
+00:00.000 --> 00:03.120
+ The following is a conversation with Chris Ermsen.
+
+00:03.120 --> 00:06.040
+ He was the CTO of the Google self driving car team,
+
+00:06.040 --> 00:08.880
+ a key engineer and leader behind the Carnegie Mellon
+
+00:08.880 --> 00:11.240
+ University, autonomous vehicle entries
+
+00:11.240 --> 00:14.120
+ in the DARPA Grand Challenges and the winner
+
+00:14.120 --> 00:16.160
+ of the DARPA Urban Challenge.
+
+00:16.160 --> 00:19.480
+ Today, he's the CEO of Aurora Innovation,
+
+00:19.480 --> 00:21.360
+ an autonomous vehicle software company.
+
+00:21.360 --> 00:23.600
+ He started with Sterling Anderson,
+
+00:23.600 --> 00:26.000
+ who was the former director of Tesla Autopilot
+
+00:26.000 --> 00:30.160
+ and drew back now Uber's former autonomy and perception lead.
+
+00:30.160 --> 00:33.160
+ Chris is one of the top roboticist and autonomous vehicle
+
+00:33.160 --> 00:37.440
+ experts in the world and a long time voice of reason
+
+00:37.440 --> 00:41.320
+ in a space that is shrouded in both mystery and hype.
+
+00:41.320 --> 00:43.600
+ He both acknowledges the incredible challenges
+
+00:43.600 --> 00:46.560
+ involved in solving the problem of autonomous driving
+
+00:46.560 --> 00:49.680
+ and is working hard to solve it.
+
+00:49.680 --> 00:52.440
+ This is the Artificial Intelligence Podcast.
+
+00:52.440 --> 00:54.720
+ If you enjoy it, subscribe on YouTube,
+
+00:54.720 --> 00:57.920
+ give it five stars on iTunes, support it on Patreon,
+
+00:57.920 --> 00:59.760
+ or simply connect with me on Twitter
+
+00:59.760 --> 01:03.280
+ at Lex Freedman spelled FRID MAN.
+
+01:03.280 --> 01:09.160
+ And now, here's my conversation with Chris Ermsen.
+
+01:09.160 --> 01:11.960
+ You were part of both the DARPA Grand Challenge
+
+01:11.960 --> 01:17.040
+ and the DARPA Urban Challenge teams at CMU with Red Whitaker.
+
+01:17.040 --> 01:19.720
+ What technical or philosophical things
+
+01:19.720 --> 01:22.280
+ have you learned from these races?
+
+01:22.280 --> 01:26.640
+ I think the high order bit was that it could be done.
+
+01:26.640 --> 01:32.880
+ I think that was the thing that was incredible about the first
+
+01:32.880 --> 01:36.440
+ of the Grand Challenges, that I remember I was a grad
+
+01:36.440 --> 01:41.440
+ student at Carnegie Mellon, and there we
+
+01:41.440 --> 01:46.320
+ was kind of this dichotomy of it seemed really hard,
+
+01:46.320 --> 01:48.800
+ so that would be cool and interesting.
+
+01:48.800 --> 01:51.720
+ But at the time, we were the only robotics
+
+01:51.720 --> 01:54.960
+ institute around, and so if we went into it and fell
+
+01:54.960 --> 01:58.320
+ in our faces, that would be embarrassing.
+
+01:58.320 --> 02:01.160
+ So I think just having the will to go do it,
+
+02:01.160 --> 02:03.360
+ to try to do this thing that at the time was marked
+
+02:03.360 --> 02:07.120
+ as darn near impossible, and then after a couple of tries,
+
+02:07.120 --> 02:11.360
+ be able to actually make it happen, I think that was really
+
+02:11.360 --> 02:12.360
+ exciting.
+
+02:12.360 --> 02:15.120
+ But at which point did you believe it was possible?
+
+02:15.120 --> 02:17.000
+ Did you, from the very beginning,
+
+02:17.000 --> 02:18.360
+ did you personally, because you're
+
+02:18.360 --> 02:20.320
+ one of the lead engineers, you actually
+
+02:20.320 --> 02:21.800
+ had to do a lot of the work?
+
+02:21.800 --> 02:23.840
+ Yeah, I was the technical director there,
+
+02:23.840 --> 02:26.120
+ and did a lot of the work, along with a bunch
+
+02:26.120 --> 02:28.440
+ of other really good people.
+
+02:28.440 --> 02:29.760
+ Did I believe it could be done?
+
+02:29.760 --> 02:31.120
+ Yeah, of course.
+
+02:31.120 --> 02:33.400
+ Why would you go do something you thought was impossible,
+
+02:33.400 --> 02:34.880
+ completely impossible?
+
+02:34.880 --> 02:36.280
+ We thought it was going to be hard.
+
+02:36.280 --> 02:38.080
+ We didn't know how we're going to be able to do it.
+
+02:38.080 --> 02:42.880
+ We didn't know if we'd be able to do it the first time.
+
+02:42.880 --> 02:46.000
+ Turns out we couldn't.
+
+02:46.000 --> 02:48.400
+ That, yeah, I guess you have to.
+
+02:48.400 --> 02:52.920
+ I think there's a certain benefit to naivete,
+
+02:52.920 --> 02:55.400
+ that if you don't know how hard something really is,
+
+02:55.400 --> 02:59.560
+ you try different things, and it gives you an opportunity
+
+02:59.560 --> 03:04.080
+ that others who are wiser maybe don't have.
+
+03:04.080 --> 03:05.680
+ What were the biggest pain points?
+
+03:05.680 --> 03:09.360
+ Mechanical, sensors, hardware, software, algorithms
+
+03:09.360 --> 03:12.760
+ for mapping, localization, just general perception,
+
+03:12.760 --> 03:15.440
+ control, like hardware, software, first of all.
+
+03:15.440 --> 03:17.840
+ I think that's the joy of this field,
+
+03:17.840 --> 03:20.040
+ is that it's all hard.
+
+03:20.040 --> 03:25.160
+ And that you have to be good at each part of it.
+
+03:25.160 --> 03:32.280
+ So for the urban challenges, if I look back at it from today,
+
+03:32.280 --> 03:36.200
+ it should be easy today.
+
+03:36.200 --> 03:38.880
+ That it was a static world.
+
+03:38.880 --> 03:40.720
+ There weren't other actors moving through it.
+
+03:40.720 --> 03:42.440
+ That is what that means.
+
+03:42.440 --> 03:47.080
+ It was out in the desert, so you get really good GPS.
+
+03:47.080 --> 03:51.320
+ So that went, and we could map it roughly.
+
+03:51.320 --> 03:55.160
+ And so in retrospect now, it's within the realm of things
+
+03:55.160 --> 03:57.800
+ we could do back then.
+
+03:57.800 --> 03:59.200
+ Just actually getting the vehicle,
+
+03:59.200 --> 04:00.680
+ and there's a bunch of engineering work
+
+04:00.680 --> 04:04.200
+ to get the vehicle so that we could control and drive it.
+
+04:04.200 --> 04:09.520
+ That's still a pain today, but it was even more so back then.
+
+04:09.520 --> 04:12.920
+ And then the uncertainty of exactly what they wanted us
+
+04:12.920 --> 04:17.080
+ to do was part of the challenge as well.
+
+04:17.080 --> 04:19.360
+ Right, you didn't actually know the track hiding it.
+
+04:19.360 --> 04:21.520
+ You knew approximately, but you didn't actually
+
+04:21.520 --> 04:23.560
+ know the route that's going to be taken.
+
+04:23.560 --> 04:26.560
+ That's right, we didn't even really,
+
+04:26.560 --> 04:28.640
+ the way the rules had been described,
+
+04:28.640 --> 04:29.840
+ you had to kind of guess.
+
+04:29.840 --> 04:33.440
+ So if you think back to that challenge,
+
+04:33.440 --> 04:37.000
+ the idea was that the government would give us,
+
+04:37.000 --> 04:40.360
+ the DARPA would give us a set of waypoints
+
+04:40.360 --> 04:44.240
+ and kind of the width that you had to stay within between the line
+
+04:44.240 --> 04:46.840
+ that went between each of those waypoints.
+
+04:46.840 --> 04:49.280
+ And so the most devious thing they could have done
+
+04:49.280 --> 04:53.720
+ is set a kilometer wide corridor across a field of scrub
+
+04:53.720 --> 04:58.520
+ brush and rocks and said, go figure it out.
+
+04:58.520 --> 05:02.200
+ Fortunately, it turned into basically driving along
+
+05:02.200 --> 05:06.800
+ a set of trails, which is much more relevant to the application
+
+05:06.800 --> 05:08.760
+ they were looking for.
+
+05:08.760 --> 05:12.080
+ But no, it was a hell of a thing back in the day.
+
+05:12.080 --> 05:16.640
+ So the legend, Red, was kind of leading that effort
+
+05:16.640 --> 05:19.120
+ in terms of just broadly speaking.
+
+05:19.120 --> 05:22.040
+ So you're a leader now.
+
+05:22.040 --> 05:25.000
+ What have you learned from Red about leadership?
+
+05:25.000 --> 05:26.360
+ I think there's a couple of things.
+
+05:26.360 --> 05:30.880
+ One is go and try those really hard things.
+
+05:30.880 --> 05:34.760
+ That's where there is an incredible opportunity.
+
+05:34.760 --> 05:36.560
+ I think the other big one, though,
+
+05:36.560 --> 05:41.720
+ is to see people for who they can be, not who they are.
+
+05:41.720 --> 05:46.080
+ It's one of the deepest lessons I learned from Red,
+
+05:46.080 --> 05:51.000
+ was that he would look at undergraduates or graduate
+
+05:51.000 --> 05:56.120
+ students and empower them to be leaders,
+
+05:56.120 --> 06:01.400
+ to have responsibility, to do great things,
+
+06:01.400 --> 06:04.760
+ that I think another person might look at them and think,
+
+06:04.760 --> 06:06.600
+ oh, well, that's just an undergraduate student.
+
+06:06.600 --> 06:08.720
+ What could they know?
+
+06:08.720 --> 06:13.520
+ And so I think that trust, but verify, have confidence
+
+06:13.520 --> 06:14.880
+ in what people can become, I think,
+
+06:14.880 --> 06:16.680
+ is a really powerful thing.
+
+06:16.680 --> 06:20.480
+ So through that, let's just fast forward through the history.
+
+06:20.480 --> 06:24.200
+ Can you maybe talk through the technical evolution
+
+06:24.200 --> 06:27.480
+ of autonomous vehicle systems from the first two
+
+06:27.480 --> 06:30.920
+ Grand Challenges to the Urban Challenge to today?
+
+06:30.920 --> 06:33.600
+ Are there major shifts in your mind,
+
+06:33.600 --> 06:37.240
+ or is it the same kind of technology just made more robust?
+
+06:37.240 --> 06:40.880
+ I think there's been some big, big steps.
+
+06:40.880 --> 06:46.600
+ So for the Grand Challenge, the real technology
+
+06:46.600 --> 06:51.400
+ that unlocked that was HD mapping.
+
+06:51.400 --> 06:55.200
+ Prior to that, a lot of the off road robotics work
+
+06:55.200 --> 06:58.920
+ had been done without any real prior model of what
+
+06:58.920 --> 07:01.400
+ the vehicle was going to encounter.
+
+07:01.400 --> 07:03.960
+ And so that innovation, that the fact
+
+07:03.960 --> 07:11.320
+ that we could get decimeter resolution models,
+
+07:11.320 --> 07:13.560
+ was really a big deal.
+
+07:13.560 --> 07:17.480
+ And that allowed us to kind of bound
+
+07:17.480 --> 07:19.680
+ the complexity of the driving problem the vehicle had
+
+07:19.680 --> 07:21.040
+ and allowed it to operate at speed,
+
+07:21.040 --> 07:23.800
+ because we could assume things about the environment
+
+07:23.800 --> 07:26.400
+ that it was going to encounter.
+
+07:26.400 --> 07:31.320
+ So that was one of the big step there.
+
+07:31.320 --> 07:38.520
+ For the Urban Challenge, one of the big technological
+
+07:38.520 --> 07:41.960
+ innovations there was the multi beam LiDAR.
+
+07:41.960 --> 07:45.720
+ And be able to generate high resolution,
+
+07:45.720 --> 07:48.680
+ mid to long range 3D models the world,
+
+07:48.680 --> 07:54.120
+ and use that for understanding the world around the vehicle.
+
+07:54.120 --> 07:59.120
+ And that was really kind of a game changing technology.
+
+07:59.120 --> 08:02.880
+ And parallel with that, we saw a bunch
+
+08:02.880 --> 08:06.640
+ of other technologies that had been kind of converging
+
+08:06.640 --> 08:08.960
+ half their day in the sun.
+
+08:08.960 --> 08:16.800
+ So Bayesian estimation had been, SLAM had been a big field
+
+08:16.800 --> 08:18.600
+ in robotics.
+
+08:18.600 --> 08:20.800
+ You would go to a conference a couple of years
+
+08:20.800 --> 08:23.800
+ before that, and every paper would effectively
+
+08:23.800 --> 08:25.640
+ have SLAM somewhere in it.
+
+08:25.640 --> 08:31.560
+ And so seeing that Bayesian estimation techniques
+
+08:31.560 --> 08:34.040
+ play out on a very visible stage,
+
+08:34.040 --> 08:38.680
+ I thought that was pretty exciting to see.
+
+08:38.680 --> 08:41.760
+ And mostly SLAM was done based on LiDAR at that time?
+
+08:41.760 --> 08:42.400
+ Well, yeah.
+
+08:42.400 --> 08:46.720
+ And in fact, we weren't really doing SLAM per se in real time,
+
+08:46.720 --> 08:48.120
+ because we had a model ahead of time.
+
+08:48.120 --> 08:51.560
+ We had a roadmap, but we were doing localization.
+
+08:51.560 --> 08:54.080
+ And we were using the LiDAR or the cameras,
+
+08:54.080 --> 08:55.920
+ depending on who exactly was doing it,
+
+08:55.920 --> 08:58.080
+ to localize to a model of the world.
+
+08:58.080 --> 09:00.720
+ And I thought that was a big step
+
+09:00.720 --> 09:07.160
+ from kind of naively trusting GPS INS before that.
+
+09:07.160 --> 09:10.400
+ And again, lots of work had been going on in this field.
+
+09:10.400 --> 09:14.080
+ Certainly, this was not doing anything particularly
+
+09:14.080 --> 09:17.400
+ innovative in SLAM or in localization,
+
+09:17.400 --> 09:20.160
+ but it was seeing that technology necessary
+
+09:20.160 --> 09:21.800
+ in a real application on a big stage.
+
+09:21.800 --> 09:23.080
+ I thought it was very cool.
+
+09:23.080 --> 09:25.600
+ So for the Urban Challenge, those already maps
+
+09:25.600 --> 09:28.120
+ constructed offline in general?
+
+09:28.120 --> 09:28.600
+ OK.
+
+09:28.600 --> 09:30.920
+ And did people do that individually?
+
+09:30.920 --> 09:33.600
+ Did individual teams do it individually?
+
+09:33.600 --> 09:36.440
+ So they had their own different approaches there?
+
+09:36.440 --> 09:41.720
+ Or did everybody kind of share that information,
+
+09:41.720 --> 09:42.880
+ at least intuitively?
+
+09:42.880 --> 09:49.560
+ So DARPA gave all the teams a model of the world, a map.
+
+09:49.560 --> 09:53.720
+ And then one of the things that we had to figure out back then
+
+09:53.720 --> 09:56.720
+ was, and it's still one of these things that trips people up
+
+09:56.720 --> 10:00.240
+ today, is actually the coordinate system.
+
+10:00.240 --> 10:03.000
+ So you get a latitude, longitude.
+
+10:03.000 --> 10:05.120
+ And to so many decimal places, you
+
+10:05.120 --> 10:07.800
+ don't really care about kind of the ellipsoid of the Earth
+
+10:07.800 --> 10:09.520
+ that's being used.
+
+10:09.520 --> 10:12.720
+ But when you want to get to 10 centimeter or centimeter
+
+10:12.720 --> 10:18.480
+ resolution, you care whether the coordinate system is NADS 83
+
+10:18.480 --> 10:22.720
+ or WGS 84, or these are different ways
+
+10:22.720 --> 10:26.720
+ to describe both the kind of nonsphericalness of the Earth,
+
+10:26.720 --> 10:31.560
+ but also kind of the actually, and I think when I can't remember
+
+10:31.560 --> 10:33.560
+ which one, the tectonic shifts that are happening
+
+10:33.560 --> 10:36.920
+ and how to transform the global datum as a function of that.
+
+10:36.920 --> 10:40.400
+ So getting a map and then actually matching it
+
+10:40.400 --> 10:41.880
+ to reality to centimeter resolution,
+
+10:41.880 --> 10:44.000
+ that was kind of interesting and fun back then.
+
+10:44.000 --> 10:46.800
+ So how much work was the perception doing there?
+
+10:46.800 --> 10:52.440
+ So how much were you relying on localization based on maps
+
+10:52.440 --> 10:55.720
+ without using perception to register to the maps?
+
+10:55.720 --> 10:57.960
+ And I guess the question is how advanced
+
+10:57.960 --> 10:59.720
+ was perception at that point?
+
+10:59.720 --> 11:01.920
+ It's certainly behind where we are today.
+
+11:01.920 --> 11:05.800
+ We're more than a decade since the urban challenge.
+
+11:05.800 --> 11:13.080
+ But the core of it was there, that we were tracking vehicles.
+
+11:13.080 --> 11:15.600
+ We had to do that at 100 plus meter range
+
+11:15.600 --> 11:18.280
+ because we had to merge with other traffic.
+
+11:18.280 --> 11:21.200
+ We were using, again, Bayesian estimates
+
+11:21.200 --> 11:23.800
+ for state of these vehicles.
+
+11:23.800 --> 11:25.560
+ We had to deal with a bunch of the problems
+
+11:25.560 --> 11:28.240
+ that you think of today of predicting
+
+11:28.240 --> 11:31.040
+ where that vehicle is going to be a few seconds into the future.
+
+11:31.040 --> 11:33.680
+ We had to deal with the fact that there
+
+11:33.680 --> 11:36.000
+ were multiple hypotheses for that because a vehicle
+
+11:36.000 --> 11:37.640
+ at an intersection might be going right
+
+11:37.640 --> 11:41.440
+ or it might be going straight or it might be making a left turn.
+
+11:41.440 --> 11:44.080
+ And we had to deal with the challenge of the fact
+
+11:44.080 --> 11:47.520
+ that our behavior was going to impact the behavior
+
+11:47.520 --> 11:48.880
+ of that other operator.
+
+11:48.880 --> 11:53.400
+ And we did a lot of that in relatively naive ways.
+
+11:53.400 --> 11:54.720
+ But it kind of worked.
+
+11:54.720 --> 11:57.000
+ Still had to have some kind of assumption.
+
+11:57.000 --> 12:00.640
+ And so where does that 10 years later, where does that take us
+
+12:00.640 --> 12:04.200
+ today from that artificial city construction
+
+12:04.200 --> 12:06.920
+ to real cities to the urban environment?
+
+12:06.920 --> 12:13.600
+ Yeah, I think the biggest thing is that the actors are truly
+
+12:13.600 --> 12:18.680
+ unpredictable, that most of the time, the drivers on the road,
+
+12:18.680 --> 12:24.000
+ the other road users are out there behaving well.
+
+12:24.000 --> 12:27.040
+ But every once in a while, they're not.
+
+12:27.040 --> 12:33.320
+ The variety of other vehicles is, you have all of them.
+
+12:33.320 --> 12:35.760
+ In terms of behavior, or terms of perception, or both?
+
+12:35.760 --> 12:38.320
+ Both.
+
+12:38.320 --> 12:40.480
+ Back then, we didn't have to deal with cyclists.
+
+12:40.480 --> 12:42.800
+ We didn't have to deal with pedestrians.
+
+12:42.800 --> 12:46.240
+ Didn't have to deal with traffic lights.
+
+12:46.240 --> 12:49.360
+ The scale over which that you have to operate is now
+
+12:49.360 --> 12:52.240
+ as much larger than the airbase that we were thinking about back
+
+12:52.240 --> 12:52.720
+ then.
+
+12:52.720 --> 12:56.280
+ So what easy question?
+
+12:56.280 --> 12:59.720
+ What do you think is the hardest part about driving?
+
+12:59.720 --> 13:00.480
+ Easy question.
+
+13:00.480 --> 13:01.320
+ Yeah.
+
+13:01.320 --> 13:02.600
+ No, I'm joking.
+
+13:02.600 --> 13:07.440
+ I'm sure nothing really jumps out at you as one thing.
+
+13:07.440 --> 13:12.920
+ But in the jump from the urban challenge to the real world,
+
+13:12.920 --> 13:16.200
+ is there something that's a particular euphorcy
+
+13:16.200 --> 13:18.480
+ as a very serious, difficult challenge?
+
+13:18.480 --> 13:21.120
+ I think the most fundamental difference
+
+13:21.120 --> 13:28.960
+ is that we're doing it for real, that in that environment,
+
+13:28.960 --> 13:31.840
+ it was both a limited complexity environment,
+
+13:31.840 --> 13:33.240
+ because certain actors weren't there,
+
+13:33.240 --> 13:35.360
+ because the roads were maintained.
+
+13:35.360 --> 13:38.720
+ There were barriers keeping people separate from robots
+
+13:38.720 --> 13:40.880
+ at the time.
+
+13:40.880 --> 13:44.480
+ And it only had to work for 60 miles, which looking at it
+
+13:44.480 --> 13:48.960
+ from 2006, it had to work for 60 miles.
+
+13:48.960 --> 13:52.720
+ Looking at it from now, we want things
+
+13:52.720 --> 13:57.200
+ that will go and drive for half a million miles.
+
+13:57.200 --> 14:00.960
+ And it's just a different game.
+
+14:00.960 --> 14:06.080
+ So how important, you said Lyder came into the game early on,
+
+14:06.080 --> 14:08.880
+ and it's really the primary driver of autonomous vehicles
+
+14:08.880 --> 14:10.240
+ today as a sensor.
+
+14:10.240 --> 14:12.880
+ So how important is the role of Lyder in the sensor suite
+
+14:12.880 --> 14:14.760
+ in the near term?
+
+14:14.760 --> 14:18.680
+ So I think it's essential.
+
+14:18.680 --> 14:20.520
+ But I also believe that cameras are essential,
+
+14:20.520 --> 14:22.160
+ and I believe the radar is essential.
+
+14:22.160 --> 14:27.400
+ I think that you really need to use the composition of data
+
+14:27.400 --> 14:28.920
+ from these different sensors if you
+
+14:28.920 --> 14:32.600
+ want the thing to really be robust.
+
+14:32.600 --> 14:35.440
+ The question I want to ask, let's see if we can untangle it,
+
+14:35.440 --> 14:40.240
+ is what are your thoughts on the Elon Musk provocative statement
+
+14:40.240 --> 14:45.840
+ that Lyder is a crutch, that is a kind of, I guess,
+
+14:45.840 --> 14:49.600
+ growing pains, and that much of the perception
+
+14:49.600 --> 14:52.160
+ task can be done with cameras?
+
+14:52.160 --> 14:56.920
+ So I think it is undeniable that people walk around
+
+14:56.920 --> 14:59.680
+ without lasers in their foreheads,
+
+14:59.680 --> 15:01.840
+ and they can get into vehicles and drive them.
+
+15:01.840 --> 15:05.560
+ And so there's an existence proof
+
+15:05.560 --> 15:10.840
+ that you can drive using passive vision.
+
+15:10.840 --> 15:12.680
+ No doubt, can't argue with that.
+
+15:12.680 --> 15:14.320
+ In terms of sensors, yeah.
+
+15:14.320 --> 15:14.800
+ So there's proof.
+
+15:14.800 --> 15:15.960
+ Yes, in terms of sensors, right?
+
+15:15.960 --> 15:18.720
+ So there's an example that we all
+
+15:18.720 --> 15:23.280
+ go do it at many of us every day.
+
+15:23.280 --> 15:28.200
+ In terms of Lyder being a crutch, sure.
+
+15:28.200 --> 15:33.080
+ But in the same way that the combustion engine
+
+15:33.080 --> 15:35.240
+ was a crutch on the path to an electric vehicle,
+
+15:35.240 --> 15:40.840
+ in the same way that any technology ultimately gets
+
+15:40.840 --> 15:44.640
+ replaced by some superior technology in the future.
+
+15:44.640 --> 15:47.720
+ And really, the way that I look at this
+
+15:47.720 --> 15:51.720
+ is that the way we get around on the ground, the way
+
+15:51.720 --> 15:55.280
+ that we use transportation is broken.
+
+15:55.280 --> 15:59.720
+ And that we have this, I think the number I saw this morning,
+
+15:59.720 --> 16:04.040
+ 37,000 Americans killed last year on our roads.
+
+16:04.040 --> 16:05.360
+ And that's just not acceptable.
+
+16:05.360 --> 16:09.440
+ And so any technology that we can bring to bear
+
+16:09.440 --> 16:12.840
+ that accelerates this technology, self driving technology,
+
+16:12.840 --> 16:15.720
+ coming to market and saving lives,
+
+16:15.720 --> 16:18.280
+ is technology we should be using.
+
+16:18.280 --> 16:24.040
+ And it feels just arbitrary to say, well, I'm not
+
+16:24.040 --> 16:27.800
+ OK with using lasers, because that's whatever.
+
+16:27.800 --> 16:30.760
+ But I am OK with using an 8 megapixel camera
+
+16:30.760 --> 16:32.880
+ or a 16 megapixel camera.
+
+16:32.880 --> 16:34.640
+ These are just bits of technology,
+
+16:34.640 --> 16:36.880
+ and we should be taking the best technology from the tool
+
+16:36.880 --> 16:41.600
+ bin that allows us to go and solve a problem.
+
+16:41.600 --> 16:45.160
+ The question I often talk to, well, obviously you do as well,
+
+16:45.160 --> 16:48.320
+ to automotive companies.
+
+16:48.320 --> 16:51.880
+ And if there's one word that comes up more often than anything,
+
+16:51.880 --> 16:55.320
+ it's cost and drive costs down.
+
+16:55.320 --> 17:01.440
+ So while it's true that it's a tragic number, the 37,000,
+
+17:01.440 --> 17:04.880
+ the question is, and I'm not the one asking this question,
+
+17:04.880 --> 17:07.160
+ because I hate this question, but we
+
+17:07.160 --> 17:11.680
+ want to find the cheapest sensor suite that
+
+17:11.680 --> 17:13.400
+ creates a safe vehicle.
+
+17:13.400 --> 17:18.240
+ So in that uncomfortable trade off,
+
+17:18.240 --> 17:23.680
+ do you foresee lidar coming down in cost in the future?
+
+17:23.680 --> 17:28.000
+ Or do you see a day where level 4 autonomy is possible
+
+17:28.000 --> 17:29.880
+ without lidar?
+
+17:29.880 --> 17:32.880
+ I see both of those, but it's really a matter of time.
+
+17:32.880 --> 17:35.080
+ And I think, really, maybe I would
+
+17:35.080 --> 17:38.760
+ talk to the question you asked about the cheapest sensor.
+
+17:38.760 --> 17:40.440
+ I don't think that's actually what you want.
+
+17:40.440 --> 17:45.720
+ What you want is a sensor suite that is economically viable.
+
+17:45.720 --> 17:49.480
+ And then after that, everything is about margin
+
+17:49.480 --> 17:52.320
+ and driving cost out of the system.
+
+17:52.320 --> 17:55.400
+ What you also want is a sensor suite that works.
+
+17:55.400 --> 18:01.280
+ And so it's great to tell a story about how it would be better
+
+18:01.280 --> 18:04.560
+ to have a self driving system with a $50 sensor instead
+
+18:04.560 --> 18:08.720
+ of a $500 sensor.
+
+18:08.720 --> 18:11.560
+ But if the $500 sensor makes it work and the $50 sensor
+
+18:11.560 --> 18:15.680
+ doesn't work, who cares?
+
+18:15.680 --> 18:21.680
+ So long as you can actually have an economic opportunity there.
+
+18:21.680 --> 18:23.760
+ And the economic opportunity is important,
+
+18:23.760 --> 18:27.800
+ because that's how you actually have a sustainable business.
+
+18:27.800 --> 18:30.440
+ And that's how you can actually see this come to scale
+
+18:30.440 --> 18:32.520
+ and be out in the world.
+
+18:32.520 --> 18:36.400
+ And so when I look at lidar, I see
+
+18:36.400 --> 18:41.200
+ a technology that has no underlying fundamentally expense
+
+18:41.200 --> 18:43.240
+ to it, fundamental expense to it.
+
+18:43.240 --> 18:46.120
+ It's going to be more expensive than an imager,
+
+18:46.120 --> 18:51.400
+ because CMOS processes or FAP processes
+
+18:51.400 --> 18:56.200
+ are dramatically more scalable than mechanical processes.
+
+18:56.200 --> 18:58.160
+ But we still should be able to drive cost
+
+18:58.160 --> 19:00.440
+ out substantially on that side.
+
+19:00.440 --> 19:05.880
+ And then I also do think that with the right business model,
+
+19:05.880 --> 19:08.440
+ you can absorb more, certainly more cost
+
+19:08.440 --> 19:09.480
+ on the below materials.
+
+19:09.480 --> 19:12.600
+ Yeah, if the sensor suite works, extra value is provided.
+
+19:12.600 --> 19:15.480
+ Thereby, you don't need to drive cost down to zero.
+
+19:15.480 --> 19:17.120
+ It's a basic economics.
+
+19:17.120 --> 19:18.840
+ You've talked about your intuition
+
+19:18.840 --> 19:22.720
+ at level two autonomy is problematic because
+
+19:22.720 --> 19:27.280
+ of the human factor of vigilance, decrement, complacency,
+
+19:27.280 --> 19:29.600
+ overtrust, and so on, just us being human.
+
+19:29.600 --> 19:33.000
+ With the overtrust system, we start doing even more
+
+19:33.000 --> 19:36.480
+ so partaking in the secondary activities like smartphone
+
+19:36.480 --> 19:38.720
+ and so on.
+
+19:38.720 --> 19:42.960
+ Have your views evolved on this point in either direction?
+
+19:42.960 --> 19:44.760
+ Can you speak to it?
+
+19:44.760 --> 19:48.240
+ So I want to be really careful, because sometimes this
+
+19:48.240 --> 19:53.000
+ gets twisted in a way that I certainly didn't intend.
+
+19:53.000 --> 19:59.360
+ So active safety systems are a really important technology
+
+19:59.360 --> 20:03.400
+ that we should be pursuing and integrating into vehicles.
+
+20:03.400 --> 20:05.680
+ And there's an opportunity in the near term
+
+20:05.680 --> 20:09.400
+ to reduce accidents, reduce fatalities, and that's
+
+20:09.400 --> 20:13.400
+ and we should be pushing on that.
+
+20:13.400 --> 20:17.280
+ Level two systems are systems where
+
+20:17.280 --> 20:19.480
+ the vehicle is controlling two axes,
+
+20:19.480 --> 20:24.800
+ so breaking and thrall slash steering.
+
+20:24.800 --> 20:27.200
+ And I think there are variants of level two systems that
+
+20:27.200 --> 20:30.200
+ are supporting the driver that absolutely we
+
+20:30.200 --> 20:32.560
+ should encourage to be out there.
+
+20:32.560 --> 20:37.920
+ Where I think there's a real challenge is in the human factors
+
+20:37.920 --> 20:40.800
+ part around this and the misconception
+
+20:40.800 --> 20:44.920
+ from the public around the capability set that that enables
+
+20:44.920 --> 20:48.000
+ and the trust that they should have in it.
+
+20:48.000 --> 20:53.880
+ And that is where I'm actually incrementally more
+
+20:53.880 --> 20:55.800
+ concerned around level three systems
+
+20:55.800 --> 20:59.960
+ and how exactly a level two system is marketed and delivered
+
+20:59.960 --> 21:03.240
+ and how much effort people have put into those human factors.
+
+21:03.240 --> 21:07.000
+ So I still believe several things around this.
+
+21:07.000 --> 21:10.760
+ One is people will over trust the technology.
+
+21:10.760 --> 21:12.720
+ We've seen over the last few weeks
+
+21:12.720 --> 21:16.280
+ a spate of people sleeping in their Tesla.
+
+21:16.280 --> 21:23.240
+ I watched an episode last night of Trevor Noah talking
+
+21:23.240 --> 21:27.160
+ about this, and this is a smart guy
+
+21:27.160 --> 21:31.040
+ who has a lot of resources at his disposal describing
+
+21:31.040 --> 21:32.880
+ a Tesla as a self driving car.
+
+21:32.880 --> 21:35.640
+ And that why shouldn't people be sleeping in their Tesla?
+
+21:35.640 --> 21:38.800
+ It's like, well, because it's not a self driving car
+
+21:38.800 --> 21:41.120
+ and it is not intended to be.
+
+21:41.120 --> 21:48.400
+ And these people will almost certainly die at some point
+
+21:48.400 --> 21:50.400
+ or hurt other people.
+
+21:50.400 --> 21:52.640
+ And so we need to really be thoughtful about how
+
+21:52.640 --> 21:56.280
+ that technology is described and brought to market.
+
+21:56.280 --> 22:00.760
+ I also think that because of the economic issue,
+
+22:00.760 --> 22:03.320
+ economic challenges we were just talking about,
+
+22:03.320 --> 22:06.960
+ that technology path will, these level two driver system
+
+22:06.960 --> 22:08.400
+ systems, that technology path will
+
+22:08.400 --> 22:11.560
+ diverge from the technology path that we
+
+22:11.560 --> 22:15.800
+ need to be on to actually deliver truly self driving
+
+22:15.800 --> 22:19.120
+ vehicles, ones where you can get in it and sleep
+
+22:19.120 --> 22:21.480
+ and have the equivalent or better safety
+
+22:21.480 --> 22:24.600
+ than a human driver behind the wheel.
+
+22:24.600 --> 22:28.440
+ Because, again, the economics are very different
+
+22:28.440 --> 22:29.800
+ in those two worlds.
+
+22:29.800 --> 22:32.720
+ And so that leads to divergent technology.
+
+22:32.720 --> 22:36.920
+ So you just don't see the economics of gradually
+
+22:36.920 --> 22:41.520
+ increasing from level two and doing so quickly enough
+
+22:41.520 --> 22:44.400
+ to where it doesn't cost safety, critical safety concerns.
+
+22:44.400 --> 22:48.600
+ You believe that it needs to diverge at this point
+
+22:48.600 --> 22:50.600
+ into different, basically different routes.
+
+22:50.600 --> 22:53.760
+ And really that comes back to what
+
+22:53.760 --> 22:56.840
+ are those L2 and L1 systems doing?
+
+22:56.840 --> 22:59.800
+ And they are driver assistance functions
+
+22:59.800 --> 23:04.360
+ where the people that are marketing that responsibly
+
+23:04.360 --> 23:07.960
+ are being very clear and putting human factors in place
+
+23:07.960 --> 23:12.400
+ such that the driver is actually responsible for the vehicle
+
+23:12.400 --> 23:15.200
+ and that the technology is there to support the driver.
+
+23:15.200 --> 23:19.880
+ And the safety cases that are built around those
+
+23:19.880 --> 23:24.320
+ are dependent on that driver attention and attentiveness.
+
+23:24.320 --> 23:30.360
+ And at that point, you can kind of give up, to some degree,
+
+23:30.360 --> 23:34.280
+ for economic reasons, you can give up on, say, false negatives.
+
+23:34.280 --> 23:36.200
+ And so the way to think about this
+
+23:36.200 --> 23:40.760
+ is for a four collision mitigation braking system,
+
+23:40.760 --> 23:45.080
+ if half the times the driver missed a vehicle in front of it,
+
+23:45.080 --> 23:47.640
+ it hit the brakes and brought the vehicle to a stop,
+
+23:47.640 --> 23:51.200
+ that would be an incredible, incredible advance
+
+23:51.200 --> 23:52.960
+ in safety on our roads, right?
+
+23:52.960 --> 23:55.080
+ That would be equivalent to seatbelts.
+
+23:55.080 --> 23:57.560
+ But it would mean that if that vehicle wasn't being monitored,
+
+23:57.560 --> 24:00.560
+ it would hit one out of two cars.
+
+24:00.560 --> 24:05.080
+ And so economically, that's a perfectly good solution
+
+24:05.080 --> 24:06.200
+ for a driver assistance system.
+
+24:06.200 --> 24:07.360
+ What you should do at that point,
+
+24:07.360 --> 24:09.200
+ if you can get it to work 50% of the time,
+
+24:09.200 --> 24:11.040
+ is drive the cost out of that so you can get it
+
+24:11.040 --> 24:13.320
+ on as many vehicles as possible.
+
+24:13.320 --> 24:16.920
+ But driving the cost out of it doesn't drive up performance
+
+24:16.920 --> 24:18.840
+ on the false negative case.
+
+24:18.840 --> 24:21.480
+ And so you'll continue to not have a technology
+
+24:21.480 --> 24:25.720
+ that could really be available for a self driven vehicle.
+
+24:25.720 --> 24:28.480
+ So clearly the communication,
+
+24:28.480 --> 24:31.640
+ and this probably applies to all four vehicles as well,
+
+24:31.640 --> 24:34.440
+ the marketing and the communication
+
+24:34.440 --> 24:37.080
+ of what the technology is actually capable of,
+
+24:37.080 --> 24:38.440
+ how hard it is, how easy it is,
+
+24:38.440 --> 24:41.040
+ all that kind of stuff is highly problematic.
+
+24:41.040 --> 24:45.680
+ So say everybody in the world was perfectly communicated
+
+24:45.680 --> 24:48.400
+ and were made to be completely aware
+
+24:48.400 --> 24:50.040
+ of every single technology out there,
+
+24:50.040 --> 24:52.880
+ what it's able to do.
+
+24:52.880 --> 24:54.160
+ What's your intuition?
+
+24:54.160 --> 24:56.920
+ And now we're maybe getting into philosophical ground.
+
+24:56.920 --> 25:00.040
+ Is it possible to have a level two vehicle
+
+25:00.040 --> 25:03.280
+ where we don't overtrust it?
+
+25:04.720 --> 25:05.840
+ I don't think so.
+
+25:05.840 --> 25:10.840
+ If people truly understood the risks and internalized it,
+
+25:11.200 --> 25:14.320
+ then sure you could do that safely,
+
+25:14.320 --> 25:16.200
+ but that's a world that doesn't exist.
+
+25:16.200 --> 25:17.560
+ The people are going to,
+
+25:19.440 --> 25:20.800
+ if the facts are put in front of them,
+
+25:20.800 --> 25:24.480
+ they're gonna then combine that with their experience.
+
+25:24.480 --> 25:28.400
+ And let's say they're using an L2 system
+
+25:28.400 --> 25:31.040
+ and they go up and down the one on one every day
+
+25:31.040 --> 25:32.800
+ and they do that for a month
+
+25:32.800 --> 25:35.200
+ and it just worked every day for a month.
+
+25:36.320 --> 25:37.400
+ Like that's pretty compelling.
+
+25:37.400 --> 25:41.880
+ At that point, just even if you know the statistics,
+
+25:41.880 --> 25:43.520
+ you're like, well, I don't know,
+
+25:43.520 --> 25:44.840
+ maybe there's something a little funny about those.
+
+25:44.840 --> 25:47.000
+ Maybe they're driving in difficult places.
+
+25:47.000 --> 25:49.960
+ Like I've seen it with my own eyes, it works.
+
+25:49.960 --> 25:52.480
+ And the problem is that that sample size that they have,
+
+25:52.480 --> 25:54.000
+ so it's 30 miles up and down,
+
+25:54.000 --> 25:58.800
+ so 60 miles times 30 days, so 60, 180, 1,800 miles.
+
+26:01.720 --> 26:05.240
+ That's a drop in the bucket compared to the one,
+
+26:05.240 --> 26:07.640
+ what 85 million miles between fatalities.
+
+26:07.640 --> 26:11.400
+ And so they don't really have a true estimate
+
+26:11.400 --> 26:14.440
+ based on their personal experience of the real risks,
+
+26:14.440 --> 26:15.640
+ but they're gonna trust it anyway,
+
+26:15.640 --> 26:17.720
+ because it's hard not to, it worked for a month.
+
+26:17.720 --> 26:18.640
+ What's gonna change?
+
+26:18.640 --> 26:21.600
+ So even if you start a perfect understanding of the system,
+
+26:21.600 --> 26:24.160
+ your own experience will make it drift.
+
+26:24.160 --> 26:25.920
+ I mean, that's a big concern.
+
+26:25.920 --> 26:29.480
+ Over a year, over two years even, it doesn't have to be months.
+
+26:29.480 --> 26:33.720
+ And I think that as this technology moves from,
+
+26:35.440 --> 26:37.800
+ what I would say is kind of the more technology savvy
+
+26:37.800 --> 26:41.480
+ ownership group to the mass market,
+
+26:41.480 --> 26:44.640
+ you may be able to have some of those folks
+
+26:44.640 --> 26:46.320
+ who are really familiar with technology,
+
+26:46.320 --> 26:48.880
+ they may be able to internalize it better.
+
+26:48.880 --> 26:50.840
+ And you're kind of immunization
+
+26:50.840 --> 26:53.400
+ against this kind of false risk assessment
+
+26:53.400 --> 26:56.960
+ might last longer, but as folks who aren't as savvy
+
+26:56.960 --> 27:00.200
+ about that read the material
+
+27:00.200 --> 27:02.200
+ and they compare that to their personal experience,
+
+27:02.200 --> 27:08.200
+ I think there that it's gonna move more quickly.
+
+27:08.200 --> 27:11.320
+ So your work, the program that you've created at Google
+
+27:11.320 --> 27:16.320
+ and now at Aurora is focused more on the second path
+
+27:16.640 --> 27:18.520
+ of creating full autonomy.
+
+27:18.520 --> 27:20.920
+ So it's such a fascinating,
+
+27:21.800 --> 27:24.600
+ I think it's one of the most interesting AI problems
+
+27:24.600 --> 27:25.640
+ of the century, right?
+
+27:25.640 --> 27:28.320
+ It's a, I just talked to a lot of people,
+
+27:28.320 --> 27:30.400
+ just regular people, I don't know, my mom
+
+27:30.400 --> 27:33.840
+ about autonomous vehicles and you begin to grapple
+
+27:33.840 --> 27:38.080
+ with ideas of giving your life control over to a machine.
+
+27:38.080 --> 27:40.040
+ It's philosophically interesting,
+
+27:40.040 --> 27:41.760
+ it's practically interesting.
+
+27:41.760 --> 27:43.720
+ So let's talk about safety.
+
+27:43.720 --> 27:46.240
+ How do you think, we demonstrate,
+
+27:46.240 --> 27:47.880
+ you've spoken about metrics in the past,
+
+27:47.880 --> 27:51.880
+ how do you think we demonstrate to the world
+
+27:51.880 --> 27:56.160
+ that an autonomous vehicle, an Aurora system is safe?
+
+27:56.160 --> 27:57.320
+ This is one where it's difficult
+
+27:57.320 --> 27:59.280
+ because there isn't a sound bite answer.
+
+27:59.280 --> 28:04.280
+ That we have to show a combination of work
+
+28:05.960 --> 28:08.360
+ that was done diligently and thoughtfully.
+
+28:08.360 --> 28:10.840
+ And this is where something like a functional safety process
+
+28:10.840 --> 28:14.360
+ as part of that is like, here's the way we did the work.
+
+28:15.320 --> 28:17.200
+ That means that we were very thorough.
+
+28:17.200 --> 28:20.560
+ So, if you believe that we, what we said about,
+
+28:20.560 --> 28:21.480
+ this is the way we did it,
+
+28:21.480 --> 28:23.440
+ then you can have some confidence that we were thorough
+
+28:23.440 --> 28:27.000
+ in the engineering work we put into the system.
+
+28:27.000 --> 28:30.160
+ And then on top of that, to kind of demonstrate
+
+28:30.160 --> 28:32.000
+ that we weren't just thorough,
+
+28:32.000 --> 28:34.000
+ we were actually good at what we did.
+
+28:35.320 --> 28:38.240
+ There'll be a kind of a collection of evidence
+
+28:38.240 --> 28:40.480
+ in terms of demonstrating that the capabilities
+
+28:40.480 --> 28:43.960
+ work the way we thought they did, statistically
+
+28:43.960 --> 28:47.200
+ and to whatever degree we can demonstrate that
+
+28:48.200 --> 28:50.320
+ both in some combination of simulation,
+
+28:50.320 --> 28:54.720
+ some combination of unit testing and decomposition testing,
+
+28:54.720 --> 28:57.000
+ and then some part of it will be on road data.
+
+28:58.200 --> 29:03.200
+ And I think the way we'll ultimately convey this
+
+29:03.320 --> 29:06.800
+ to the public is there'll be clearly some conversation
+
+29:06.800 --> 29:08.240
+ with the public about it,
+
+29:08.240 --> 29:12.080
+ but we'll kind of invoke the kind of the trusted nodes
+
+29:12.080 --> 29:14.360
+ and that we'll spend more time being able to go
+
+29:14.360 --> 29:17.280
+ into more depth with folks like NHTSA
+
+29:17.280 --> 29:19.760
+ and other federal and state regulatory bodies
+
+29:19.760 --> 29:22.600
+ and kind of given that they are operating
+
+29:22.600 --> 29:25.120
+ in the public interest and they're trusted
+
+29:26.240 --> 29:28.680
+ that if we can show enough work to them
+
+29:28.680 --> 29:30.040
+ that they're convinced,
+
+29:30.040 --> 29:33.840
+ then I think we're in a pretty good place.
+
+29:33.840 --> 29:35.040
+ That means that you work with people
+
+29:35.040 --> 29:36.960
+ that are essentially experts at safety
+
+29:36.960 --> 29:39.040
+ to try to discuss and show,
+
+29:39.040 --> 29:41.800
+ do you think the answer is probably no,
+
+29:41.800 --> 29:44.360
+ but just in case, do you think there exists a metric?
+
+29:44.360 --> 29:46.360
+ So currently people have been using
+
+29:46.360 --> 29:48.200
+ a number of disengagement.
+
+29:48.200 --> 29:50.160
+ And it quickly turns into a marketing scheme
+
+29:50.160 --> 29:54.320
+ to sort of you alter the experiments you run to.
+
+29:54.320 --> 29:56.320
+ I think you've spoken that you don't like.
+
+29:56.320 --> 29:57.160
+ Don't love it.
+
+29:57.160 --> 29:59.720
+ No, in fact, I was on the record telling DMV
+
+29:59.720 --> 30:02.000
+ that I thought this was not a great metric.
+
+30:02.000 --> 30:05.360
+ Do you think it's possible to create a metric,
+
+30:05.360 --> 30:09.480
+ a number that could demonstrate safety
+
+30:09.480 --> 30:12.400
+ outside of fatalities?
+
+30:12.400 --> 30:16.640
+ So I do and I think that it won't be just one number.
+
+30:16.640 --> 30:21.320
+ So as we are internally grappling with this
+
+30:21.320 --> 30:23.600
+ and at some point we'll be able to talk
+
+30:23.600 --> 30:25.080
+ more publicly about it,
+
+30:25.080 --> 30:28.560
+ is how do we think about human performance
+
+30:28.560 --> 30:32.200
+ in different tasks, say detecting traffic lights
+
+30:32.200 --> 30:36.240
+ or safely making a left turn across traffic?
+
+30:37.720 --> 30:40.040
+ And what do we think the failure rates
+
+30:40.040 --> 30:42.520
+ are for those different capabilities for people?
+
+30:42.520 --> 30:44.760
+ And then demonstrating to ourselves
+
+30:44.760 --> 30:48.480
+ and then ultimately folks in regulatory role
+
+30:48.480 --> 30:50.760
+ and then ultimately the public,
+
+30:50.760 --> 30:52.400
+ that we have confidence that our system
+
+30:52.400 --> 30:54.800
+ will work better than that.
+
+30:54.800 --> 30:57.040
+ And so these individual metrics
+
+30:57.040 --> 31:00.720
+ will kind of tell a compelling story ultimately.
+
+31:01.760 --> 31:03.920
+ I do think at the end of the day,
+
+31:03.920 --> 31:06.640
+ what we care about in terms of safety
+
+31:06.640 --> 31:11.640
+ is life saved and injuries reduced.
+
+31:11.640 --> 31:15.320
+ And then ultimately kind of casualty dollars
+
+31:16.440 --> 31:19.360
+ that people aren't having to pay to get their car fixed.
+
+31:19.360 --> 31:22.680
+ And I do think that in aviation,
+
+31:22.680 --> 31:25.880
+ they look at a kind of an event pyramid
+
+31:25.880 --> 31:28.600
+ where a crash is at the top of that
+
+31:28.600 --> 31:30.440
+ and that's the worst event obviously.
+
+31:30.440 --> 31:34.240
+ And then there's injuries and near miss events and whatnot
+
+31:34.240 --> 31:37.320
+ and violation of operating procedures.
+
+31:37.320 --> 31:40.160
+ And you kind of build a statistical model
+
+31:40.160 --> 31:44.440
+ of the relevance of the low severity things
+
+31:44.440 --> 31:45.280
+ and the high severity things.
+
+31:45.280 --> 31:46.120
+ And I think that's something
+
+31:46.120 --> 31:48.240
+ where we'll be able to look at as well
+
+31:48.240 --> 31:51.920
+ because an event per 85 million miles
+
+31:51.920 --> 31:54.480
+ is statistically a difficult thing
+
+31:54.480 --> 31:59.440
+ even at the scale of the US to kind of compare directly.
+
+31:59.440 --> 32:02.280
+ And that event, the fatality that's connected
+
+32:02.280 --> 32:07.280
+ to an autonomous vehicle is significantly,
+
+32:07.480 --> 32:09.160
+ at least currently magnified
+
+32:09.160 --> 32:12.320
+ in the amount of attention you get.
+
+32:12.320 --> 32:15.080
+ So that speaks to public perception.
+
+32:15.080 --> 32:16.720
+ I think the most popular topic
+
+32:16.720 --> 32:19.520
+ about autonomous vehicles in the public
+
+32:19.520 --> 32:23.080
+ is the trolley problem formulation, right?
+
+32:23.080 --> 32:27.040
+ Which has, let's not get into that too much
+
+32:27.040 --> 32:29.600
+ but is misguided in many ways.
+
+32:29.600 --> 32:32.320
+ But it speaks to the fact that people are grappling
+
+32:32.320 --> 32:36.160
+ with this idea of giving control over to a machine.
+
+32:36.160 --> 32:41.160
+ So how do you win the hearts and minds of the people
+
+32:41.560 --> 32:43.600
+ that autonomy is something
+
+32:43.600 --> 32:45.480
+ that could be a part of their lives?
+
+32:45.480 --> 32:47.640
+ I think you let them experience it, right?
+
+32:47.640 --> 32:50.440
+ I think it's right.
+
+32:50.440 --> 32:52.720
+ I think people should be skeptical.
+
+32:52.720 --> 32:55.680
+ I think people should ask questions.
+
+32:55.680 --> 32:57.000
+ I think they should doubt
+
+32:58.040 --> 33:00.960
+ because this is something new and different.
+
+33:00.960 --> 33:01.960
+ They haven't touched it yet.
+
+33:01.960 --> 33:03.680
+ And I think it's perfectly reasonable.
+
+33:03.680 --> 33:07.360
+ And but at the same time,
+
+33:07.360 --> 33:09.360
+ it's clear there's an opportunity to make the road safer.
+
+33:09.360 --> 33:12.480
+ It's clear that we can improve access to mobility.
+
+33:12.480 --> 33:15.160
+ It's clear that we can reduce the cost of mobility.
+
+33:16.680 --> 33:19.520
+ And that once people try that
+
+33:19.520 --> 33:22.800
+ and understand that it's safe
+
+33:22.800 --> 33:24.480
+ and are able to use in their daily lives,
+
+33:24.480 --> 33:28.080
+ I think it's one of these things that will just be obvious.
+
+33:28.080 --> 33:32.280
+ And I've seen this practically in demonstrations
+
+33:32.280 --> 33:35.640
+ that I've given where I've had people come in
+
+33:35.640 --> 33:38.600
+ and they're very skeptical.
+
+33:38.600 --> 33:39.960
+ And they get in the vehicle.
+
+33:39.960 --> 33:42.640
+ My favorite one is taking somebody out on the freeway
+
+33:42.640 --> 33:46.080
+ and we're on the one on one driving at 65 miles an hour.
+
+33:46.080 --> 33:48.560
+ And after 10 minutes, they kind of turn and ask,
+
+33:48.560 --> 33:49.560
+ is that all it does?
+
+33:49.560 --> 33:52.160
+ And you're like, it's self driving car.
+
+33:52.160 --> 33:54.920
+ I'm not sure exactly what you thought it would do, right?
+
+33:54.920 --> 33:57.960
+ But it becomes mundane,
+
+33:58.920 --> 34:01.560
+ which is exactly what you want to technology
+
+34:01.560 --> 34:02.760
+ like this to be, right?
+
+34:02.760 --> 34:04.680
+ We don't really...
+
+34:04.680 --> 34:07.320
+ When I turn the light switch on in here,
+
+34:07.320 --> 34:12.040
+ I don't think about the complexity of those electrons
+
+34:12.040 --> 34:14.240
+ being pushed down a wire from wherever it was
+
+34:14.240 --> 34:15.880
+ and being generated.
+
+34:15.880 --> 34:19.120
+ It's like, I just get annoyed if it doesn't work, right?
+
+34:19.120 --> 34:21.440
+ And what I value is the fact
+
+34:21.440 --> 34:23.120
+ that I can do other things in this space.
+
+34:23.120 --> 34:24.600
+ I can see my colleagues.
+
+34:24.600 --> 34:26.200
+ I can read stuff on a paper.
+
+34:26.200 --> 34:29.240
+ I can not be afraid of the dark.
+
+34:29.240 --> 34:32.840
+ And I think that's what we want this technology to be like
+
+34:32.840 --> 34:34.160
+ is it's in the background
+
+34:34.160 --> 34:36.520
+ and people get to have those life experiences
+
+34:36.520 --> 34:37.880
+ and do so safely.
+
+34:37.880 --> 34:41.600
+ So putting this technology in the hands of people
+
+34:41.600 --> 34:45.800
+ speaks to scale of deployment, right?
+
+34:45.800 --> 34:50.360
+ So what do you think the dreaded question about the future
+
+34:50.360 --> 34:52.840
+ because nobody can predict the future?
+
+34:52.840 --> 34:57.080
+ But just maybe speak poetically about
+
+34:57.080 --> 35:00.600
+ when do you think we'll see a large scale deployment
+
+35:00.600 --> 35:05.600
+ of autonomous vehicles, 10,000, those kinds of numbers.
+
+35:06.360 --> 35:08.280
+ We'll see that within 10 years.
+
+35:09.280 --> 35:10.600
+ I'm pretty confident.
+
+35:10.600 --> 35:11.920
+ We...
+
+35:13.920 --> 35:15.920
+ What's an impressive scale?
+
+35:15.920 --> 35:19.000
+ What moment, so you've done the DARPA Challenge
+
+35:19.000 --> 35:20.240
+ where there's one vehicle,
+
+35:20.240 --> 35:22.000
+ at which moment does it become,
+
+35:22.000 --> 35:23.720
+ wow, this is serious scale?
+
+35:23.720 --> 35:27.960
+ So I think the moment it gets serious is when
+
+35:27.960 --> 35:32.040
+ we really do have a driverless vehicle
+
+35:32.040 --> 35:33.880
+ operating on public roads
+
+35:34.760 --> 35:37.760
+ and that we can do that kind of continuously.
+
+35:37.760 --> 35:38.640
+ Without a safety driver?
+
+35:38.640 --> 35:40.240
+ Without a safety driver in the vehicle.
+
+35:40.240 --> 35:41.320
+ I think at that moment,
+
+35:41.320 --> 35:44.160
+ we've kind of crossed the zero to one threshold.
+
+35:45.720 --> 35:50.000
+ And then it is about how do we continue to scale that?
+
+35:50.000 --> 35:53.720
+ How do we build the right business models?
+
+35:53.720 --> 35:56.040
+ How do we build the right customer experience around it
+
+35:56.040 --> 35:59.680
+ so that it is actually a useful product out in the world?
+
+36:00.720 --> 36:03.360
+ And I think that is really,
+
+36:03.360 --> 36:05.720
+ at that point, it moves from a,
+
+36:05.720 --> 36:08.960
+ what is this kind of mixed science engineering project
+
+36:08.960 --> 36:12.120
+ into engineering and commercialization
+
+36:12.120 --> 36:15.600
+ and really starting to deliver on the value
+
+36:15.600 --> 36:18.000
+ that we all see here.
+
+36:18.000 --> 36:20.680
+ And actually making that real in the world.
+
+36:20.680 --> 36:22.240
+ What do you think that deployment looks like?
+
+36:22.240 --> 36:24.920
+ Where do we first see the inkling of
+
+36:24.920 --> 36:28.600
+ no safety driver, one or two cars here and there?
+
+36:28.600 --> 36:29.760
+ Is it on the highway?
+
+36:29.760 --> 36:33.200
+ Is it in specific routes in the urban environment?
+
+36:33.200 --> 36:36.960
+ I think it's gonna be urban, suburban type environments.
+
+36:37.920 --> 36:38.920
+ You know, with Aurora,
+
+36:38.920 --> 36:41.560
+ when we thought about how to tackle this,
+
+36:42.400 --> 36:45.040
+ it was kind of invoke to think about trucking
+
+36:46.000 --> 36:47.760
+ as opposed to urban driving.
+
+36:47.760 --> 36:51.240
+ And again, the human intuition around this
+
+36:51.240 --> 36:55.360
+ is that freeways are easier to drive on
+
+36:57.040 --> 36:59.240
+ because everybody's kind of going in the same direction
+
+36:59.240 --> 37:01.560
+ and lanes are a little wider, et cetera.
+
+37:01.560 --> 37:03.280
+ And I think that that intuition is pretty good,
+
+37:03.280 --> 37:06.000
+ except we don't really care about most of the time.
+
+37:06.000 --> 37:08.360
+ We care about all of the time.
+
+37:08.360 --> 37:10.840
+ And when you're driving on a freeway with a truck,
+
+37:10.840 --> 37:15.840
+ say 70 miles an hour and you've got 70,000 pound load
+
+37:15.840 --> 37:17.840
+ to do with you, that's just an incredible amount
+
+37:17.840 --> 37:18.840
+ of kinetic energy.
+
+37:18.840 --> 37:21.440
+ And so when that goes wrong, it goes really wrong.
+
+37:22.600 --> 37:27.600
+ And that those challenges that you see occur more rarely
+
+37:27.760 --> 37:31.040
+ so you don't get to learn as quickly.
+
+37:31.040 --> 37:33.640
+ And they're incrementally more difficult
+
+37:33.640 --> 37:35.920
+ than urban driving, but they're not easier
+
+37:35.920 --> 37:37.400
+ than urban driving.
+
+37:37.400 --> 37:41.600
+ And so I think this happens in moderate speed,
+
+37:41.600 --> 37:43.840
+ urban environments, because there,
+
+37:43.840 --> 37:46.560
+ if two vehicles crash at 25 miles per hour,
+
+37:46.560 --> 37:50.040
+ it's not good, but probably everybody walks away.
+
+37:51.000 --> 37:53.680
+ And those events where there's the possibility
+
+37:53.680 --> 37:55.720
+ for that occurring happen frequently.
+
+37:55.720 --> 37:57.920
+ So we get to learn more rapidly.
+
+37:57.920 --> 38:01.320
+ We get to do that with lower risk for everyone.
+
+38:02.440 --> 38:04.280
+ And then we can deliver value to people
+
+38:04.280 --> 38:05.800
+ that need to get from one place to another.
+
+38:05.800 --> 38:08.160
+ And then once we've got that solved,
+
+38:08.160 --> 38:10.000
+ then the kind of the freeway driving part of this
+
+38:10.000 --> 38:12.440
+ just falls out, but we're able to learn
+
+38:12.440 --> 38:15.160
+ more safely, more quickly in the urban environment.
+
+38:15.160 --> 38:18.480
+ So 10 years and then scale 20, 30 years.
+
+38:18.480 --> 38:21.440
+ I mean, who knows if it's sufficiently compelling
+
+38:21.440 --> 38:24.320
+ experience is created, it can be faster and slower.
+
+38:24.320 --> 38:27.120
+ Do you think there could be breakthroughs
+
+38:27.120 --> 38:29.880
+ and what kind of breakthroughs might there be
+
+38:29.880 --> 38:32.360
+ that completely change that timeline?
+
+38:32.360 --> 38:35.320
+ Again, not only am I asking to predict the future,
+
+38:35.320 --> 38:37.280
+ I'm asking you to predict breakthroughs
+
+38:37.280 --> 38:38.280
+ that haven't happened yet.
+
+38:38.280 --> 38:41.800
+ So what's the, I think another way to ask that would be
+
+38:41.800 --> 38:44.240
+ if I could wave a magic wand,
+
+38:44.240 --> 38:46.640
+ what part of the system would I make work today
+
+38:46.640 --> 38:48.600
+ to accelerate it as quickly as possible?
+
+38:48.600 --> 38:49.440
+ Right.
+
+38:52.080 --> 38:54.080
+ Don't say infrastructure, please don't say infrastructure.
+
+38:54.080 --> 38:56.280
+ No, it's definitely not infrastructure.
+
+38:56.280 --> 39:00.520
+ It's really that perception forecasting capability.
+
+39:00.520 --> 39:04.760
+ So if tomorrow you could give me a perfect model
+
+39:04.760 --> 39:07.520
+ of what's happening and what will happen
+
+39:07.520 --> 39:11.400
+ for the next five seconds around a vehicle
+
+39:11.400 --> 39:14.480
+ on the roadway, that would accelerate things
+
+39:14.480 --> 39:15.320
+ pretty dramatically.
+
+39:15.320 --> 39:17.560
+ Are you interested in staying up at night?
+
+39:17.560 --> 39:21.680
+ Are you mostly bothered by cars, pedestrians, or cyclists?
+
+39:21.680 --> 39:25.920
+ So I worry most about the vulnerable road users
+
+39:25.920 --> 39:28.000
+ about the combination of cyclists and cars, right?
+
+39:28.000 --> 39:29.480
+ Just cyclists and pedestrians
+
+39:29.480 --> 39:31.880
+ because they're not in armor.
+
+39:33.240 --> 39:36.480
+ The cars, they're bigger, they've got protection
+
+39:36.480 --> 39:39.440
+ for the people and so the ultimate risk is lower there.
+
+39:39.440 --> 39:44.080
+ Whereas a pedestrian or cyclist, they're out on the road
+
+39:44.080 --> 39:46.520
+ and they don't have any protection.
+
+39:46.520 --> 39:49.760
+ And so we need to pay extra attention to that.
+
+39:49.760 --> 39:54.120
+ Do you think about a very difficult technical challenge
+
+39:55.760 --> 39:58.560
+ of the fact that pedestrians,
+
+39:58.560 --> 40:01.400
+ if you try to protect pedestrians by being careful
+
+40:01.400 --> 40:04.600
+ and slow, they'll take advantage of that.
+
+40:04.600 --> 40:07.560
+ So the game theoretic dance.
+
+40:07.560 --> 40:10.880
+ Does that worry you from a technical perspective
+
+40:10.880 --> 40:12.520
+ how we solve that?
+
+40:12.520 --> 40:14.600
+ Because as humans, the way we solve that
+
+40:14.600 --> 40:17.280
+ is kind of nudge our way through the pedestrians,
+
+40:17.280 --> 40:20.040
+ which doesn't feel from a technical perspective
+
+40:20.040 --> 40:22.320
+ as a appropriate algorithm.
+
+40:23.240 --> 40:25.960
+ But do you think about how we solve that problem?
+
+40:25.960 --> 40:30.960
+ Yeah, I think there's two different concepts there.
+
+40:31.400 --> 40:35.880
+ So one is, am I worried that because these vehicles
+
+40:35.880 --> 40:37.640
+ are self driving, people will kind of step on the road
+
+40:37.640 --> 40:38.680
+ and take advantage of them.
+
+40:38.680 --> 40:43.680
+ And I've heard this and I don't really believe it
+
+40:43.800 --> 40:46.000
+ because if I'm driving down the road
+
+40:46.000 --> 40:48.920
+ and somebody steps in front of me, I'm going to stop.
+
+40:48.920 --> 40:49.760
+ Right?
+
+40:49.760 --> 40:53.720
+ Like even if I'm annoyed, I'm not gonna just drive
+
+40:53.720 --> 40:55.200
+ through a person stood on the road.
+
+40:55.200 --> 40:56.440
+ Right.
+
+40:56.440 --> 41:00.440
+ And so I think today people can take advantage of this
+
+41:00.440 --> 41:02.600
+ and you do see some people do it.
+
+41:02.600 --> 41:04.200
+ I guess there's an incremental risk
+
+41:04.200 --> 41:05.920
+ because maybe they have lower confidence
+
+41:05.920 --> 41:06.760
+ that I'm going to see them
+
+41:06.760 --> 41:09.360
+ than they might have for an automated vehicle.
+
+41:09.360 --> 41:12.080
+ And so maybe that shifts it a little bit.
+
+41:12.080 --> 41:14.400
+ But I think people don't want to get hit by cars.
+
+41:14.400 --> 41:17.120
+ And so I think that I'm not that worried
+
+41:17.120 --> 41:18.800
+ about people walking out of the one on one
+
+41:18.800 --> 41:21.840
+ and creating chaos more than they would today.
+
+41:24.400 --> 41:27.040
+ Regarding kind of the nudging through a big stream
+
+41:27.040 --> 41:30.040
+ of pedestrians leaving a concert or something.
+
+41:30.040 --> 41:33.480
+ I think that is further down the technology pipeline.
+
+41:33.480 --> 41:36.920
+ I think that you're right, that's tricky.
+
+41:36.920 --> 41:38.560
+ I don't think it's necessarily,
+
+41:40.320 --> 41:43.360
+ I think the algorithm people use for this is pretty simple.
+
+41:43.360 --> 41:44.200
+ Right?
+
+41:44.200 --> 41:45.040
+ It's kind of just move forward slowly
+
+41:45.040 --> 41:47.600
+ and if somebody's really close and stop.
+
+41:47.600 --> 41:50.840
+ And I think that that probably can be replicated
+
+41:50.840 --> 41:54.040
+ pretty easily and particularly given that it's,
+
+41:54.040 --> 41:57.240
+ you don't do this at 30 miles an hour, you do it at one,
+
+41:57.240 --> 41:59.080
+ that even in those situations,
+
+41:59.080 --> 42:01.200
+ the risk is relatively minimal.
+
+42:01.200 --> 42:03.440
+ But it's not something we're thinking
+
+42:03.440 --> 42:04.560
+ about in any serious way.
+
+42:04.560 --> 42:08.000
+ And probably that's less an algorithm problem
+
+42:08.000 --> 42:10.160
+ more creating a human experience.
+
+42:10.160 --> 42:14.320
+ So the HCI people that create a visual display
+
+42:14.320 --> 42:16.280
+ that you're pleasantly as a pedestrian,
+
+42:16.280 --> 42:17.680
+ nudged out of the way.
+
+42:17.680 --> 42:21.960
+ That's an experience problem, not an algorithm problem.
+
+42:22.880 --> 42:25.480
+ Who's the main competitor to Aurora today?
+
+42:25.480 --> 42:28.600
+ And how do you out compete them in the long run?
+
+42:28.600 --> 42:31.200
+ So we really focus a lot on what we're doing here.
+
+42:31.200 --> 42:34.440
+ I think that, I've said this a few times
+
+42:34.440 --> 42:37.960
+ that this is a huge difficult problem
+
+42:37.960 --> 42:40.280
+ and it's great that a bunch of companies are tackling it
+
+42:40.280 --> 42:42.320
+ because I think it's so important for society
+
+42:42.320 --> 42:43.760
+ that somebody gets there.
+
+42:45.200 --> 42:49.040
+ So we don't spend a whole lot of time
+
+42:49.040 --> 42:51.560
+ like thinking tactically about who's out there
+
+42:51.560 --> 42:55.480
+ and how do we beat that person individually?
+
+42:55.480 --> 42:58.680
+ What are we trying to do to go faster ultimately?
+
+42:58.680 --> 43:02.600
+ Well, part of it is the leisure team we have
+
+43:02.600 --> 43:04.160
+ has got pretty tremendous experience.
+
+43:04.160 --> 43:06.400
+ And so we kind of understand the landscape
+
+43:06.400 --> 43:09.120
+ and understand where the cul de sacs are to some degree.
+
+43:09.120 --> 43:10.920
+ And we try and avoid those.
+
+43:12.600 --> 43:14.240
+ I think there's a part of it
+
+43:14.240 --> 43:16.240
+ just this great team we've built.
+
+43:16.240 --> 43:19.040
+ People, this is a technology and a company
+
+43:19.040 --> 43:22.280
+ that people believe in the mission of.
+
+43:22.280 --> 43:24.760
+ And so it allows us to attract just awesome people
+
+43:24.760 --> 43:25.680
+ to go work.
+
+43:26.760 --> 43:28.000
+ We've got a culture, I think,
+
+43:28.000 --> 43:30.440
+ that people appreciate, that allows them to focus,
+
+43:30.440 --> 43:33.080
+ allows them to really spend time solving problems.
+
+43:33.080 --> 43:35.880
+ And I think that keeps them energized.
+
+43:35.880 --> 43:40.880
+ And then we've invested heavily in the infrastructure
+
+43:43.520 --> 43:46.520
+ and architectures that we think will ultimately accelerate us.
+
+43:46.520 --> 43:50.640
+ So because of the folks we're able to bring in early on,
+
+43:50.640 --> 43:53.520
+ because of the great investors we have,
+
+43:53.520 --> 43:56.760
+ we don't spend all of our time doing demos
+
+43:56.760 --> 43:58.680
+ and kind of leaping from one demo to the next.
+
+43:58.680 --> 44:02.800
+ We've been given the freedom to invest in
+
+44:03.960 --> 44:05.480
+ infrastructure to do machine learning,
+
+44:05.480 --> 44:08.600
+ infrastructure to pull data from our on road testing,
+
+44:08.600 --> 44:11.480
+ infrastructure to use that to accelerate engineering.
+
+44:11.480 --> 44:14.480
+ And I think that early investment
+
+44:14.480 --> 44:17.320
+ and continuing investment in those kind of tools
+
+44:17.320 --> 44:19.400
+ will ultimately allow us to accelerate
+
+44:19.400 --> 44:21.920
+ and do something pretty incredible.
+
+44:21.920 --> 44:23.400
+ Chris, beautifully put.
+
+44:23.400 --> 44:24.640
+ It's a good place to end.
+
+44:24.640 --> 44:26.520
+ Thank you so much for talking today.
+
+44:26.520 --> 44:27.360
+ Thank you very much.
+
+44:27.360 --> 44:57.200
+ I hope you enjoyed it.
+

vtt/episode_030_small.vtt

@@ -0,0 +1,3584 @@
+WEBVTT
+
+00:00.000 --> 00:03.440
+ The following is a conversation with Kevin Scott,
+
+00:03.440 --> 00:06.080
+ the CTO of Microsoft.
+
+00:06.080 --> 00:08.560
+ Before that, he was the senior vice president
+
+00:08.560 --> 00:11.080
+ of engineering and operations at LinkedIn,
+
+00:11.080 --> 00:13.520
+ and before that, he oversaw mobile ads
+
+00:13.520 --> 00:14.960
+ engineering at Google.
+
+00:15.960 --> 00:19.000
+ He also has a podcast called Behind the Tech
+
+00:19.000 --> 00:21.880
+ with Kevin Scott, which I'm a fan of.
+
+00:21.880 --> 00:24.280
+ This was a fun and wide ranging conversation
+
+00:24.280 --> 00:26.680
+ that covered many aspects of computing.
+
+00:26.680 --> 00:28.840
+ It happened over a month ago,
+
+00:28.840 --> 00:31.000
+ before the announcement of Microsoft's investment
+
+00:31.000 --> 00:34.440
+ OpenAI that a few people have asked me about.
+
+00:34.440 --> 00:38.120
+ I'm sure there'll be one or two people in the future
+
+00:38.120 --> 00:41.400
+ that'll talk with me about the impact of that investment.
+
+00:42.280 --> 00:45.440
+ This is the Artificial Intelligence podcast.
+
+00:45.440 --> 00:47.680
+ If you enjoy it, subscribe on YouTube,
+
+00:47.680 --> 00:49.440
+ give it five stars on iTunes,
+
+00:49.440 --> 00:50.960
+ support it on a Patreon,
+
+00:50.960 --> 00:53.000
+ or simply connect with me on Twitter,
+
+00:53.000 --> 00:57.680
+ at Lex Freedman, spelled FRIDMAM.
+
+00:57.680 --> 00:59.240
+ And I'd like to give a special thank you
+
+00:59.240 --> 01:01.960
+ to Tom and Elanti Bighausen
+
+01:01.960 --> 01:04.600
+ for their support of the podcast on Patreon.
+
+01:04.600 --> 01:06.080
+ Thanks Tom and Elanti.
+
+01:06.080 --> 01:08.400
+ Hope I didn't mess up your last name too bad.
+
+01:08.400 --> 01:10.520
+ Your support means a lot,
+
+01:10.520 --> 01:13.480
+ and inspires me to keep this series going.
+
+01:13.480 --> 01:18.160
+ And now, here's my conversation with Kevin Scott.
+
+01:18.160 --> 01:20.760
+ You've described yourself as a kid in a candy store
+
+01:20.760 --> 01:23.000
+ at Microsoft because of all the interesting projects
+
+01:23.000 --> 01:24.200
+ that are going on.
+
+01:24.200 --> 01:28.000
+ Can you try to do the impossible task
+
+01:28.000 --> 01:31.760
+ and give a brief whirlwind view
+
+01:31.760 --> 01:34.520
+ of all the spaces that Microsoft is working in?
+
+01:35.520 --> 01:37.440
+ Both research and product.
+
+01:37.440 --> 01:42.440
+ If you include research, it becomes even more difficult.
+
+01:46.480 --> 01:48.880
+ So, I think broadly speaking,
+
+01:48.880 --> 01:53.720
+ Microsoft's product portfolio includes everything
+
+01:53.720 --> 01:56.920
+ from big cloud business,
+
+01:56.920 --> 01:59.360
+ like a big set of SaaS services.
+
+01:59.360 --> 02:01.720
+ We have sort of the original,
+
+02:01.720 --> 02:05.560
+ or like some of what are among the original
+
+02:05.560 --> 02:09.640
+ productivity software products that everybody uses.
+
+02:09.640 --> 02:11.200
+ We have an operating system business.
+
+02:11.200 --> 02:13.560
+ We have a hardware business
+
+02:13.560 --> 02:17.240
+ where we make everything from computer mice
+
+02:17.240 --> 02:20.760
+ and headphones to high end,
+
+02:20.760 --> 02:23.520
+ high end personal computers and laptops.
+
+02:23.520 --> 02:27.680
+ We have a fairly broad ranging research group
+
+02:27.680 --> 02:29.680
+ where we have people doing everything
+
+02:29.680 --> 02:31.880
+ from economics research.
+
+02:31.880 --> 02:35.920
+ So, there's this really smart young economist,
+
+02:35.920 --> 02:39.760
+ Glenn Weil, who like my group works with a lot,
+
+02:39.760 --> 02:42.880
+ who's doing this research on these things
+
+02:42.880 --> 02:45.120
+ called radical markets.
+
+02:45.120 --> 02:48.120
+ Like he's written an entire technical book
+
+02:48.120 --> 02:51.120
+ about this whole notion of radical markets.
+
+02:51.120 --> 02:53.520
+ So, like the research group sort of spans from that
+
+02:53.520 --> 02:56.840
+ to human computer interaction, to artificial intelligence.
+
+02:56.840 --> 03:01.040
+ And we have GitHub, we have LinkedIn.
+
+03:01.040 --> 03:05.800
+ We have a search advertising and news business
+
+03:05.800 --> 03:07.360
+ and like probably a bunch of stuff
+
+03:07.360 --> 03:11.240
+ that I'm embarrassingly not recounting in this list.
+
+03:11.240 --> 03:12.920
+ On gaming to Xbox and so on, right?
+
+03:12.920 --> 03:14.120
+ Yeah, gaming for sure.
+
+03:14.120 --> 03:17.320
+ Like I was having a super fun conversation
+
+03:17.320 --> 03:19.520
+ this morning with Phil Spencer.
+
+03:19.520 --> 03:21.280
+ So, when I was in college,
+
+03:21.280 --> 03:25.560
+ there was this game that Lucas Arts made
+
+03:25.560 --> 03:27.600
+ called Day of the Tentacle,
+
+03:27.600 --> 03:30.160
+ that my friends and I played forever.
+
+03:30.160 --> 03:33.920
+ And like we're doing some interesting collaboration now
+
+03:33.920 --> 03:37.920
+ with the folks who made Day of the Tentacle.
+
+03:37.920 --> 03:40.840
+ And I was like completely nerding out with Tim Schaeffer,
+
+03:40.840 --> 03:43.880
+ like the guy who wrote Day of the Tentacle this morning,
+
+03:43.880 --> 03:45.840
+ just a complete fanboy,
+
+03:45.840 --> 03:49.880
+ which you know, sort of it like happens a lot.
+
+03:49.880 --> 03:53.320
+ Like, you know, Microsoft has been doing so much stuff
+
+03:53.320 --> 03:56.000
+ at such breadth for such a long period of time
+
+03:56.000 --> 03:59.680
+ that, you know, like being CTO,
+
+03:59.680 --> 04:02.200
+ like most of the time my job is very, very serious
+
+04:02.200 --> 04:05.640
+ and sometimes that like I get caught up
+
+04:05.640 --> 04:09.200
+ in like how amazing it is
+
+04:09.200 --> 04:11.520
+ to be able to have the conversations
+
+04:11.520 --> 04:14.640
+ that I have with the people I get to have them with.
+
+04:14.640 --> 04:17.040
+ You had to reach back into the sentimental
+
+04:17.040 --> 04:21.640
+ and what's the radical markets and the economics?
+
+04:21.640 --> 04:24.760
+ So the idea with radical markets is like,
+
+04:24.760 --> 04:29.760
+ can you come up with new market based mechanisms to,
+
+04:32.320 --> 04:33.800
+ you know, I think we have this,
+
+04:33.800 --> 04:35.240
+ we're having this debate right now,
+
+04:35.240 --> 04:40.040
+ like does capitalism work, like free markets work?
+
+04:40.040 --> 04:43.000
+ Can the incentive structures
+
+04:43.000 --> 04:46.360
+ that are built into these systems produce outcomes
+
+04:46.360 --> 04:51.360
+ that are creating sort of equitably distributed benefits
+
+04:51.560 --> 04:53.520
+ for every member of society?
+
+04:55.400 --> 04:58.720
+ You know, and I think it's a reasonable set of questions
+
+04:58.720 --> 04:59.560
+ to be asking.
+
+04:59.560 --> 05:02.160
+ And so what Glenn, and so like, you know,
+
+05:02.160 --> 05:04.400
+ one mode of thought there, like if you have doubts
+
+05:04.400 --> 05:06.720
+ that the markets are actually working,
+
+05:06.720 --> 05:08.560
+ you can sort of like tip towards like,
+
+05:08.560 --> 05:10.800
+ okay, let's become more socialist
+
+05:10.800 --> 05:14.240
+ and like have central planning and governments
+
+05:14.240 --> 05:15.800
+ or some other central organization
+
+05:15.800 --> 05:18.280
+ is like making a bunch of decisions
+
+05:18.280 --> 05:22.040
+ about how sort of work gets done
+
+05:22.040 --> 05:25.400
+ and like where the investments
+
+05:25.400 --> 05:28.880
+ and where the outputs of those investments get distributed.
+
+05:28.880 --> 05:32.160
+ Glenn's notion is like lean more
+
+05:32.160 --> 05:35.800
+ into like the market based mechanism.
+
+05:35.800 --> 05:37.920
+ So like for instance,
+
+05:37.920 --> 05:39.600
+ this is one of the more radical ideas,
+
+05:39.600 --> 05:44.600
+ like suppose that you had a radical pricing mechanism
+
+05:45.160 --> 05:47.120
+ for assets like real estate
+
+05:47.120 --> 05:52.120
+ where you could be bid out of your position
+
+05:53.600 --> 05:58.600
+ in your home, you know, for instance.
+
+05:58.720 --> 06:01.120
+ So like if somebody came along and said,
+
+06:01.120 --> 06:04.400
+ you know, like I can find higher economic utility
+
+06:04.400 --> 06:05.760
+ for this piece of real estate
+
+06:05.760 --> 06:08.720
+ that you're running your business in,
+
+06:08.720 --> 06:13.040
+ like then like you either have to, you know,
+
+06:13.040 --> 06:16.440
+ sort of bid to sort of stay
+
+06:16.440 --> 06:19.960
+ or like the thing that's got the higher economic utility,
+
+06:19.960 --> 06:21.440
+ you know, sort of takes over the asset
+
+06:21.440 --> 06:23.720
+ and which would make it very difficult
+
+06:23.720 --> 06:27.600
+ to have the same sort of rent seeking behaviors
+
+06:27.600 --> 06:29.000
+ that you've got right now
+
+06:29.000 --> 06:34.000
+ because like if you did speculative bidding,
+
+06:34.000 --> 06:39.000
+ like you would very quickly like lose a whole lot of money.
+
+06:40.440 --> 06:43.520
+ And so like the prices of the assets would be sort of
+
+06:43.520 --> 06:47.600
+ like very closely indexed to like the value
+
+06:47.600 --> 06:49.720
+ that they can produce.
+
+06:49.720 --> 06:52.680
+ And like because like you'd have this sort of real time
+
+06:52.680 --> 06:55.320
+ mechanism that would force you to sort of mark the value
+
+06:55.320 --> 06:56.800
+ of the asset to the market,
+
+06:56.800 --> 06:58.560
+ then it could be taxed appropriately.
+
+06:58.560 --> 07:00.400
+ Like you couldn't sort of sit on this thing and say,
+
+07:00.400 --> 07:03.040
+ oh, like this house is only worth 10,000 bucks
+
+07:03.040 --> 07:06.600
+ when like everything around it is worth 10 million.
+
+07:06.600 --> 07:07.440
+ That's really interesting.
+
+07:07.440 --> 07:08.720
+ So it's an incentive structure
+
+07:08.720 --> 07:13.200
+ that where the prices match the value much better.
+
+07:13.200 --> 07:14.040
+ Yeah.
+
+07:14.040 --> 07:16.320
+ And Glenn does a much, much better job than I do
+
+07:16.320 --> 07:18.920
+ at selling and I probably picked the world's worst example,
+
+07:18.920 --> 07:20.360
+ you know, and, and, and, but like,
+
+07:20.360 --> 07:24.520
+ and it's intentionally provocative, you know,
+
+07:24.520 --> 07:26.480
+ so like this whole notion, like I, you know,
+
+07:26.480 --> 07:28.920
+ like I'm not sure whether I like this notion
+
+07:28.920 --> 07:31.120
+ that like we can have a set of market mechanisms
+
+07:31.120 --> 07:35.360
+ where I could get bid out of, out of my property, you know,
+
+07:35.360 --> 07:37.680
+ but, but, you know, like if you're thinking about something
+
+07:37.680 --> 07:42.480
+ like Elizabeth Warren's wealth tax, for instance,
+
+07:42.480 --> 07:45.600
+ like you would have, I mean, it'd be really interesting
+
+07:45.600 --> 07:50.080
+ in like how you would actually set the price on the assets.
+
+07:50.080 --> 07:52.040
+ And like you might have to have a mechanism like that
+
+07:52.040 --> 07:54.160
+ if you put a tax like that in place.
+
+07:54.160 --> 07:56.440
+ It's really interesting that that kind of research,
+
+07:56.440 --> 07:59.800
+ at least tangentially touching Microsoft research.
+
+07:59.800 --> 08:00.640
+ Yeah.
+
+08:00.640 --> 08:02.560
+ So if you're really thinking broadly,
+
+08:02.560 --> 08:07.560
+ maybe you can speak to this connects to AI.
+
+08:08.400 --> 08:10.680
+ So we have a candidate, Andrew Yang,
+
+08:10.680 --> 08:13.480
+ who kind of talks about artificial intelligence
+
+08:13.480 --> 08:16.640
+ and the concern that people have about, you know,
+
+08:16.640 --> 08:19.000
+ automations impact on society.
+
+08:19.000 --> 08:22.680
+ And arguably Microsoft is at the cutting edge
+
+08:22.680 --> 08:25.040
+ of innovation in all these kinds of ways.
+
+08:25.040 --> 08:27.080
+ And so it's pushing AI forward.
+
+08:27.080 --> 08:30.040
+ How do you think about combining all our conversations
+
+08:30.040 --> 08:32.840
+ together here with radical markets and socialism
+
+08:32.840 --> 08:37.520
+ and innovation in AI that Microsoft is doing?
+
+08:37.520 --> 08:42.520
+ And then Andrew Yang's worry that that will,
+
+08:43.520 --> 08:46.840
+ that will result in job loss for the lower and so on.
+
+08:46.840 --> 08:47.680
+ How do you think about that?
+
+08:47.680 --> 08:51.160
+ I think it's sort of one of the most important questions
+
+08:51.160 --> 08:55.320
+ in technology, like maybe even in society right now
+
+08:55.320 --> 09:00.320
+ about how is AI going to develop over the course
+
+09:00.720 --> 09:02.000
+ of the next several decades
+
+09:02.000 --> 09:03.600
+ and like what's it gonna be used for
+
+09:03.600 --> 09:06.560
+ and like what benefits will it produce
+
+09:06.560 --> 09:08.520
+ and what negative impacts will it produce
+
+09:08.520 --> 09:13.520
+ and you know, who gets to steer this whole thing?
+
+09:13.720 --> 09:16.320
+ You know, I'll say at the highest level,
+
+09:17.240 --> 09:22.240
+ one of the real joys of getting to do what I do at Microsoft
+
+09:22.240 --> 09:27.240
+ is Microsoft has this heritage as a platform company.
+
+09:27.560 --> 09:31.040
+ And so, you know, like Bill has this thing
+
+09:31.040 --> 09:32.880
+ that he said a bunch of years ago
+
+09:32.880 --> 09:36.440
+ where the measure of a successful platform
+
+09:36.440 --> 09:39.800
+ is that it produces far more economic value
+
+09:39.800 --> 09:41.840
+ for the people who build on top of the platform
+
+09:41.840 --> 09:46.840
+ than is created for the platform owner or builder.
+
+09:47.320 --> 09:50.920
+ And I think we have to think about AI that way.
+
+09:50.920 --> 09:55.920
+ Like it has to be a platform that other people can use
+
+09:56.280 --> 10:01.280
+ to build businesses, to fulfill their creative objectives,
+
+10:01.280 --> 10:04.640
+ to be entrepreneurs, to solve problems that they have
+
+10:04.640 --> 10:07.680
+ in their work and in their lives.
+
+10:07.680 --> 10:11.960
+ It can't be a thing where there are a handful of companies
+
+10:11.960 --> 10:16.440
+ sitting in a very small handful of cities geographically
+
+10:16.440 --> 10:19.120
+ who are making all the decisions
+
+10:19.120 --> 10:24.120
+ about what goes into the AI and like,
+
+10:24.240 --> 10:26.920
+ and then on top of like all this infrastructure,
+
+10:26.920 --> 10:31.000
+ then build all of the commercially valuable uses for it.
+
+10:31.000 --> 10:34.400
+ So like, I think like that's bad from a, you know,
+
+10:34.400 --> 10:36.520
+ sort of, you know, economics
+
+10:36.520 --> 10:39.720
+ and sort of equitable distribution of value perspective,
+
+10:39.720 --> 10:42.080
+ like, you know, sort of back to this whole notion of,
+
+10:42.080 --> 10:44.560
+ you know, like, do the markets work?
+
+10:44.560 --> 10:47.600
+ But I think it's also bad from an innovation perspective
+
+10:47.600 --> 10:51.360
+ because like I have infinite amounts of faith
+
+10:51.360 --> 10:53.880
+ in human beings that if you, you know,
+
+10:53.880 --> 10:58.280
+ give folks powerful tools, they will go do interesting things.
+
+10:58.280 --> 11:02.320
+ And it's more than just a few tens of thousands of people
+
+11:02.320 --> 11:03.360
+ with the interesting tools,
+
+11:03.360 --> 11:05.400
+ it should be millions of people with the tools.
+
+11:05.400 --> 11:07.200
+ So it's sort of like, you know,
+
+11:07.200 --> 11:10.200
+ you think about the steam engine
+
+11:10.200 --> 11:13.800
+ and the late 18th century, like it was, you know,
+
+11:13.800 --> 11:16.800
+ maybe the first large scale substitute for human labor
+
+11:16.800 --> 11:19.120
+ that we've built like a machine.
+
+11:19.120 --> 11:21.680
+ And, you know, in the beginning,
+
+11:21.680 --> 11:23.520
+ when these things are getting deployed,
+
+11:23.520 --> 11:28.320
+ the folks who got most of the value from the steam engines
+
+11:28.320 --> 11:30.160
+ were the folks who had capital
+
+11:30.160 --> 11:31.600
+ so they could afford to build them.
+
+11:31.600 --> 11:34.720
+ And like they built factories around them in businesses
+
+11:34.720 --> 11:38.680
+ and the experts who knew how to build and maintain them.
+
+11:38.680 --> 11:42.880
+ But access to that technology democratized over time.
+
+11:42.880 --> 11:47.040
+ Like now like an engine is not a,
+
+11:47.040 --> 11:48.800
+ it's not like a differentiated thing.
+
+11:48.800 --> 11:50.280
+ Like there isn't one engine company
+
+11:50.280 --> 11:51.560
+ that builds all the engines
+
+11:51.560 --> 11:53.120
+ and all of the things that use engines
+
+11:53.120 --> 11:54.240
+ are made by this company.
+
+11:54.240 --> 11:57.440
+ And like they get all the economics from all of that.
+
+11:57.440 --> 11:59.320
+ Like, no, like fully demarcated.
+
+11:59.320 --> 12:00.600
+ Like they're probably, you know,
+
+12:00.600 --> 12:02.360
+ we're sitting here in this room
+
+12:02.360 --> 12:03.680
+ and like even though they don't,
+
+12:03.680 --> 12:05.280
+ they're probably things, you know,
+
+12:05.280 --> 12:09.120
+ like the MIMS gyroscope that are in both of our,
+
+12:09.120 --> 12:11.480
+ like there's like little engines, you know,
+
+12:11.480 --> 12:14.520
+ sort of everywhere, they're just a component
+
+12:14.520 --> 12:16.240
+ in how we build the modern world.
+
+12:16.240 --> 12:17.680
+ Like AI needs to get there.
+
+12:17.680 --> 12:20.200
+ Yeah, so that's a really powerful way to think.
+
+12:20.200 --> 12:25.120
+ If we think of AI as a platform versus a tool
+
+12:25.120 --> 12:27.600
+ that Microsoft owns as a platform
+
+12:27.600 --> 12:30.120
+ that enables creation on top of it,
+
+12:30.120 --> 12:31.520
+ that's the way to democratize it.
+
+12:31.520 --> 12:34.200
+ That's really interesting actually.
+
+12:34.200 --> 12:36.040
+ And Microsoft throughout its history
+
+12:36.040 --> 12:38.240
+ has been positioned well to do that.
+
+12:38.240 --> 12:41.640
+ And the, you know, the tieback to this radical markets thing,
+
+12:41.640 --> 12:46.640
+ like the, so my team has been working with Glenn
+
+12:47.800 --> 12:51.120
+ on this and Jaren Lanier actually.
+
+12:51.120 --> 12:56.120
+ So Jaren is the like the sort of father of virtual reality.
+
+12:56.440 --> 12:59.480
+ Like he's one of the most interesting human beings
+
+12:59.480 --> 13:01.760
+ on the planet, like a sweet, sweet guy.
+
+13:02.840 --> 13:07.120
+ And so Jaren and Glenn and folks in my team
+
+13:07.120 --> 13:10.360
+ have been working on this notion of data as labor
+
+13:10.360 --> 13:13.160
+ or like they call it data dignity as well.
+
+13:13.160 --> 13:16.880
+ And so the idea is that if you, you know,
+
+13:16.880 --> 13:18.600
+ again, going back to this, you know,
+
+13:18.600 --> 13:20.800
+ sort of industrial analogy,
+
+13:20.800 --> 13:23.560
+ if you think about data as the raw material
+
+13:23.560 --> 13:27.640
+ that is consumed by the machine of AI
+
+13:27.640 --> 13:30.560
+ in order to do useful things,
+
+13:30.560 --> 13:34.400
+ then like we're not doing a really great job right now
+
+13:34.400 --> 13:37.760
+ in having transparent marketplaces for valuing
+
+13:37.760 --> 13:39.800
+ those data contributions.
+
+13:39.800 --> 13:42.680
+ So like, and we all make them like explicitly,
+
+13:42.680 --> 13:43.600
+ like you go to LinkedIn,
+
+13:43.600 --> 13:46.160
+ you sort of set up your profile on LinkedIn,
+
+13:46.160 --> 13:47.800
+ like that's an explicit contribution.
+
+13:47.800 --> 13:49.480
+ Like, you know exactly the information
+
+13:49.480 --> 13:50.720
+ that you're putting into the system.
+
+13:50.720 --> 13:53.000
+ And like you put it there because you have
+
+13:53.000 --> 13:55.520
+ some nominal notion of like what value
+
+13:55.520 --> 13:56.640
+ you're going to get in return,
+
+13:56.640 --> 13:57.720
+ but it's like only nominal.
+
+13:57.720 --> 13:59.680
+ Like you don't know exactly what value
+
+13:59.680 --> 14:02.040
+ you're getting in return, like services free, you know,
+
+14:02.040 --> 14:04.600
+ like it's low amount of like perceived.
+
+14:04.600 --> 14:06.680
+ And then you've got all this indirect contribution
+
+14:06.680 --> 14:08.960
+ that you're making just by virtue of interacting
+
+14:08.960 --> 14:13.160
+ with all of the technology that's in your daily life.
+
+14:13.160 --> 14:16.120
+ And so like what Glenn and Jaren
+
+14:16.120 --> 14:19.440
+ and this data dignity team are trying to do is like,
+
+14:19.440 --> 14:22.240
+ can we figure out a set of mechanisms
+
+14:22.240 --> 14:26.000
+ that let us value those data contributions
+
+14:26.000 --> 14:28.200
+ so that you could create an economy
+
+14:28.200 --> 14:31.480
+ and like a set of controls and incentives
+
+14:31.480 --> 14:36.480
+ that would allow people to like maybe even in the limit
+
+14:36.840 --> 14:38.880
+ like earn part of their living
+
+14:38.880 --> 14:41.000
+ through the data that they're creating.
+
+14:41.000 --> 14:42.680
+ And like you can sort of see it in explicit ways.
+
+14:42.680 --> 14:46.000
+ There are these companies like Scale AI
+
+14:46.000 --> 14:49.960
+ and like they're a whole bunch of them in China right now
+
+14:49.960 --> 14:52.400
+ that are basically data labeling companies.
+
+14:52.400 --> 14:54.560
+ So like you're doing supervised machine learning,
+
+14:54.560 --> 14:57.400
+ you need lots and lots of label training data.
+
+14:58.600 --> 15:01.440
+ And like those people are getting like who work
+
+15:01.440 --> 15:03.600
+ for those companies are getting compensated
+
+15:03.600 --> 15:06.360
+ for their data contributions into the system.
+
+15:06.360 --> 15:07.720
+ And so...
+
+15:07.720 --> 15:10.280
+ That's easier to put a number on their contribution
+
+15:10.280 --> 15:11.960
+ because they're explicitly labeling data.
+
+15:11.960 --> 15:12.800
+ Correct.
+
+15:12.800 --> 15:14.360
+ But you're saying that we're all contributing data
+
+15:14.360 --> 15:15.720
+ in different kinds of ways.
+
+15:15.720 --> 15:19.640
+ And it's fascinating to start to explicitly try
+
+15:19.640 --> 15:20.880
+ to put a number on it.
+
+15:20.880 --> 15:22.600
+ Do you think that's possible?
+
+15:22.600 --> 15:23.640
+ I don't know, it's hard.
+
+15:23.640 --> 15:25.480
+ It really is.
+
+15:25.480 --> 15:30.480
+ Because, you know, we don't have as much transparency
+
+15:30.480 --> 15:35.480
+ as I think we need in like how the data is getting used.
+
+15:37.240 --> 15:38.720
+ And it's, you know, super complicated.
+
+15:38.720 --> 15:41.000
+ Like, you know, we, you know,
+
+15:41.000 --> 15:42.880
+ I think as technologists sort of appreciate
+
+15:42.880 --> 15:44.160
+ like some of the subtlety there.
+
+15:44.160 --> 15:47.880
+ It's like, you know, the data, the data gets created
+
+15:47.880 --> 15:51.400
+ and then it gets, you know, it's not valuable.
+
+15:51.400 --> 15:56.000
+ Like the data exhaust that you give off
+
+15:56.000 --> 15:58.480
+ or the, you know, the explicit data
+
+15:58.480 --> 16:03.240
+ that I am putting into the system isn't valuable.
+
+16:03.240 --> 16:05.160
+ It's super valuable atomically.
+
+16:05.160 --> 16:08.360
+ Like it's only valuable when you sort of aggregate it together
+
+16:08.360 --> 16:10.440
+ into, you know, sort of large numbers.
+
+16:10.440 --> 16:11.960
+ It's true even for these like folks
+
+16:11.960 --> 16:14.880
+ who are getting compensated for like labeling things.
+
+16:14.880 --> 16:16.480
+ Like for supervised machine learning now,
+
+16:16.480 --> 16:20.080
+ like you need lots of labels to train, you know,
+
+16:20.080 --> 16:22.080
+ a model that performs well.
+
+16:22.080 --> 16:24.440
+ And so, you know, I think that's one of the challenges.
+
+16:24.440 --> 16:26.120
+ It's like, how do you, you know,
+
+16:26.120 --> 16:28.000
+ how do you sort of figure out like
+
+16:28.000 --> 16:31.480
+ because this data is getting combined in so many ways,
+
+16:31.480 --> 16:33.880
+ like through these combinations,
+
+16:33.880 --> 16:35.880
+ like how the value is flowing.
+
+16:35.880 --> 16:38.520
+ Yeah, that's, that's fascinating.
+
+16:38.520 --> 16:39.360
+ Yeah.
+
+16:39.360 --> 16:41.880
+ And it's fascinating that you're thinking about this.
+
+16:41.880 --> 16:44.160
+ And I wasn't even going into this competition
+
+16:44.160 --> 16:48.200
+ expecting the breadth of research really
+
+16:48.200 --> 16:50.600
+ that Microsoft broadly is thinking about.
+
+16:50.600 --> 16:52.360
+ You are thinking about in Microsoft.
+
+16:52.360 --> 16:57.360
+ So if we go back to 89 when Microsoft released Office
+
+16:57.360 --> 17:00.920
+ or 1990 when they released Windows 3.0,
+
+17:00.920 --> 17:04.960
+ how's the, in your view,
+
+17:04.960 --> 17:07.280
+ I know you weren't there the entire, you know,
+
+17:07.280 --> 17:09.760
+ through its history, but how has the company changed
+
+17:09.760 --> 17:12.840
+ in the 30 years since as you look at it now?
+
+17:12.840 --> 17:17.080
+ The good thing is it's started off as a platform company.
+
+17:17.080 --> 17:19.960
+ Like it's still a platform company,
+
+17:19.960 --> 17:22.640
+ like the parts of the business that are like thriving
+
+17:22.640 --> 17:26.560
+ and most successful or those that are building platforms,
+
+17:26.560 --> 17:29.000
+ like the mission of the company now is,
+
+17:29.000 --> 17:30.120
+ the mission's changed.
+
+17:30.120 --> 17:32.480
+ It's like changing a very interesting way.
+
+17:32.480 --> 17:36.280
+ So, you know, back in 89.90,
+
+17:36.280 --> 17:39.040
+ like they were still on the original mission,
+
+17:39.040 --> 17:43.840
+ which was like put a PC on every desk and in every home.
+
+17:43.840 --> 17:47.480
+ Like, and it was basically about democratizing access
+
+17:47.480 --> 17:50.000
+ to this new personal computing technology,
+
+17:50.000 --> 17:52.680
+ which when Bill started the company,
+
+17:52.680 --> 17:57.680
+ integrated circuit microprocessors were a brand new thing
+
+17:57.680 --> 18:00.120
+ and like people were building, you know,
+
+18:00.120 --> 18:03.840
+ homebrew computers, you know, from kits,
+
+18:03.840 --> 18:07.520
+ like the way people build ham radios right now.
+
+18:08.520 --> 18:10.680
+ And I think this is sort of the interesting thing
+
+18:10.680 --> 18:12.840
+ for folks who build platforms in general.
+
+18:12.840 --> 18:16.840
+ Bill saw the opportunity there
+
+18:16.840 --> 18:18.720
+ and what personal computers could do.
+
+18:18.720 --> 18:20.440
+ And it was like, it was sort of a reach.
+
+18:20.440 --> 18:21.680
+ Like you just sort of imagined
+
+18:21.680 --> 18:23.880
+ like where things were, you know,
+
+18:23.880 --> 18:24.880
+ when they started the company
+
+18:24.880 --> 18:26.120
+ versus where things are now.
+
+18:26.120 --> 18:29.400
+ Like in success, when you democratize a platform,
+
+18:29.400 --> 18:31.000
+ it just sort of vanishes into the platform.
+
+18:31.000 --> 18:32.480
+ You don't pay attention to it anymore.
+
+18:32.480 --> 18:35.600
+ Like operating systems aren't a thing anymore.
+
+18:35.600 --> 18:38.040
+ Like they're super important, like completely critical.
+
+18:38.040 --> 18:41.760
+ And like, you know, when you see one, you know, fail,
+
+18:41.760 --> 18:43.520
+ like you just, you sort of understand,
+
+18:43.520 --> 18:45.320
+ but like, you know, it's not a thing where you're,
+
+18:45.320 --> 18:47.920
+ you're not like waiting for, you know,
+
+18:47.920 --> 18:50.480
+ the next operating system thing
+
+18:50.480 --> 18:52.960
+ in the same way that you were in 1995, right?
+
+18:52.960 --> 18:54.280
+ Like in 1995, like, you know,
+
+18:54.280 --> 18:56.000
+ we had Rolling Stones on the stage
+
+18:56.000 --> 18:57.600
+ with the Windows 95 roll out.
+
+18:57.600 --> 18:59.320
+ Like it was like the biggest thing in the world.
+
+18:59.320 --> 19:01.080
+ Everybody would like lined up for it
+
+19:01.080 --> 19:03.400
+ the way that people used to line up for iPhone.
+
+19:03.400 --> 19:05.120
+ But like, you know, eventually,
+
+19:05.120 --> 19:07.160
+ and like this isn't necessarily a bad thing.
+
+19:07.160 --> 19:09.000
+ Like it just sort of, you know,
+
+19:09.000 --> 19:12.880
+ the success is that it's sort of, it becomes ubiquitous.
+
+19:12.880 --> 19:14.800
+ It's like everywhere and like human beings
+
+19:14.800 --> 19:16.640
+ when their technology becomes ubiquitous,
+
+19:16.640 --> 19:18.240
+ they just sort of start taking it for granted.
+
+19:18.240 --> 19:23.240
+ So the mission now that Satya rearticulated
+
+19:23.640 --> 19:25.280
+ five plus years ago now
+
+19:25.280 --> 19:27.360
+ when he took over as CEO of the company,
+
+19:29.320 --> 19:33.480
+ our mission is to empower every individual
+
+19:33.480 --> 19:37.760
+ and every organization in the world to be more successful.
+
+19:39.200 --> 19:43.160
+ And so, you know, again, like that's a platform mission.
+
+19:43.160 --> 19:46.320
+ And like the way that we do it now is different.
+
+19:46.320 --> 19:48.680
+ It's like we have a hyperscale cloud
+
+19:48.680 --> 19:51.680
+ that people are building their applications on top of.
+
+19:51.680 --> 19:53.680
+ Like we have a bunch of AI infrastructure
+
+19:53.680 --> 19:56.280
+ that people are building their AI applications on top of.
+
+19:56.280 --> 20:01.280
+ We have, you know, we have a productivity suite of software
+
+20:02.280 --> 20:05.800
+ like Microsoft Dynamics, which, you know,
+
+20:05.800 --> 20:07.440
+ some people might not think is the sexiest thing
+
+20:07.440 --> 20:10.040
+ in the world, but it's like helping people figure out
+
+20:10.040 --> 20:12.720
+ how to automate all of their business processes
+
+20:12.720 --> 20:16.800
+ and workflows and to, you know, like help those businesses
+
+20:16.800 --> 20:19.120
+ using it to like grow and be more successful.
+
+20:19.120 --> 20:24.120
+ So it's a much broader vision in a way now
+
+20:24.240 --> 20:25.480
+ than it was back then.
+
+20:25.480 --> 20:27.400
+ Like it was sort of very particular thing.
+
+20:27.400 --> 20:29.280
+ And like now, like we live in this world
+
+20:29.280 --> 20:31.320
+ where technology is so powerful
+
+20:31.320 --> 20:36.320
+ and it's like such a basic fact of life
+
+20:36.320 --> 20:39.760
+ that it, you know, that it both exists
+
+20:39.760 --> 20:42.760
+ and is going to get better and better over time
+
+20:42.760 --> 20:46.000
+ or at least more and more powerful over time.
+
+20:46.000 --> 20:48.200
+ So like, you know, what you have to do as a platform player
+
+20:48.200 --> 20:49.920
+ is just much bigger.
+
+20:49.920 --> 20:50.760
+ Right.
+
+20:50.760 --> 20:52.600
+ There's so many directions in which you can transform.
+
+20:52.600 --> 20:55.160
+ You didn't mention mixed reality too.
+
+20:55.160 --> 20:59.200
+ You know, that's probably early days
+
+20:59.200 --> 21:00.680
+ or depends how you think of it.
+
+21:00.680 --> 21:02.240
+ But if we think in a scale of centuries,
+
+21:02.240 --> 21:04.120
+ it's the early days of mixed reality.
+
+21:04.120 --> 21:04.960
+ Oh, for sure.
+
+21:04.960 --> 21:08.280
+ And so yeah, with how it lands,
+
+21:08.280 --> 21:10.600
+ the Microsoft is doing some really interesting work there.
+
+21:10.600 --> 21:13.560
+ Do you touch that part of the effort?
+
+21:13.560 --> 21:14.840
+ What's the thinking?
+
+21:14.840 --> 21:17.640
+ Do you think of mixed reality as a platform too?
+
+21:17.640 --> 21:18.480
+ Oh, sure.
+
+21:18.480 --> 21:21.320
+ When we look at what the platforms of the future could be.
+
+21:21.320 --> 21:23.880
+ So like fairly obvious that like AI is one,
+
+21:23.880 --> 21:26.600
+ like you don't have to, I mean, like that's,
+
+21:26.600 --> 21:29.160
+ you know, you sort of say it to like someone
+
+21:29.160 --> 21:31.920
+ and you know, like they get it.
+
+21:31.920 --> 21:36.280
+ But like we also think of the like mixed reality
+
+21:36.280 --> 21:39.560
+ and quantum is like these two interesting,
+
+21:39.560 --> 21:40.920
+ you know, potentially.
+
+21:40.920 --> 21:41.800
+ Quantum computing.
+
+21:41.800 --> 21:42.640
+ Yeah.
+
+21:42.640 --> 21:44.520
+ Okay, so let's get crazy then.
+
+21:44.520 --> 21:48.920
+ So you're talking about some futuristic things here.
+
+21:48.920 --> 21:50.920
+ Well, the mixed reality Microsoft is really,
+
+21:50.920 --> 21:52.600
+ it's not even futuristic, it's here.
+
+21:52.600 --> 21:53.440
+ It is.
+
+21:53.440 --> 21:54.280
+ Incredible stuff.
+
+21:54.280 --> 21:56.680
+ And look, and it's having an impact right now.
+
+21:56.680 --> 21:58.720
+ Like one of the more interesting things
+
+21:58.720 --> 22:01.280
+ that's happened with mixed reality over the past
+
+22:01.280 --> 22:04.120
+ couple of years that I didn't clearly see
+
+22:04.120 --> 22:08.400
+ is that it's become the computing device
+
+22:08.400 --> 22:13.160
+ for folks who, for doing their work
+
+22:13.160 --> 22:16.040
+ who haven't used any computing device at all
+
+22:16.040 --> 22:16.960
+ to do their work before.
+
+22:16.960 --> 22:19.800
+ So technicians and service folks
+
+22:19.800 --> 22:24.200
+ and people who are doing like machine maintenance
+
+22:24.200 --> 22:25.280
+ on factory floors.
+
+22:25.280 --> 22:28.760
+ So like they, you know, because they're mobile
+
+22:28.760 --> 22:30.280
+ and like they're out in the world
+
+22:30.280 --> 22:32.320
+ and they're working with their hands
+
+22:32.320 --> 22:34.080
+ and, you know, sort of servicing these
+
+22:34.080 --> 22:36.520
+ like very complicated things.
+
+22:36.520 --> 22:39.440
+ They're, they don't use their mobile phone
+
+22:39.440 --> 22:41.440
+ and like they don't carry a laptop with them.
+
+22:41.440 --> 22:43.480
+ And, you know, they're not tethered to a desk.
+
+22:43.480 --> 22:46.920
+ And so mixed reality, like where it's getting
+
+22:46.920 --> 22:48.840
+ traction right now, where HoloLens is selling
+
+22:48.840 --> 22:53.840
+ a lot of units is for these sorts of applications
+
+22:53.880 --> 22:55.440
+ for these workers and it's become like,
+
+22:55.440 --> 22:58.040
+ I mean, like the people love it.
+
+22:58.040 --> 23:00.600
+ They're like, oh my God, like this is like,
+
+23:00.600 --> 23:02.840
+ for them like the same sort of productivity boosts
+
+23:02.840 --> 23:05.520
+ that, you know, like an office worker had
+
+23:05.520 --> 23:08.200
+ when they got their first personal computer.
+
+23:08.200 --> 23:09.800
+ Yeah, but you did mention,
+
+23:09.800 --> 23:13.400
+ it's certainly obvious AI as a platform,
+
+23:13.400 --> 23:15.560
+ but can we dig into it a little bit?
+
+23:15.560 --> 23:18.320
+ How does AI begin to infuse some of the products
+
+23:18.320 --> 23:19.480
+ in Microsoft?
+
+23:19.480 --> 23:24.480
+ So currently providing training of, for example,
+
+23:25.040 --> 23:26.760
+ neural networks in the cloud
+
+23:26.760 --> 23:30.960
+ or providing pre trained models
+
+23:30.960 --> 23:35.360
+ or just even providing computing resources
+
+23:35.360 --> 23:37.520
+ and whatever different inference
+
+23:37.520 --> 23:39.320
+ that you want to do using neural networks.
+
+23:39.320 --> 23:40.160
+ Yep.
+
+23:40.160 --> 23:43.560
+ Well, how do you think of AI infusing the,
+
+23:43.560 --> 23:45.880
+ as a platform that Microsoft can provide?
+
+23:45.880 --> 23:48.320
+ Yeah, I mean, I think it's, it's super interesting.
+
+23:48.320 --> 23:49.560
+ It's like everywhere.
+
+23:49.560 --> 23:54.560
+ And like we run these, we run these review meetings now
+
+23:54.560 --> 23:59.560
+ where it's me and Satya and like members of Satya's
+
+24:01.480 --> 24:04.600
+ leadership team and like a cross functional group
+
+24:04.600 --> 24:06.200
+ of folks across the entire company
+
+24:06.200 --> 24:11.200
+ who are working on like either AI infrastructure
+
+24:11.840 --> 24:15.520
+ or like have some substantial part of their,
+
+24:16.480 --> 24:21.480
+ of their product work using AI in some significant way.
+
+24:21.480 --> 24:23.440
+ Now, the important thing to understand is like,
+
+24:23.440 --> 24:27.040
+ when you think about like how the AI is going to manifest
+
+24:27.040 --> 24:29.600
+ in like an experience for something
+
+24:29.600 --> 24:30.760
+ that's going to make it better,
+
+24:30.760 --> 24:35.760
+ like I think you don't want the AI in this
+
+24:35.760 --> 24:37.760
+ to be the first order thing.
+
+24:37.760 --> 24:40.600
+ It's like whatever the product is and like the thing
+
+24:40.600 --> 24:42.440
+ that is trying to help you do,
+
+24:42.440 --> 24:44.560
+ like the AI just sort of makes it better.
+
+24:44.560 --> 24:46.840
+ And you know, this is a gross exaggeration,
+
+24:46.840 --> 24:50.680
+ but like I, yeah, people get super excited about it.
+
+24:50.680 --> 24:53.280
+ They're super excited about like where the AI is showing up
+
+24:53.280 --> 24:55.440
+ in products and I'm like, do you get that excited
+
+24:55.440 --> 24:59.880
+ about like where you're using a hash table like in your code?
+
+24:59.880 --> 25:03.200
+ Like it's just another, it's a very interesting
+
+25:03.200 --> 25:05.800
+ programming tool, but it's sort of like it's an engineering
+
+25:05.800 --> 25:09.560
+ tool and so like it shows up everywhere.
+
+25:09.560 --> 25:12.920
+ So like we've got dozens and dozens of features now
+
+25:12.920 --> 25:17.400
+ in office that are powered by like fairly sophisticated
+
+25:17.400 --> 25:22.200
+ machine learning, our search engine wouldn't work at all
+
+25:22.200 --> 25:24.840
+ if you took the machine learning out of it.
+
+25:24.840 --> 25:28.560
+ The like increasingly, you know,
+
+25:28.560 --> 25:33.560
+ things like content moderation on our Xbox and xCloud
+
+25:34.800 --> 25:35.960
+ platform.
+
+25:37.000 --> 25:39.160
+ When you mean moderation to me, like the recommender
+
+25:39.160 --> 25:41.760
+ is like showing what you want to look at next.
+
+25:41.760 --> 25:44.000
+ No, no, no, it's like anti bullying stuff.
+
+25:44.000 --> 25:47.040
+ So the usual social network stuff that you have to deal with.
+
+25:47.040 --> 25:47.880
+ Yeah, correct.
+
+25:47.880 --> 25:50.080
+ But it's like really it's targeted,
+
+25:50.080 --> 25:52.280
+ it's targeted towards a gaming audience.
+
+25:52.280 --> 25:55.320
+ So it's like a very particular type of thing where,
+
+25:55.320 --> 25:59.480
+ you know, the the line between playful banter
+
+25:59.480 --> 26:02.280
+ and like legitimate bullying is like a subtle one.
+
+26:02.280 --> 26:06.080
+ And like you have to, it's sort of tough.
+
+26:06.080 --> 26:09.080
+ Like I have, I love to, if we could dig into it
+
+26:09.080 --> 26:11.720
+ because you're also, you led the engineering efforts
+
+26:11.720 --> 26:14.920
+ of LinkedIn and if we look at,
+
+26:14.920 --> 26:17.640
+ if we look at LinkedIn as a social network
+
+26:17.640 --> 26:21.760
+ and if we look at the Xbox gaming as the social components,
+
+26:21.760 --> 26:24.840
+ the very different kinds of, I imagine communication
+
+26:24.840 --> 26:26.880
+ going on on the two platforms, right?
+
+26:26.880 --> 26:29.520
+ And the line in terms of bullying and so on
+
+26:29.520 --> 26:31.480
+ is different on the two platforms.
+
+26:31.480 --> 26:33.480
+ So how do you, I mean,
+
+26:33.480 --> 26:36.240
+ such a fascinating philosophical discussion
+
+26:36.240 --> 26:37.240
+ of where that line is.
+
+26:37.240 --> 26:39.840
+ I don't think anyone knows the right answer.
+
+26:39.840 --> 26:42.040
+ Twitter folks are under fire now,
+
+26:42.040 --> 26:45.120
+ Jack at Twitter for trying to find that line.
+
+26:45.120 --> 26:46.920
+ Nobody knows what that line is,
+
+26:46.920 --> 26:51.720
+ but how do you try to find the line for,
+
+26:52.480 --> 26:57.480
+ you know, trying to prevent abusive behavior
+
+26:58.040 --> 27:00.200
+ and at the same time let people be playful
+
+27:00.200 --> 27:02.880
+ and joke around and that kind of thing.
+
+27:02.880 --> 27:04.640
+ I think in a certain way, like, you know,
+
+27:04.640 --> 27:09.640
+ if you have what I would call vertical social networks,
+
+27:09.640 --> 27:12.200
+ it gets to be a little bit easier.
+
+27:12.200 --> 27:14.440
+ So like if you have a clear notion
+
+27:14.440 --> 27:17.960
+ of like what your social network should be used for
+
+27:17.960 --> 27:22.280
+ or like what you are designing a community around,
+
+27:22.280 --> 27:25.800
+ then you don't have as many dimensions
+
+27:25.800 --> 27:28.960
+ to your sort of content safety problem
+
+27:28.960 --> 27:33.720
+ as, you know, as you do in a general purpose platform.
+
+27:33.720 --> 27:37.520
+ I mean, so like on LinkedIn,
+
+27:37.520 --> 27:39.920
+ like the whole social network is about
+
+27:39.920 --> 27:41.560
+ connecting people with opportunity,
+
+27:41.560 --> 27:43.160
+ whether it's helping them find a job
+
+27:43.160 --> 27:46.280
+ or to, you know, sort of find mentors
+
+27:46.280 --> 27:49.320
+ or to, you know, sort of help them
+
+27:49.320 --> 27:52.120
+ like find their next sales lead
+
+27:52.120 --> 27:56.160
+ or to just sort of allow them to broadcast
+
+27:56.160 --> 27:59.440
+ their, you know, sort of professional identity
+
+27:59.440 --> 28:04.440
+ to their network of peers and collaborators
+
+28:04.440 --> 28:05.880
+ and, you know, sort of professional community.
+
+28:05.880 --> 28:07.400
+ Like that is, I mean, like in some ways,
+
+28:07.400 --> 28:08.960
+ like that's very, very broad,
+
+28:08.960 --> 28:12.480
+ but in other ways, it's sort of, you know, it's narrow.
+
+28:12.480 --> 28:17.480
+ And so like you can build AIs like machine learning systems
+
+28:18.360 --> 28:23.360
+ that are, you know, capable with those boundaries
+
+28:23.360 --> 28:26.200
+ of making better automated decisions about like,
+
+28:26.200 --> 28:28.240
+ what is, you know, sort of inappropriate
+
+28:28.240 --> 28:30.440
+ and offensive comment or dangerous comment
+
+28:30.440 --> 28:31.920
+ or illegal content.
+
+28:31.920 --> 28:34.800
+ When you have some constraints,
+
+28:34.800 --> 28:37.400
+ you know, same thing with, you know,
+
+28:37.400 --> 28:40.880
+ same thing with like the gaming social network.
+
+28:40.880 --> 28:42.680
+ So for instance, like it's about playing games,
+
+28:42.680 --> 28:44.880
+ about having fun and like the thing
+
+28:44.880 --> 28:47.240
+ that you don't want to have happen on the platform.
+
+28:47.240 --> 28:49.160
+ It's why bullying is such an important thing.
+
+28:49.160 --> 28:50.600
+ Like bullying is not fun.
+
+28:50.600 --> 28:53.400
+ So you want to do everything in your power
+
+28:53.400 --> 28:56.240
+ to encourage that not to happen.
+
+28:56.240 --> 29:00.320
+ And yeah, but I think that's a really important thing
+
+29:00.320 --> 29:03.920
+ but I think it's sort of a tough problem in general.
+
+29:03.920 --> 29:05.280
+ It's one where I think, you know,
+
+29:05.280 --> 29:07.120
+ eventually we're gonna have to have
+
+29:09.120 --> 29:13.800
+ some sort of clarification from our policy makers
+
+29:13.800 --> 29:17.400
+ about what it is that we should be doing,
+
+29:17.400 --> 29:20.880
+ like where the lines are, because it's tough.
+
+29:20.880 --> 29:23.760
+ Like you don't, like in democracy, right?
+
+29:23.760 --> 29:26.680
+ Like you don't want, you want some sort
+
+29:26.680 --> 29:28.880
+ of democratic involvement.
+
+29:28.880 --> 29:30.440
+ Like people should have a say
+
+29:30.440 --> 29:34.680
+ in like where the lines are drawn.
+
+29:34.680 --> 29:36.920
+ Like you don't want a bunch of people
+
+29:36.920 --> 29:39.480
+ making like unilateral decisions.
+
+29:39.480 --> 29:43.120
+ And like we are in a state right now
+
+29:43.120 --> 29:44.760
+ for some of these platforms where you actually
+
+29:44.760 --> 29:46.280
+ do have to make unilateral decisions
+
+29:46.280 --> 29:48.640
+ where the policy making isn't gonna happen fast enough
+
+29:48.640 --> 29:52.520
+ in order to like prevent very bad things from happening.
+
+29:52.520 --> 29:55.200
+ But like we need the policy making side of that
+
+29:55.200 --> 29:58.480
+ to catch up I think as quickly as possible
+
+29:58.480 --> 30:00.680
+ because you want that whole process
+
+30:00.680 --> 30:02.000
+ to be a democratic thing,
+
+30:02.000 --> 30:05.760
+ not a, you know, not some sort of weird thing
+
+30:05.760 --> 30:08.040
+ where you've got a non representative group
+
+30:08.040 --> 30:10.440
+ of people making decisions that have, you know,
+
+30:10.440 --> 30:12.520
+ like national and global impact.
+
+30:12.520 --> 30:14.720
+ And it's fascinating because the digital space
+
+30:14.720 --> 30:17.520
+ is different than the physical space
+
+30:17.520 --> 30:19.800
+ in which nations and governments were established.
+
+30:19.800 --> 30:23.960
+ And so what policy looks like globally,
+
+30:23.960 --> 30:25.760
+ what bullying looks like globally,
+
+30:25.760 --> 30:28.360
+ what healthy communication looks like globally
+
+30:28.360 --> 30:31.920
+ is an open question and we're all figuring it out together.
+
+30:31.920 --> 30:32.760
+ Which is fascinating.
+
+30:32.760 --> 30:37.160
+ Yeah, I mean with, you know, sort of fake news for instance
+
+30:37.160 --> 30:42.160
+ and deep fakes and fake news generated by humans.
+
+30:42.320 --> 30:44.600
+ Yeah, so we can talk about deep fakes.
+
+30:44.600 --> 30:46.120
+ Like I think that is another like, you know,
+
+30:46.120 --> 30:48.280
+ sort of very interesting level of complexity.
+
+30:48.280 --> 30:51.480
+ But like if you think about just the written word, right?
+
+30:51.480 --> 30:54.400
+ Like we have, you know, we invented Papyrus
+
+30:54.400 --> 30:56.760
+ what 3000 years ago where we, you know,
+
+30:56.760 --> 31:01.160
+ you could sort of put word on paper.
+
+31:01.160 --> 31:06.160
+ And then 500 years ago, like we get the printing press
+
+31:07.240 --> 31:11.480
+ like where the word gets a little bit more ubiquitous.
+
+31:11.480 --> 31:14.600
+ And then like you really, really didn't get ubiquitous
+
+31:14.600 --> 31:18.400
+ printed word until the end of the 19th century
+
+31:18.400 --> 31:20.720
+ when the offset press was invented.
+
+31:20.720 --> 31:22.360
+ And then, you know, just sort of explodes
+
+31:22.360 --> 31:25.360
+ and like, you know, the cross product of that
+
+31:25.360 --> 31:28.960
+ and the industrial revolutions need
+
+31:28.960 --> 31:32.880
+ for educated citizens resulted in like
+
+31:32.880 --> 31:34.720
+ this rapid expansion of literacy
+
+31:34.720 --> 31:36.000
+ and the rapid expansion of the word.
+
+31:36.000 --> 31:39.680
+ But like we had 3000 years up to that point
+
+31:39.680 --> 31:44.040
+ to figure out like how to, you know, like what's,
+
+31:44.040 --> 31:46.880
+ what's journalism, what's editorial integrity?
+
+31:46.880 --> 31:50.120
+ Like what's, you know, what's scientific peer review?
+
+31:50.120 --> 31:52.840
+ And so like you built all of this mechanism
+
+31:52.840 --> 31:57.080
+ to like try to filter through all of the noise
+
+31:57.080 --> 32:00.600
+ that the technology made possible to like, you know,
+
+32:00.600 --> 32:04.000
+ sort of getting to something that society could cope with.
+
+32:04.000 --> 32:06.600
+ And like, if you think about just the piece,
+
+32:06.600 --> 32:09.800
+ the PC didn't exist 50 years ago.
+
+32:09.800 --> 32:11.800
+ And so in like this span of, you know,
+
+32:11.800 --> 32:16.160
+ like half a century, like we've gone from no digital,
+
+32:16.160 --> 32:18.320
+ you know, no ubiquitous digital technology
+
+32:18.320 --> 32:21.080
+ to like having a device that sits in your pocket
+
+32:21.080 --> 32:23.760
+ where you can sort of say whatever is on your mind
+
+32:23.760 --> 32:26.800
+ to like what would Mary have
+
+32:26.800 --> 32:31.800
+ and Mary Meeker just released her new like slide deck last week.
+
+32:32.440 --> 32:37.360
+ You know, we've got 50% penetration of the internet
+
+32:37.360 --> 32:38.520
+ to the global population.
+
+32:38.520 --> 32:40.280
+ Like there are like three and a half billion people
+
+32:40.280 --> 32:41.720
+ who are connected now.
+
+32:41.720 --> 32:43.720
+ So it's like, it's crazy, crazy.
+
+32:43.720 --> 32:45.000
+ They're like inconceivable,
+
+32:45.000 --> 32:46.480
+ like how fast all of this happened.
+
+32:46.480 --> 32:48.720
+ So, you know, it's not surprising
+
+32:48.720 --> 32:51.000
+ that we haven't figured out what to do yet,
+
+32:51.000 --> 32:55.640
+ but like we gotta really like lean into this set of problems
+
+32:55.640 --> 33:00.200
+ because like we basically have three millennia worth of work
+
+33:00.200 --> 33:02.520
+ to do about how to deal with all of this
+
+33:02.520 --> 33:05.800
+ and like probably what amounts to the next decade
+
+33:05.800 --> 33:07.040
+ worth of time.
+
+33:07.040 --> 33:09.960
+ So since we're on the topic of tough, you know,
+
+33:09.960 --> 33:11.600
+ tough challenging problems,
+
+33:11.600 --> 33:15.200
+ let's look at more on the tooling side in AI
+
+33:15.200 --> 33:18.440
+ that Microsoft is looking at as face recognition software.
+
+33:18.440 --> 33:21.840
+ So there's a lot of powerful positive use cases
+
+33:21.840 --> 33:24.240
+ for face recognition, but there's some negative ones
+
+33:24.240 --> 33:27.200
+ and we're seeing those in different governments
+
+33:27.200 --> 33:28.160
+ in the world.
+
+33:28.160 --> 33:30.240
+ So how do you, how does Microsoft think
+
+33:30.240 --> 33:33.880
+ about the use of face recognition software
+
+33:33.880 --> 33:38.880
+ as a platform in governments and companies?
+
+33:39.400 --> 33:42.280
+ Yeah, how do we strike an ethical balance here?
+
+33:42.280 --> 33:47.280
+ Yeah, I think we've articulated a clear point of view.
+
+33:47.280 --> 33:51.840
+ So Brad Smith wrote a blog post last fall,
+
+33:51.840 --> 33:54.120
+ I believe that sort of like outline,
+
+33:54.120 --> 33:57.000
+ like very specifically what, you know,
+
+33:57.000 --> 33:59.280
+ what our point of view is there.
+
+33:59.280 --> 34:02.240
+ And, you know, I think we believe that there are certain uses
+
+34:02.240 --> 34:04.680
+ to which face recognition should not be put
+
+34:04.680 --> 34:09.160
+ and we believe again that there's a need for regulation there.
+
+34:09.160 --> 34:12.440
+ Like the government should like really come in and say
+
+34:12.440 --> 34:15.720
+ that, you know, this is where the lines are.
+
+34:15.720 --> 34:18.600
+ And like we very much wanted to like figuring out
+
+34:18.600 --> 34:20.680
+ where the lines are should be a democratic process.
+
+34:20.680 --> 34:23.240
+ But in the short term, like we've drawn some lines
+
+34:23.240 --> 34:26.640
+ where, you know, we push back against uses
+
+34:26.640 --> 34:29.440
+ of face recognition technology.
+
+34:29.440 --> 34:32.480
+ You know, like this city of San Francisco, for instance,
+
+34:32.480 --> 34:36.480
+ I think has completely outlawed any government agency
+
+34:36.480 --> 34:39.560
+ from using face recognition tech.
+
+34:39.560 --> 34:44.560
+ And like that may prove to be a little bit overly broad.
+
+34:44.560 --> 34:48.840
+ But for like certain law enforcement things,
+
+34:48.840 --> 34:53.840
+ like you really, I would personally rather be overly
+
+34:54.040 --> 34:57.400
+ sort of cautious in terms of restricting use of it
+
+34:57.400 --> 34:58.920
+ until like we have, you know,
+
+34:58.920 --> 35:02.160
+ sort of defined a reasonable, you know,
+
+35:02.160 --> 35:04.880
+ democratically determined regulatory framework
+
+35:04.880 --> 35:08.840
+ for like where we could and should use it.
+
+35:08.840 --> 35:10.880
+ And, you know, the other thing there is
+
+35:11.960 --> 35:14.000
+ like we've got a bunch of research that we're doing
+
+35:14.000 --> 35:18.400
+ and a bunch of progress that we've made on bias there.
+
+35:18.400 --> 35:20.880
+ And like there are all sorts of like weird biases
+
+35:20.880 --> 35:23.640
+ that these models can have like all the way
+
+35:23.640 --> 35:26.920
+ from like the most noteworthy one where, you know,
+
+35:26.920 --> 35:31.680
+ you may have underrepresented minorities
+
+35:31.680 --> 35:34.680
+ who are like underrepresented in the training data.
+
+35:34.680 --> 35:39.240
+ And then you start learning like strange things.
+
+35:39.240 --> 35:42.160
+ But like they're even, you know, other weird things
+
+35:42.160 --> 35:46.480
+ like we've, I think we've seen in the public research
+
+35:46.480 --> 35:49.520
+ like models can learn strange things
+
+35:49.520 --> 35:54.520
+ like all doctors or men for instance.
+
+35:54.520 --> 35:59.520
+ Yeah, I mean, and so like it really is a thing where
+
+36:00.760 --> 36:03.600
+ it's very important for everybody
+
+36:03.600 --> 36:08.440
+ who is working on these things before they push publish,
+
+36:08.440 --> 36:12.800
+ they launch the experiment, they, you know, push the code
+
+36:12.800 --> 36:17.120
+ to, you know, online or they even publish the paper
+
+36:17.120 --> 36:20.040
+ that they are at least starting to think
+
+36:20.040 --> 36:25.040
+ about what some of the potential negative consequences
+
+36:25.040 --> 36:25.880
+ are some of this stuff.
+
+36:25.880 --> 36:29.040
+ I mean, this is where, you know, like the deep fake stuff
+
+36:29.040 --> 36:32.360
+ I find very worrisome just because
+
+36:32.360 --> 36:37.360
+ they're going to be some very good beneficial uses
+
+36:39.800 --> 36:44.800
+ of like GAN generated imagery.
+
+36:46.080 --> 36:48.440
+ And like, and funny enough, like one of the places
+
+36:48.440 --> 36:52.920
+ where it's actually useful is we're using the technology
+
+36:52.920 --> 36:57.920
+ right now to generate synthetic, synthetic visual data
+
+36:58.640 --> 37:01.160
+ for training some of the face recognition models
+
+37:01.160 --> 37:03.440
+ to get rid of the bias.
+
+37:03.440 --> 37:05.800
+ So like that's one like super good use of the tech,
+
+37:05.800 --> 37:09.640
+ but like, you know, it's getting good enough now
+
+37:09.640 --> 37:12.320
+ where, you know, it's going to sort of challenge
+
+37:12.320 --> 37:15.400
+ a normal human beings ability to like now you're just sort
+
+37:15.400 --> 37:19.320
+ of say like it's very expensive for someone
+
+37:19.320 --> 37:23.280
+ to fabricate a photorealistic fake video.
+
+37:24.200 --> 37:26.920
+ And like GANs are going to make it fantastically cheap
+
+37:26.920 --> 37:30.440
+ to fabricate a photorealistic fake video.
+
+37:30.440 --> 37:33.920
+ And so like what you assume you can sort of trust
+
+37:33.920 --> 37:38.400
+ is true versus like be skeptical about is about to change.
+
+37:38.400 --> 37:40.560
+ And like we're not ready for it, I don't think.
+
+37:40.560 --> 37:42.000
+ The nature of truth, right?
+
+37:42.000 --> 37:46.360
+ That's, it's also exciting because I think both you
+
+37:46.360 --> 37:49.600
+ and I probably would agree that the way to solve,
+
+37:49.600 --> 37:52.080
+ to take on that challenge is with technology.
+
+37:52.080 --> 37:52.920
+ Yeah. Right.
+
+37:52.920 --> 37:56.800
+ There's probably going to be ideas of ways to verify
+
+37:56.800 --> 38:00.800
+ which kind of video is legitimate, which kind is not.
+
+38:00.800 --> 38:03.880
+ So to me, that's an exciting possibility.
+
+38:03.880 --> 38:07.160
+ Most likely for just the comedic genius
+
+38:07.160 --> 38:10.960
+ that the internet usually creates with these kinds of videos.
+
+38:10.960 --> 38:13.960
+ And hopefully will not result in any serious harm.
+
+38:13.960 --> 38:17.680
+ Yeah. And it could be, you know, like I think
+
+38:17.680 --> 38:22.680
+ we will have technology to that may be able to detect
+
+38:23.040 --> 38:24.440
+ whether or not something's fake or real.
+
+38:24.440 --> 38:29.440
+ Although the fakes are pretty convincing
+
+38:30.160 --> 38:34.360
+ even like when you subject them to machine scrutiny.
+
+38:34.360 --> 38:37.800
+ But, you know, we also have these increasingly
+
+38:37.800 --> 38:40.520
+ interesting social networks, you know,
+
+38:40.520 --> 38:45.520
+ that are under fire right now for some of the bad things
+
+38:45.800 --> 38:46.640
+ that they do.
+
+38:46.640 --> 38:47.720
+ Like one of the things you could choose to do
+
+38:47.720 --> 38:51.760
+ with a social network is like you could,
+
+38:51.760 --> 38:55.560
+ you could use crypto and the networks
+
+38:55.560 --> 38:59.960
+ to like have content signed where you could have a like
+
+38:59.960 --> 39:02.160
+ full chain of custody that accompanied
+
+39:02.160 --> 39:03.920
+ every piece of content.
+
+39:03.920 --> 39:06.800
+ So like when you're viewing something
+
+39:06.800 --> 39:09.640
+ and like you want to ask yourself like how, you know,
+
+39:09.640 --> 39:11.040
+ how much can I trust this?
+
+39:11.040 --> 39:12.400
+ Like you can click something
+
+39:12.400 --> 39:15.640
+ and like have a verified chain of custody that shows like,
+
+39:15.640 --> 39:19.040
+ oh, this is coming from, you know, from this source.
+
+39:19.040 --> 39:24.040
+ And it's like signed by like someone whose identity I trust.
+
+39:24.080 --> 39:25.400
+ Yeah, I think having that, you know,
+
+39:25.400 --> 39:28.040
+ having that chain of custody like being able to like say,
+
+39:28.040 --> 39:31.200
+ oh, here's this video, like it may or may not
+
+39:31.200 --> 39:33.760
+ been produced using some of this deep fake technology.
+
+39:33.760 --> 39:35.640
+ But if you've got a verified chain of custody
+
+39:35.640 --> 39:37.800
+ where you can sort of trace it all the way back
+
+39:37.800 --> 39:39.960
+ to an identity and you can decide whether or not
+
+39:39.960 --> 39:41.520
+ like I trust this identity.
+
+39:41.520 --> 39:43.360
+ Like, oh no, this is really from the White House
+
+39:43.360 --> 39:45.480
+ or like this is really from the, you know,
+
+39:45.480 --> 39:48.840
+ the office of this particular presidential candidate
+
+39:48.840 --> 39:50.960
+ or it's really from, you know,
+
+39:50.960 --> 39:55.520
+ Jeff Wiener CEO of LinkedIn or Satya Nadella CEO of Microsoft.
+
+39:55.520 --> 39:58.400
+ Like that might be like one way
+
+39:58.400 --> 39:59.960
+ that you can solve some of the problems.
+
+39:59.960 --> 40:01.800
+ So like that's not the super high tech.
+
+40:01.800 --> 40:04.480
+ Like we've had all of this technology forever.
+
+40:04.480 --> 40:06.720
+ And but I think you're right.
+
+40:06.720 --> 40:11.120
+ Like it has to be some sort of technological thing
+
+40:11.120 --> 40:15.840
+ because the underlying tech that is used to create this
+
+40:15.840 --> 40:18.800
+ is not going to do anything but get better over time
+
+40:18.800 --> 40:21.160
+ and the genie is sort of out of the bottle.
+
+40:21.160 --> 40:22.800
+ There's no stuffing it back in.
+
+40:22.800 --> 40:24.520
+ And there's a social component
+
+40:24.520 --> 40:26.600
+ which I think is really healthy for democracy
+
+40:26.600 --> 40:30.200
+ where people will be skeptical about the thing they watch.
+
+40:30.200 --> 40:31.040
+ Yeah.
+
+40:31.040 --> 40:34.160
+ In general, so, you know, which is good.
+
+40:34.160 --> 40:37.280
+ Skepticism in general is good for your personal content.
+
+40:37.280 --> 40:40.400
+ So deep fakes in that sense are creating
+
+40:40.400 --> 40:44.800
+ global skepticism about can they trust what they read?
+
+40:44.800 --> 40:46.880
+ It encourages further research.
+
+40:46.880 --> 40:48.840
+ I come from the Soviet Union
+
+40:49.800 --> 40:53.320
+ where basically nobody trusted the media
+
+40:53.320 --> 40:55.120
+ because you knew it was propaganda.
+
+40:55.120 --> 40:59.160
+ And that kind of skepticism encouraged further research
+
+40:59.160 --> 41:02.360
+ about ideas supposed to just trusting anyone's source.
+
+41:02.360 --> 41:05.440
+ Well, like I think it's one of the reasons why the,
+
+41:05.440 --> 41:09.440
+ you know, the scientific method and our apparatus
+
+41:09.440 --> 41:11.480
+ of modern science is so good.
+
+41:11.480 --> 41:15.360
+ Like because you don't have to trust anything.
+
+41:15.360 --> 41:18.520
+ Like you, like the whole notion of, you know,
+
+41:18.520 --> 41:21.320
+ like modern science beyond the fact that, you know,
+
+41:21.320 --> 41:23.440
+ this is a hypothesis and this is an experiment
+
+41:23.440 --> 41:24.840
+ to test the hypothesis.
+
+41:24.840 --> 41:27.360
+ And, you know, like this is a peer review process
+
+41:27.360 --> 41:30.080
+ for scrutinizing published results.
+
+41:30.080 --> 41:33.280
+ But like stuff's also supposed to be reproducible.
+
+41:33.280 --> 41:35.240
+ So like, you know, it's been vetted by this process,
+
+41:35.240 --> 41:38.000
+ but like you also are expected to publish enough detail
+
+41:38.000 --> 41:41.480
+ where, you know, if you are sufficiently skeptical
+
+41:41.480 --> 41:44.720
+ of the thing, you can go try to like reproduce it yourself.
+
+41:44.720 --> 41:47.560
+ And like, I don't know what it is.
+
+41:47.560 --> 41:49.920
+ Like, I think a lot of engineers are like this
+
+41:49.920 --> 41:52.600
+ where like, you know, sort of this, like your brain
+
+41:52.600 --> 41:55.520
+ is sort of wired for skepticism.
+
+41:55.520 --> 41:58.000
+ Like you don't just first order trust everything
+
+41:58.000 --> 42:00.040
+ that you see and encounter.
+
+42:00.040 --> 42:02.560
+ And like you're sort of curious to understand,
+
+42:02.560 --> 42:04.480
+ you know, the next thing.
+
+42:04.480 --> 42:09.080
+ But like, I think it's an entirely healthy thing.
+
+42:09.080 --> 42:12.280
+ And like we need a little bit more of that right now.
+
+42:12.280 --> 42:16.200
+ So I'm not a large business owner.
+
+42:16.200 --> 42:23.200
+ So I'm just, I'm just a huge fan of many of Microsoft products.
+
+42:23.200 --> 42:25.360
+ I mean, I still, actually in terms of,
+
+42:25.360 --> 42:27.000
+ I generate a lot of graphics and images
+
+42:27.000 --> 42:28.640
+ and I still use PowerPoint to do that.
+
+42:28.640 --> 42:30.440
+ It beats Illustrator for me.
+
+42:30.440 --> 42:34.480
+ Even professional sort of, it's fascinating.
+
+42:34.480 --> 42:39.560
+ So I wonder what is the future of, let's say,
+
+42:39.560 --> 42:41.920
+ windows and office look like?
+
+42:41.920 --> 42:43.840
+ Is do you see it?
+
+42:43.840 --> 42:45.880
+ I mean, I remember looking forward to XP.
+
+42:45.880 --> 42:48.200
+ Was it exciting when XP was released?
+
+42:48.200 --> 42:51.080
+ Just like you said, I don't remember when 95 was released.
+
+42:51.080 --> 42:53.800
+ But XP for me was a big celebration.
+
+42:53.800 --> 42:56.000
+ And when 10 came out, I was like,
+
+42:56.000 --> 42:58.040
+ okay, well, it's nice, it's a nice improvement.
+
+42:58.040 --> 43:02.600
+ But so what do you see the future of these products?
+
+43:02.600 --> 43:04.640
+ You know, I think there's a bunch of excitement.
+
+43:04.640 --> 43:07.160
+ I mean, on the office front,
+
+43:07.160 --> 43:13.440
+ there's going to be this like increasing productivity
+
+43:13.440 --> 43:17.080
+ wins that are coming out of some of these AI powered features
+
+43:17.080 --> 43:19.000
+ that are coming, like the products will sort of get
+
+43:19.000 --> 43:21.120
+ smarter and smarter in like a very subtle way.
+
+43:21.120 --> 43:24.120
+ Like there's not going to be this big bang moment
+
+43:24.120 --> 43:27.080
+ where, you know, like Clippy is going to reemerge
+
+43:27.080 --> 43:27.960
+ and it's going to be...
+
+43:27.960 --> 43:28.680
+ Wait a minute.
+
+43:28.680 --> 43:30.520
+ Okay, well, I have to wait, wait, wait.
+
+43:30.520 --> 43:31.960
+ It's Clippy coming back.
+
+43:31.960 --> 43:34.560
+ Well, quite seriously.
+
+43:34.560 --> 43:37.920
+ So injection of AI, there's not much,
+
+43:37.920 --> 43:39.040
+ or at least I'm not familiar,
+
+43:39.040 --> 43:41.200
+ sort of assistive type of stuff going on
+
+43:41.200 --> 43:43.600
+ inside the office products,
+
+43:43.600 --> 43:47.600
+ like a Clippy style assistant, personal assistant.
+
+43:47.600 --> 43:50.560
+ Do you think that there's a possibility
+
+43:50.560 --> 43:52.000
+ of that in the future?
+
+43:52.000 --> 43:54.680
+ So I think there are a bunch of like very small ways
+
+43:54.680 --> 43:57.320
+ in which like machine learning power
+
+43:57.320 --> 44:00.080
+ and assistive things are in the product right now.
+
+44:00.080 --> 44:04.800
+ So there are a bunch of interesting things,
+
+44:04.800 --> 44:09.280
+ like the auto response stuff's getting better and better
+
+44:09.280 --> 44:12.160
+ and it's like getting to the point where, you know,
+
+44:12.160 --> 44:14.960
+ it can auto respond with like, okay,
+
+44:14.960 --> 44:19.080
+ let this person is clearly trying to schedule a meeting
+
+44:19.080 --> 44:21.520
+ so it looks at your calendar and it automatically
+
+44:21.520 --> 44:24.080
+ like tries to find like a time and a space
+
+44:24.080 --> 44:26.240
+ that's mutually interesting.
+
+44:26.240 --> 44:31.240
+ Like we have this notion of Microsoft search
+
+44:33.520 --> 44:34.960
+ where it's like not just web search,
+
+44:34.960 --> 44:38.200
+ but it's like search across like all of your information
+
+44:38.200 --> 44:43.200
+ that's sitting inside of like your Office 365 tenant
+
+44:43.320 --> 44:46.880
+ and like, you know, potentially in other products.
+
+44:46.880 --> 44:49.680
+ And like we have this thing called the Microsoft Graph
+
+44:49.680 --> 44:53.400
+ that is basically a API federator that, you know,
+
+44:53.400 --> 44:57.960
+ sort of like gets you hooked up across the entire breadth
+
+44:57.960 --> 44:59.760
+ of like all of the, you know,
+
+44:59.760 --> 45:01.640
+ like what were information silos
+
+45:01.640 --> 45:04.720
+ before they got woven together with the graph.
+
+45:05.680 --> 45:07.880
+ Like that is like getting increasing
+
+45:07.880 --> 45:09.160
+ with increasing effectiveness,
+
+45:09.160 --> 45:11.280
+ sort of plumbed into the,
+
+45:11.280 --> 45:13.120
+ into some of these auto response things
+
+45:13.120 --> 45:15.840
+ where you're going to be able to see the system
+
+45:15.840 --> 45:18.200
+ like automatically retrieve information for you.
+
+45:18.200 --> 45:21.160
+ Like if, you know, like I frequently send out,
+
+45:21.160 --> 45:24.080
+ you know, emails to folks where like I can't find a paper
+
+45:24.080 --> 45:25.400
+ or a document or whatnot.
+
+45:25.400 --> 45:26.840
+ There's no reason why the system won't be able
+
+45:26.840 --> 45:27.680
+ to do that for you.
+
+45:27.680 --> 45:29.560
+ And like, I think the,
+
+45:29.560 --> 45:33.640
+ it's building towards like having things that look more
+
+45:33.640 --> 45:37.880
+ like like a fully integrated, you know, assistant,
+
+45:37.880 --> 45:40.720
+ but like you'll have a bunch of steps
+
+45:40.720 --> 45:42.800
+ that you will see before you,
+
+45:42.800 --> 45:45.120
+ like it will not be this like big bang thing
+
+45:45.120 --> 45:47.400
+ where like Clippy comes back and you've got this like,
+
+45:47.400 --> 45:49.360
+ you know, manifestation of, you know,
+
+45:49.360 --> 45:52.000
+ like a fully, fully powered assistant.
+
+45:53.320 --> 45:56.920
+ So I think that's, that's definitely coming out.
+
+45:56.920 --> 45:58.680
+ Like all of the, you know, collaboration,
+
+45:58.680 --> 46:00.720
+ co authoring stuff's getting better.
+
+46:00.720 --> 46:02.200
+ You know, it's like really interesting.
+
+46:02.200 --> 46:07.200
+ Like if you look at how we use the office product portfolio
+
+46:08.320 --> 46:10.840
+ at Microsoft, like more and more of it is happening
+
+46:10.840 --> 46:14.480
+ inside of like teams as a canvas.
+
+46:14.480 --> 46:17.160
+ And like it's this thing where, you know,
+
+46:17.160 --> 46:19.840
+ that you've got collaboration is like
+
+46:19.840 --> 46:21.560
+ at the center of the product.
+
+46:21.560 --> 46:26.560
+ And like we, we, we built some like really cool stuff
+
+46:26.720 --> 46:29.440
+ that's some of, which is about to be open source
+
+46:29.440 --> 46:33.120
+ that are sort of framework level things for doing,
+
+46:33.120 --> 46:35.600
+ for doing co authoring.
+
+46:35.600 --> 46:36.440
+ That's awesome.
+
+46:36.440 --> 46:38.920
+ So in, is there a cloud component to that?
+
+46:38.920 --> 46:41.880
+ So on the web or is it,
+
+46:41.880 --> 46:43.640
+ forgive me if I don't already know this,
+
+46:43.640 --> 46:45.600
+ but with office 365,
+
+46:45.600 --> 46:48.480
+ we still, the collaboration we do, if we're doing Word,
+
+46:48.480 --> 46:50.640
+ we're still sending the file around.
+
+46:50.640 --> 46:51.480
+ No, no, no, no.
+
+46:51.480 --> 46:53.400
+ So this is,
+
+46:53.400 --> 46:55.240
+ we're already a little bit better than that.
+
+46:55.240 --> 46:57.360
+ And like, you know, so like the fact that you're unaware
+
+46:57.360 --> 46:59.120
+ of it means we've got a better job to do,
+
+46:59.120 --> 47:01.960
+ like helping you discover, discover this stuff.
+
+47:02.880 --> 47:06.360
+ But yeah, I mean, it's already like got a huge,
+
+47:06.360 --> 47:07.200
+ huge cloud component.
+
+47:07.200 --> 47:09.680
+ And like part of, you know, part of this framework stuff,
+
+47:09.680 --> 47:12.640
+ I think we're calling it, like I,
+
+47:12.640 --> 47:14.520
+ like we've been working on it for a couple of years.
+
+47:14.520 --> 47:17.200
+ So like, I know the, the internal OLA code name for it,
+
+47:17.200 --> 47:18.640
+ but I think when we launched it to build,
+
+47:18.640 --> 47:20.720
+ it's called the fluid framework.
+
+47:21.920 --> 47:25.080
+ And, but like what fluid lets you do is like,
+
+47:25.080 --> 47:27.920
+ you can go into a conversation that you're having in teams
+
+47:27.920 --> 47:30.280
+ and like reference, like part of a spreadsheet
+
+47:30.280 --> 47:32.600
+ that you're working on,
+
+47:32.600 --> 47:35.600
+ where somebody's like sitting in the Excel canvas,
+
+47:35.600 --> 47:37.760
+ like working on the spreadsheet with a, you know,
+
+47:37.760 --> 47:39.120
+ charter whatnot.
+
+47:39.120 --> 47:42.000
+ And like, you can sort of embed like part of the spreadsheet
+
+47:42.000 --> 47:43.240
+ in the team's conversation,
+
+47:43.240 --> 47:46.520
+ where like you can dynamically update in like all
+
+47:46.520 --> 47:49.400
+ of the changes that you're making to the,
+
+47:49.400 --> 47:51.280
+ to this object or like, you know,
+
+47:51.280 --> 47:54.680
+ coordinate and everything is sort of updating in real time.
+
+47:54.680 --> 47:58.000
+ So like you can be in whatever canvas is most convenient
+
+47:58.000 --> 48:00.400
+ for you to get your work done.
+
+48:00.400 --> 48:03.400
+ So out of my own sort of curiosity as an engineer,
+
+48:03.400 --> 48:06.280
+ I know what it's like to sort of lead a team
+
+48:06.280 --> 48:08.280
+ of 10, 15 engineers.
+
+48:08.280 --> 48:11.680
+ Microsoft has, I don't know what the numbers are,
+
+48:11.680 --> 48:14.920
+ maybe 15, maybe 60,000 engineers, maybe 40.
+
+48:14.920 --> 48:16.160
+ I don't know exactly what the number is.
+
+48:16.160 --> 48:17.000
+ It's a lot.
+
+48:17.000 --> 48:18.520
+ It's tens of thousands.
+
+48:18.520 --> 48:20.640
+ Right. This is more than 10 or 15.
+
+48:23.640 --> 48:28.640
+ I mean, you've led different sizes,
+
+48:28.720 --> 48:30.560
+ mostly large sizes of engineers.
+
+48:30.560 --> 48:33.840
+ What does it take to lead such a large group
+
+48:33.840 --> 48:37.480
+ into a continue innovation,
+
+48:37.480 --> 48:40.240
+ continue being highly productive
+
+48:40.240 --> 48:43.200
+ and yet develop all kinds of new ideas
+
+48:43.200 --> 48:45.120
+ and yet maintain like, what does it take
+
+48:45.120 --> 48:49.000
+ to lead such a large group of brilliant people?
+
+48:49.000 --> 48:52.080
+ I think the thing that you learn
+
+48:52.080 --> 48:55.120
+ as you manage larger and larger scale
+
+48:55.120 --> 48:57.920
+ is that there are three things
+
+48:57.920 --> 49:00.480
+ that are like very, very important
+
+49:00.480 --> 49:02.360
+ for big engineering teams.
+
+49:02.360 --> 49:06.320
+ Like one is like having some sort of forethought
+
+49:06.320 --> 49:09.840
+ about what it is that you're going to be building
+
+49:09.840 --> 49:11.040
+ over large periods of time.
+
+49:11.040 --> 49:11.880
+ Like not exactly.
+
+49:11.880 --> 49:13.760
+ Like you don't need to know that like,
+
+49:13.760 --> 49:16.440
+ I'm putting all my chips on this one product
+
+49:16.440 --> 49:17.760
+ and like this is going to be the thing.
+
+49:17.760 --> 49:21.440
+ But it's useful to know what sort of capabilities
+
+49:21.440 --> 49:23.080
+ you think you're going to need to have
+
+49:23.080 --> 49:24.720
+ to build the products of the future
+
+49:24.720 --> 49:28.000
+ and then like invest in that infrastructure.
+
+49:28.000 --> 49:31.520
+ Like whether, and I'm not just talking about storage systems
+
+49:31.520 --> 49:33.480
+ or cloud APIs, it's also like,
+
+49:33.480 --> 49:35.360
+ what is your development process look like?
+
+49:35.360 --> 49:36.720
+ What tools do you want?
+
+49:36.720 --> 49:39.560
+ Like what culture do you want to build
+
+49:39.560 --> 49:42.760
+ around like how you're sort of collaborating together
+
+49:42.760 --> 49:45.720
+ to like make complicated technical things?
+
+49:45.720 --> 49:48.080
+ And so like having an opinion and investing in that
+
+49:48.080 --> 49:50.480
+ is like, it just gets more and more important.
+
+49:50.480 --> 49:54.520
+ And like the sooner you can get a concrete set of opinions,
+
+49:54.520 --> 49:57.680
+ like the better you're going to be.
+
+49:57.680 --> 50:01.600
+ Like you can wing it for a while at small scales.
+
+50:01.600 --> 50:03.160
+ Like, you know, when you start a company,
+
+50:03.160 --> 50:06.320
+ like you don't have to be like super specific about it.
+
+50:06.320 --> 50:10.000
+ But like the biggest miseries that I've ever seen
+
+50:10.000 --> 50:12.640
+ as an engineering leader are in places
+
+50:12.640 --> 50:14.440
+ where you didn't have a clear enough opinion
+
+50:14.440 --> 50:16.800
+ about those things soon enough.
+
+50:16.800 --> 50:20.240
+ And then you just sort of go create a bunch of technical debt
+
+50:20.240 --> 50:24.000
+ and like culture debt that is excruciatingly painful
+
+50:24.000 --> 50:25.760
+ to clean up.
+
+50:25.760 --> 50:28.640
+ So like that's one bundle of things.
+
+50:28.640 --> 50:33.640
+ Like the other, you know, another bundle of things is
+
+50:33.640 --> 50:37.440
+ like it's just really, really important to
+
+50:38.960 --> 50:43.960
+ like have a clear mission that's not just some cute crap
+
+50:45.520 --> 50:48.880
+ you say because like you think you should have a mission,
+
+50:48.880 --> 50:52.880
+ but like something that clarifies for people
+
+50:52.880 --> 50:55.680
+ like where it is that you're headed together.
+
+50:57.160 --> 50:58.520
+ Like I know it's like probably
+
+50:58.520 --> 51:00.320
+ like a little bit too popular right now,
+
+51:00.320 --> 51:05.320
+ but Yval Harari's book, Sapiens,
+
+51:07.240 --> 51:12.240
+ one of the central ideas in his book is that
+
+51:12.440 --> 51:16.840
+ like storytelling is like the quintessential thing
+
+51:16.840 --> 51:20.480
+ for coordinating the activities of large groups of people.
+
+51:20.480 --> 51:22.320
+ Like once you get past Dunbar's number
+
+51:23.360 --> 51:25.800
+ and like I've really, really seen that
+
+51:25.800 --> 51:27.320
+ just managing engineering teams.
+
+51:27.320 --> 51:32.080
+ Like you can just brute force things
+
+51:32.080 --> 51:35.160
+ when you're less than 120, 150 folks
+
+51:35.160 --> 51:37.520
+ where you can sort of know and trust
+
+51:37.520 --> 51:40.920
+ and understand what the dynamics are between all the people.
+
+51:40.920 --> 51:41.840
+ But like past that,
+
+51:41.840 --> 51:45.440
+ like things just sort of start to catastrophically fail
+
+51:45.440 --> 51:48.760
+ if you don't have some sort of set of shared goals
+
+51:48.760 --> 51:50.480
+ that you're marching towards.
+
+51:50.480 --> 51:52.960
+ And so like even though it sounds touchy feely
+
+51:52.960 --> 51:55.640
+ and you know, like a bunch of technical people
+
+51:55.640 --> 51:58.200
+ will sort of balk at the idea that like you need
+
+51:58.200 --> 52:01.680
+ to like have a clear, like the missions
+
+52:01.680 --> 52:03.560
+ like very, very, very important.
+
+52:03.560 --> 52:04.640
+ Yval's right, right?
+
+52:04.640 --> 52:07.520
+ Stories, that's how our society,
+
+52:07.520 --> 52:09.360
+ that's the fabric that connects us all of us
+
+52:09.360 --> 52:11.120
+ is these powerful stories.
+
+52:11.120 --> 52:13.440
+ And that works for companies too, right?
+
+52:13.440 --> 52:14.520
+ It works for everything.
+
+52:14.520 --> 52:16.520
+ Like I mean, even down to like, you know,
+
+52:16.520 --> 52:18.280
+ you sort of really think about like our currency
+
+52:18.280 --> 52:19.960
+ for instance is a story.
+
+52:19.960 --> 52:23.360
+ Our constitution is a story, our laws are story.
+
+52:23.360 --> 52:27.840
+ I mean, like we believe very, very, very strongly in them
+
+52:27.840 --> 52:29.960
+ and thank God we do.
+
+52:29.960 --> 52:33.040
+ But like they are, they're just abstract things.
+
+52:33.040 --> 52:34.000
+ Like they're just words.
+
+52:34.000 --> 52:36.520
+ Like if we don't believe in them, they're nothing.
+
+52:36.520 --> 52:39.440
+ And in some sense, those stories are platforms
+
+52:39.440 --> 52:43.040
+ and the kinds some of which Microsoft is creating, right?
+
+52:43.040 --> 52:46.360
+ Yeah, platforms in which we define the future.
+
+52:46.360 --> 52:48.600
+ So last question, what do you,
+
+52:48.600 --> 52:50.080
+ let's get philosophical maybe,
+
+52:50.080 --> 52:51.480
+ bigger than even Microsoft.
+
+52:51.480 --> 52:56.280
+ What do you think the next 2030 plus years
+
+52:56.280 --> 53:00.120
+ looks like for computing, for technology, for devices?
+
+53:00.120 --> 53:03.760
+ Do you have crazy ideas about the future of the world?
+
+53:04.600 --> 53:06.400
+ Yeah, look, I think we, you know,
+
+53:06.400 --> 53:09.480
+ we're entering this time where we've got,
+
+53:10.640 --> 53:13.360
+ we have technology that is progressing
+
+53:13.360 --> 53:15.800
+ at the fastest rate that it ever has.
+
+53:15.800 --> 53:20.800
+ And you've got, you get some really big social problems
+
+53:20.800 --> 53:25.800
+ like society scale problems that we have to tackle.
+
+53:26.320 --> 53:28.720
+ And so, you know, I think we're gonna rise to the challenge
+
+53:28.720 --> 53:30.560
+ and like figure out how to intersect
+
+53:30.560 --> 53:32.400
+ like all of the power of this technology
+
+53:32.400 --> 53:35.320
+ with all of the big challenges that are facing us,
+
+53:35.320 --> 53:37.840
+ whether it's, you know, global warming,
+
+53:37.840 --> 53:41.000
+ whether it's like the biggest remainder of the population
+
+53:41.000 --> 53:46.000
+ boom is in Africa for the next 50 years or so.
+
+53:46.800 --> 53:49.360
+ And like global warming is gonna make it increasingly
+
+53:49.360 --> 53:52.600
+ difficult to feed the global population in particular,
+
+53:52.600 --> 53:54.200
+ like in this place where you're gonna have
+
+53:54.200 --> 53:56.600
+ like the biggest population boom.
+
+53:57.720 --> 54:01.520
+ I think we, you know, like AI is gonna,
+
+54:01.520 --> 54:03.560
+ like if we push it in the right direction,
+
+54:03.560 --> 54:05.680
+ like it can do like incredible things
+
+54:05.680 --> 54:10.160
+ to empower all of us to achieve our full potential
+
+54:10.160 --> 54:15.160
+ and to, you know, like live better lives.
+
+54:15.160 --> 54:20.160
+ But like that also means focus on like
+
+54:20.520 --> 54:22.040
+ some super important things,
+
+54:22.040 --> 54:23.960
+ like how can you apply it to healthcare
+
+54:23.960 --> 54:28.960
+ to make sure that, you know, like our quality and cost of,
+
+54:29.640 --> 54:32.080
+ and sort of ubiquity of health coverage
+
+54:32.080 --> 54:35.080
+ is better and better over time.
+
+54:35.080 --> 54:37.960
+ Like that's more and more important every day
+
+54:37.960 --> 54:40.880
+ is like in the United States
+
+54:40.880 --> 54:43.280
+ and like the rest of the industrialized world.
+
+54:43.280 --> 54:45.720
+ So Western Europe, China, Japan, Korea,
+
+54:45.720 --> 54:48.880
+ like you've got this population bubble
+
+54:48.880 --> 54:52.880
+ of like aging working, you know, working age folks
+
+54:52.880 --> 54:56.200
+ who are, you know, at some point over the next 20, 30 years
+
+54:56.200 --> 54:58.000
+ they're gonna be largely retired
+
+54:58.000 --> 55:00.160
+ and like you're gonna have more retired people
+
+55:00.160 --> 55:01.200
+ than working age people.
+
+55:01.200 --> 55:02.520
+ And then like you've got, you know,
+
+55:02.520 --> 55:04.800
+ sort of natural questions about who's gonna take care
+
+55:04.800 --> 55:07.120
+ of all the old folks and who's gonna do all the work.
+
+55:07.120 --> 55:11.040
+ And the answers to like all of these sorts of questions
+
+55:11.040 --> 55:13.200
+ like where you're sort of running into, you know,
+
+55:13.200 --> 55:16.080
+ like constraints of the, you know,
+
+55:16.080 --> 55:20.080
+ the world and of society has always been like
+
+55:20.080 --> 55:23.000
+ what tech is gonna like help us get around this.
+
+55:23.000 --> 55:26.360
+ You know, like when I was a kid in the 70s and 80s,
+
+55:26.360 --> 55:29.800
+ like we talked all the time about like population boom,
+
+55:29.800 --> 55:32.200
+ population boom, like we're gonna,
+
+55:32.200 --> 55:34.360
+ like we're not gonna be able to like feed the planet.
+
+55:34.360 --> 55:36.800
+ And like we were like right in the middle
+
+55:36.800 --> 55:38.200
+ of the green revolution
+
+55:38.200 --> 55:43.200
+ where like this massive technology driven increase
+
+55:44.560 --> 55:47.520
+ and crop productivity like worldwide.
+
+55:47.520 --> 55:49.320
+ And like some of that was like taking some of the things
+
+55:49.320 --> 55:52.560
+ that we knew in the West and like getting them distributed
+
+55:52.560 --> 55:55.760
+ to the, you know, to the developing world.
+
+55:55.760 --> 55:59.360
+ And like part of it were things like, you know,
+
+55:59.360 --> 56:03.280
+ just smarter biology like helping us increase.
+
+56:03.280 --> 56:06.760
+ And like we don't talk about like, yeah,
+
+56:06.760 --> 56:10.320
+ overpopulation anymore because like we can more or less,
+
+56:10.320 --> 56:12.000
+ we sort of figured out how to feed the world.
+
+56:12.000 --> 56:14.760
+ Like that's a technology story.
+
+56:14.760 --> 56:19.480
+ And so like I'm super, super hopeful about the future
+
+56:19.480 --> 56:24.080
+ and in the ways where we will be able to apply technology
+
+56:24.080 --> 56:28.040
+ to solve some of these super challenging problems.
+
+56:28.040 --> 56:31.360
+ Like I've, like one of the things
+
+56:31.360 --> 56:34.680
+ that I'm trying to spend my time doing right now
+
+56:34.680 --> 56:36.600
+ is trying to get everybody else to be hopeful
+
+56:36.600 --> 56:38.720
+ as well because, you know, back to Harari,
+
+56:38.720 --> 56:41.160
+ like we are the stories that we tell.
+
+56:41.160 --> 56:44.320
+ Like if we, you know, if we get overly pessimistic right now
+
+56:44.320 --> 56:49.320
+ about like the potential future of technology, like we,
+
+56:49.320 --> 56:53.680
+ you know, like we may fail to fail to get all the things
+
+56:53.680 --> 56:56.880
+ in place that we need to like have our best possible future.
+
+56:56.880 --> 56:59.440
+ And that kind of hopeful optimism.
+
+56:59.440 --> 57:03.160
+ I'm glad that you have it because you're leading large groups
+
+57:03.160 --> 57:05.600
+ of engineers that are actually defining
+
+57:05.600 --> 57:06.720
+ that are writing that story,
+
+57:06.720 --> 57:08.320
+ that are helping build that future,
+
+57:08.320 --> 57:10.000
+ which is super exciting.
+
+57:10.000 --> 57:12.320
+ And I agree with everything you said,
+
+57:12.320 --> 57:14.840
+ except I do hope Clippy comes back.
+
+57:16.400 --> 57:17.760
+ We miss him.
+
+57:17.760 --> 57:19.360
+ I speak for the people.
+
+57:19.360 --> 57:21.800
+ So, Kellen, thank you so much for talking to me.
+
+57:21.800 --> 57:22.640
+ Thank you so much for having me.
+
+57:22.640 --> 57:43.640
+ It was a pleasure.
+

vtt/episode_032_small.vtt

@@ -0,0 +1,3392 @@
+WEBVTT
+
+00:00.000 --> 00:03.360
+ The following is a conversation with Paola Arlada.
+
+00:03.360 --> 00:06.400
+ She is a professor of stem cell and regenerative biology
+
+00:06.400 --> 00:09.760
+ at Harvard University and is interested in understanding
+
+00:09.760 --> 00:13.160
+ the molecular laws that govern the birth, differentiation,
+
+00:13.160 --> 00:16.600
+ and assembly of the human brain's cerebral cortex.
+
+00:16.600 --> 00:18.400
+ She explores the complexity of the brain
+
+00:18.400 --> 00:21.000
+ by studying and engineering elements
+
+00:21.000 --> 00:22.880
+ of how the brain develops.
+
+00:22.880 --> 00:25.640
+ This was a fascinating conversation to me.
+
+00:25.640 --> 00:28.560
+ It's part of the Artificial Intelligence podcast.
+
+00:28.560 --> 00:30.600
+ If you enjoy it, subscribe on YouTube,
+
+00:30.600 --> 00:32.360
+ give it five stars on iTunes.
+
+00:32.360 --> 00:35.760
+ Support on Patreon or simply connect with me on Twitter
+
+00:35.760 --> 00:39.960
+ at Lex Freedman, spelled F R I D M A N.
+
+00:39.960 --> 00:43.120
+ And I'd like to give a special thank you to Amy Jeffers
+
+00:43.120 --> 00:45.560
+ for her support of the podcast on Patreon.
+
+00:45.560 --> 00:47.680
+ She's an artist and you should definitely check out
+
+00:47.680 --> 00:52.680
+ her Instagram at LoveTruthGood, three beautiful words.
+
+00:52.720 --> 00:55.640
+ Your support means a lot and inspires me
+
+00:55.640 --> 00:57.720
+ to keep the series going.
+
+00:57.720 --> 01:01.720
+ And now here's my conversation with Paola Arlada.
+
+01:03.120 --> 01:05.320
+ You studied the development of the human brain
+
+01:05.320 --> 01:06.680
+ for many years.
+
+01:06.680 --> 01:10.520
+ So let me ask you an out of the box question first.
+
+01:11.480 --> 01:14.600
+ How likely is it that there's intelligent life out there
+
+01:14.600 --> 01:17.160
+ in the universe outside of earth
+
+01:17.160 --> 01:19.320
+ with something like the human brain?
+
+01:19.320 --> 01:20.760
+ So I can put it another way.
+
+01:20.760 --> 01:24.320
+ How unlikely is the human brain?
+
+01:24.320 --> 01:28.360
+ How difficult is it to build a thing
+
+01:28.360 --> 01:30.400
+ through the evolutionary process?
+
+01:30.400 --> 01:32.520
+ Well, it has happened here, right?
+
+01:32.520 --> 01:33.360
+ On this planet.
+
+01:33.360 --> 01:34.200
+ Once, yes.
+
+01:34.200 --> 01:35.040
+ Once.
+
+01:35.040 --> 01:39.960
+ So that simply tells you that it could, of course,
+
+01:39.960 --> 01:44.120
+ happen again, other places is only a matter of probability.
+
+01:44.120 --> 01:46.600
+ What the probability that you would get a brain
+
+01:46.600 --> 01:51.200
+ like the ones that we have, like the human brain.
+
+01:51.200 --> 01:54.000
+ So how difficult is it to make the human brain?
+
+01:54.000 --> 01:55.400
+ It's pretty difficult.
+
+01:56.480 --> 02:00.960
+ But most importantly, I guess we know very little
+
+02:00.960 --> 02:04.440
+ about how this process really happens.
+
+02:04.440 --> 02:06.320
+ And there is a reason for that,
+
+02:06.320 --> 02:09.160
+ actually multiple reasons for that.
+
+02:09.160 --> 02:13.400
+ Most of what we know about how the mammalian brains
+
+02:13.400 --> 02:15.680
+ or the brain of mammals develop,
+
+02:15.680 --> 02:18.960
+ comes from studying in labs other brains,
+
+02:18.960 --> 02:22.560
+ not our own brain, the brain of mice, for example.
+
+02:22.560 --> 02:25.400
+ But if I showed you a picture of a mouse brain
+
+02:25.400 --> 02:28.400
+ and then you put it next to a picture of a human brain,
+
+02:28.400 --> 02:31.160
+ they don't look at all like each other.
+
+02:31.160 --> 02:33.040
+ So they're very different.
+
+02:33.040 --> 02:36.240
+ And therefore, there is a limit to what you can learn
+
+02:36.240 --> 02:40.560
+ about how the human brain is made by studying the mouse brain.
+
+02:40.560 --> 02:43.320
+ There is a huge value in studying the mouse brain.
+
+02:43.320 --> 02:45.080
+ There are many things that we have learned,
+
+02:45.080 --> 02:46.600
+ but it's not the same thing.
+
+02:46.600 --> 02:49.200
+ So in having studied the human brain
+
+02:49.200 --> 02:51.440
+ or through the mouse and through other methodologies
+
+02:51.440 --> 02:54.760
+ that we'll talk about, do you have a sense?
+
+02:54.760 --> 02:57.560
+ I mean, you're one of the experts in the world.
+
+02:57.560 --> 03:01.160
+ How much do you feel you know about the brain?
+
+03:01.160 --> 03:05.920
+ And how often do you find yourself in awe
+
+03:05.920 --> 03:07.840
+ of this mysterious thing?
+
+03:07.840 --> 03:12.200
+ Yeah, you pretty much find yourself in awe all the time.
+
+03:12.200 --> 03:15.360
+ It's an amazing process.
+
+03:15.360 --> 03:17.960
+ It's a process by which,
+
+03:17.960 --> 03:20.720
+ by means that we don't fully understand
+
+03:20.720 --> 03:23.640
+ at the very beginning of embryogenesis,
+
+03:23.640 --> 03:28.800
+ the structure called the neural tube literally self assembles.
+
+03:28.800 --> 03:30.440
+ And it happens in an embryo
+
+03:30.440 --> 03:33.720
+ and it can happen also from stem cells in a dish.
+
+03:33.720 --> 03:34.960
+ Okay.
+
+03:34.960 --> 03:38.720
+ And then from there, these stem cells that are present
+
+03:38.720 --> 03:41.640
+ within the neural tube give rise to all of the thousands
+
+03:41.640 --> 03:43.440
+ and thousands of different cell types
+
+03:43.440 --> 03:46.560
+ that are present in the brain through time, right?
+
+03:46.560 --> 03:51.760
+ With the interesting, very intriguing, interesting observation
+
+03:51.760 --> 03:56.280
+ is that the time that it takes for the human brain to be made,
+
+03:56.280 --> 04:01.480
+ it's human time, meaning that for me and you,
+
+04:01.480 --> 04:04.560
+ it took almost nine months of gestation to build the brain
+
+04:04.560 --> 04:08.080
+ and then another 20 years of learning postnatally
+
+04:08.080 --> 04:09.360
+ to get the brain that we have today
+
+04:09.360 --> 04:11.360
+ that allows us to this conversation.
+
+04:11.360 --> 04:14.960
+ A mouse takes 20 days or so
+
+04:14.960 --> 04:22.200
+ for an embryo to be born and so the brain is built
+
+04:22.200 --> 04:24.840
+ in a much shorter period of time and the beauty of it
+
+04:24.840 --> 04:27.160
+ is that if you take mouse stem cells
+
+04:27.160 --> 04:29.880
+ and you put them in a cultured dish,
+
+04:29.880 --> 04:34.840
+ the brain organoid that you get from a mouse is formed faster
+
+04:34.840 --> 04:39.120
+ that if you took human stem cells and put them in the dish
+
+04:39.120 --> 04:42.120
+ and let them make a human brain organoid.
+
+04:42.120 --> 04:45.560
+ So the very developmental process is...
+
+04:45.560 --> 04:49.000
+ Controlled by the speed of the species.
+
+04:49.000 --> 04:54.080
+ Which means it's by its own purpose, it's not accidental
+
+04:54.080 --> 04:59.800
+ or there is something in that temporal dynamic to that development.
+
+04:59.800 --> 05:04.000
+ Exactly, that is very important for us to get the brain we have
+
+05:04.000 --> 05:08.120
+ and we can speculate for why that is.
+
+05:08.120 --> 05:13.000
+ It takes us a long time as human beings after we're born
+
+05:13.000 --> 05:16.040
+ to learn all the things that we have to learn
+
+05:16.040 --> 05:17.920
+ to have the adult brain.
+
+05:17.920 --> 05:20.160
+ It's actually 20 years, think about it.
+
+05:20.160 --> 05:23.440
+ From when a baby is born to when a teenager
+
+05:23.440 --> 05:27.160
+ goes through puberty to adults, it's a long time.
+
+05:27.160 --> 05:32.320
+ Do you think you can maybe talk through the first few months
+
+05:32.320 --> 05:35.080
+ and then on to the first 20 years
+
+05:35.080 --> 05:37.640
+ and then for the rest of their lives?
+
+05:37.640 --> 05:41.080
+ What does the development of the human brain look like?
+
+05:41.080 --> 05:43.360
+ What are the different stages?
+
+05:43.360 --> 05:46.720
+ At the beginning you have to build a brain, right?
+
+05:46.720 --> 05:48.840
+ And the brain is made of cells.
+
+05:48.840 --> 05:51.960
+ What's the very beginning? Which beginning are we talking about?
+
+05:51.960 --> 05:53.280
+ In the embryo.
+
+05:53.280 --> 05:56.000
+ As the embryo is developing in the womb,
+
+05:56.000 --> 05:59.680
+ in addition to making all of the other tissues of the embryo,
+
+05:59.680 --> 06:02.040
+ the muscle, the heart, the blood,
+
+06:02.040 --> 06:04.960
+ the embryo is also building the brain.
+
+06:04.960 --> 06:08.480
+ And it builds from a very simple structure
+
+06:08.480 --> 06:10.000
+ called the neural tube,
+
+06:10.000 --> 06:13.160
+ which is basically nothing but a tube of cells
+
+06:13.160 --> 06:15.800
+ that spans sort of the length of the embryo
+
+06:15.800 --> 06:20.840
+ from the head all the way to the tail, let's say, of the embryo.
+
+06:20.840 --> 06:23.360
+ And then over in human beings,
+
+06:23.360 --> 06:25.520
+ over many months of gestation,
+
+06:25.520 --> 06:28.240
+ from that neural tube,
+
+06:28.240 --> 06:32.480
+ which contains a stem cell like cells of the brain,
+
+06:32.480 --> 06:37.000
+ you will make many, many other building blocks of the brain.
+
+06:37.000 --> 06:39.600
+ So all of the other cell types,
+
+06:39.600 --> 06:43.080
+ because there are many, many different types of cells in the brain,
+
+06:43.080 --> 06:46.640
+ that will form specific structures of the brain.
+
+06:46.640 --> 06:49.800
+ So you can think about embryonic development of the brain
+
+06:49.800 --> 06:54.440
+ as just the time in which you are making the building blocks, the cells.
+
+06:54.440 --> 06:56.840
+ Are the stem cells relatively homogeneous,
+
+06:56.840 --> 06:59.240
+ like uniform, or are they all different types?
+
+06:59.240 --> 07:01.280
+ It's a very good question. It's exactly how it works.
+
+07:01.280 --> 07:05.200
+ You start with a more homogeneous,
+
+07:05.200 --> 07:09.120
+ perhaps more multipotent type of stem cell.
+
+07:09.120 --> 07:13.520
+ That multipotent means that it has the potential
+
+07:13.520 --> 07:17.360
+ to make many, many different types of other cells.
+
+07:17.360 --> 07:21.440
+ And then with time, these progenitors become more heterogeneous,
+
+07:21.440 --> 07:22.840
+ which means more diverse.
+
+07:22.840 --> 07:26.200
+ There are going to be many different types of these stem cells.
+
+07:26.200 --> 07:28.760
+ And also they will give rise to progeny,
+
+07:28.760 --> 07:31.760
+ to other cells that are not stem cells,
+
+07:31.760 --> 07:33.320
+ that are specific cells of the brain,
+
+07:33.320 --> 07:36.080
+ that are very different from the mother stem cell.
+
+07:36.080 --> 07:40.360
+ And now you think about this process of making cells from the stem cells
+
+07:40.360 --> 07:43.760
+ over many, many months of development for humans.
+
+07:43.760 --> 07:48.200
+ And what you're doing here, building the cells that physically make the brain,
+
+07:48.200 --> 07:52.400
+ and then you arrange them in specific structures
+
+07:52.400 --> 07:55.440
+ that are present in the final brain.
+
+07:55.440 --> 07:59.280
+ So you can think about the embryonic development of the brain
+
+07:59.280 --> 08:02.120
+ as the time where you're building the bricks.
+
+08:02.120 --> 08:05.520
+ You're putting the bricks together to form buildings,
+
+08:05.520 --> 08:08.120
+ structures, regions of the brain,
+
+08:08.120 --> 08:13.040
+ and where you make the connections between these many different types of cells,
+
+08:13.040 --> 08:15.200
+ especially nerve cells, neurons, right,
+
+08:15.200 --> 08:19.240
+ that transmit action potentials and electricity.
+
+08:19.240 --> 08:22.280
+ I've heard you also say somewhere, I think, correct me if I'm wrong,
+
+08:22.280 --> 08:25.080
+ that the order of the way this builds matters.
+
+08:25.080 --> 08:26.040
+ Oh, yes.
+
+08:26.040 --> 08:29.880
+ If you are an engineer and you think about development,
+
+08:29.880 --> 08:34.960
+ you can think of it as, well, I could also take all the cells
+
+08:34.960 --> 08:37.960
+ and bring them all together into a brain in the end.
+
+08:37.960 --> 08:40.320
+ But development is much more than that.
+
+08:40.320 --> 08:43.880
+ So the cells are made in a very specific order
+
+08:43.880 --> 08:47.400
+ that subserve the final product that you need to get.
+
+08:47.400 --> 08:52.200
+ And so, for example, all of the nerve cells, the neurons, are made first.
+
+08:52.200 --> 08:56.760
+ And all of the supportive cells of the neurons, like the glia, is made later.
+
+08:56.760 --> 09:02.080
+ And there is a reason for that because they have to assemble together in specific ways.
+
+09:02.080 --> 09:05.720
+ But you also may say, well, why don't we just put them all together in the end?
+
+09:05.720 --> 09:08.920
+ It's because as they develop next to each other,
+
+09:08.920 --> 09:11.280
+ they influence their own development.
+
+09:11.280 --> 09:15.400
+ So it's a different thing for a glia to be made alone in a dish
+
+09:15.400 --> 09:19.360
+ than a glia cell be made in a developing embryo
+
+09:19.360 --> 09:23.680
+ with all these other cells around it that produce all these other signals.
+
+09:23.680 --> 09:27.840
+ First of all, that's mind blowing, that this development process.
+
+09:27.840 --> 09:29.760
+ From my perspective in artificial intelligence,
+
+09:29.760 --> 09:33.480
+ you often think of how incredible the final product is,
+
+09:33.480 --> 09:35.200
+ the final product, the brain.
+
+09:35.200 --> 09:40.400
+ But you just, you're making me realize that the final product is just,
+
+09:40.400 --> 09:44.480
+ is the beautiful thing is the actual development process.
+
+09:44.480 --> 09:51.640
+ Do we know the code that drives that development?
+
+09:51.640 --> 09:53.640
+ Do we have any sense?
+
+09:53.640 --> 09:59.320
+ First of all, thank you for saying that it's really the formation of the brain.
+
+09:59.320 --> 10:05.120
+ It's really its development, this incredibly choreographed dance
+
+10:05.120 --> 10:10.560
+ that happens the same way every time each one of us builds the brain, right?
+
+10:10.560 --> 10:14.880
+ And that builds an organ that allows us to do what we're doing today, right?
+
+10:14.880 --> 10:16.360
+ That is mind blowing.
+
+10:16.360 --> 10:21.880
+ And this is why developmental neurobiologists never get tired of studying that.
+
+10:21.880 --> 10:23.800
+ Now, you're asking about the code.
+
+10:23.800 --> 10:26.520
+ What drives this? How is this done?
+
+10:26.520 --> 10:29.720
+ Well, it's millions of years of evolution
+
+10:29.720 --> 10:33.120
+ of really fine tuning gene expression programs
+
+10:33.120 --> 10:37.480
+ that allow certain cells to be made at a certain time
+
+10:37.480 --> 10:41.680
+ and to become a certain cell type,
+
+10:41.680 --> 10:47.360
+ but also mechanical forces of pressure bending.
+
+10:47.360 --> 10:50.280
+ This embryo is not just, it will not stay a tube,
+
+10:50.280 --> 10:52.000
+ this brain for very long.
+
+10:52.000 --> 10:55.720
+ At some point, this tube in the front of the embryo will expand
+
+10:55.720 --> 10:58.160
+ to make the primordium of the brain, right?
+
+10:58.160 --> 11:02.880
+ Now, the forces that control the cells feel,
+
+11:02.880 --> 11:04.840
+ and this is another beautiful thing,
+
+11:04.840 --> 11:09.240
+ the very force that they feel, which is different from a week before,
+
+11:09.240 --> 11:11.080
+ a week ago, will tell the cell,
+
+11:11.080 --> 11:13.640
+ oh, you're being squished in a certain way,
+
+11:13.640 --> 11:16.560
+ begin to produce these new genes,
+
+11:16.560 --> 11:18.400
+ because now you are at the corner,
+
+11:18.400 --> 11:23.160
+ or you are in a stretch of cells or whatever it is.
+
+11:23.160 --> 11:26.000
+ And so that mechanical physical force
+
+11:26.000 --> 11:29.440
+ shapes the fate of the cell as well.
+
+11:29.440 --> 11:31.880
+ So it's not only chemical, it's also mechanical.
+
+11:31.880 --> 11:38.360
+ So from my perspective, biology is this incredibly complex mess,
+
+11:38.360 --> 11:40.160
+ gooey mess.
+
+11:40.160 --> 11:43.440
+ So you're seeing mechanical forces.
+
+11:43.440 --> 11:50.000
+ How different is a computer or any kind of mechanical machine
+
+11:50.000 --> 11:53.840
+ that humans build and the biological systems?
+
+11:53.840 --> 11:57.000
+ Have you been, because you've worked a lot with biological systems,
+
+11:57.000 --> 11:59.960
+ are they as much of a mess as it seems
+
+11:59.960 --> 12:03.520
+ from a perspective of an engineer, a mechanical engineer?
+
+12:03.520 --> 12:11.680
+ Yeah, they are much more prone to taking alternative routes, right?
+
+12:11.680 --> 12:18.160
+ So if you, we go back to printing a brain versus developing a brain,
+
+12:18.160 --> 12:20.440
+ of course, if you print a brain,
+
+12:20.440 --> 12:23.960
+ given that you start with the same building blocks, the same cells,
+
+12:23.960 --> 12:28.600
+ you could potentially print it the same way every time.
+
+12:28.600 --> 12:32.520
+ But that final brain may not work the same way
+
+12:32.520 --> 12:34.440
+ as a brain built during development does,
+
+12:34.440 --> 12:38.680
+ because the very same building blocks that you're using
+
+12:38.680 --> 12:41.440
+ developed in a completely different environment, right?
+
+12:41.440 --> 12:43.000
+ That was not the environment of the brain.
+
+12:43.000 --> 12:47.120
+ Therefore, they're going to be different just by definition.
+
+12:47.120 --> 12:51.840
+ So if you instead use development to build, let's say, a brain
+
+12:51.840 --> 12:55.840
+ organoid, which maybe we will be talking about in a few minutes.
+
+12:55.840 --> 12:57.000
+ Those things are fascinating.
+
+12:57.000 --> 13:01.960
+ Yes, so if you use processes of development,
+
+13:01.960 --> 13:06.480
+ then when you watch it, you can see that sometimes things can go wrong
+
+13:06.480 --> 13:07.520
+ in some organoids.
+
+13:07.520 --> 13:10.880
+ And by wrong, I mean different one organoid from the next.
+
+13:10.880 --> 13:14.840
+ While if you think about that embryo, it always goes right.
+
+13:14.840 --> 13:18.960
+ So it's this development, it's for as complex as it is.
+
+13:18.960 --> 13:23.680
+ Every time a baby is born has, you know, with very few exceptions,
+
+13:23.680 --> 13:26.200
+ the brain is like the next baby.
+
+13:26.200 --> 13:31.320
+ But it's not the same if you develop it in a dish.
+
+13:31.320 --> 13:33.840
+ And first of all, we don't even develop a brain.
+
+13:33.840 --> 13:36.080
+ You develop something much simpler in the dish.
+
+13:36.080 --> 13:39.680
+ But there are more options for building things differently,
+
+13:39.680 --> 13:46.080
+ which really tells you that evolution has played a really
+
+13:46.080 --> 13:53.280
+ tight game here for how in the end the brain is built in vivo.
+
+13:53.280 --> 13:55.360
+ So just a quick maybe dumb question,
+
+13:55.360 --> 14:01.120
+ but it seems like the building process is not a dictatorship.
+
+14:01.120 --> 14:06.680
+ It seems like there's not a centralized high level mechanism
+
+14:06.680 --> 14:10.320
+ that says, OK, this cell built itself the wrong way.
+
+14:10.320 --> 14:11.560
+ I'm going to kill it.
+
+14:11.560 --> 14:15.480
+ It seems like there's a really strong distributed mechanism.
+
+14:15.480 --> 14:18.440
+ Is that in your sense for what you have?
+
+14:18.440 --> 14:20.920
+ There are a lot of possibilities, right?
+
+14:20.920 --> 14:25.080
+ And if you think about, for example, different species,
+
+14:25.080 --> 14:28.920
+ building their brain, each brain is a little bit different.
+
+14:28.920 --> 14:31.360
+ So the brain of a lizard is very different from that
+
+14:31.360 --> 14:36.560
+ of a chicken, from that of one of us, and so on and so forth.
+
+14:36.560 --> 14:40.960
+ And still is a brain, but it was built differently.
+
+14:40.960 --> 14:44.120
+ Starting from stem cells, they pretty much
+
+14:44.120 --> 14:46.040
+ had the same potential.
+
+14:46.040 --> 14:49.400
+ But in the end, evolution builds different brains
+
+14:49.400 --> 14:51.520
+ in different species, because that
+
+14:51.520 --> 14:54.040
+ serves in a way the purpose of that species
+
+14:54.040 --> 14:56.680
+ and the well being of that organism.
+
+14:56.680 --> 15:00.720
+ And so there are many possibilities,
+
+15:00.720 --> 15:04.880
+ but then there is a way, and you were talking about a code.
+
+15:04.880 --> 15:07.560
+ Nobody knows what the entire code of development is.
+
+15:07.560 --> 15:08.680
+ Of course, we don't.
+
+15:08.680 --> 15:13.400
+ We know bits and pieces of very specific aspects
+
+15:13.400 --> 15:15.680
+ of development of the brain, what genes are involved
+
+15:15.680 --> 15:18.520
+ to make a certain cell types, how those two cells interact
+
+15:18.520 --> 15:21.480
+ to make the next level structure that we might know,
+
+15:21.480 --> 15:24.560
+ but the entirety of it, how it's so well controlled.
+
+15:24.560 --> 15:26.160
+ It's really mind blowing.
+
+15:26.160 --> 15:29.120
+ So in the first two months in the embryo,
+
+15:29.120 --> 15:32.720
+ or whatever, the first few weeks, few months.
+
+15:32.720 --> 15:37.120
+ So yeah, the building blocks are constructed,
+
+15:37.120 --> 15:40.040
+ the actual, the different regions of the brain,
+
+15:40.040 --> 15:42.760
+ I guess, in the nervous system.
+
+15:42.760 --> 15:46.520
+ Well, this continues way longer than just the first few months.
+
+15:46.520 --> 15:50.480
+ So over the very first few months,
+
+15:50.480 --> 15:52.080
+ you build a lot of these cells,
+
+15:52.080 --> 15:56.800
+ but then there is continuous building of new cell types
+
+15:56.800 --> 15:58.480
+ all the way through birth.
+
+15:58.480 --> 16:00.400
+ And then even postnatally,
+
+16:01.520 --> 16:04.000
+ I don't know if you've ever heard of myelin.
+
+16:04.000 --> 16:06.720
+ Myelin is this sort of insulation
+
+16:06.720 --> 16:09.800
+ that is built around the cables of the neurons
+
+16:09.800 --> 16:12.200
+ so that the electricity can go really fast from.
+
+16:12.200 --> 16:13.520
+ The axons, I guess they're called.
+
+16:13.520 --> 16:15.720
+ The axons are called axons, exactly.
+
+16:15.720 --> 16:20.720
+ And so as human beings, we myelinate ourselves
+
+16:22.920 --> 16:27.000
+ postnatally, a kid, a six year old kid,
+
+16:27.000 --> 16:29.720
+ as barely started the process of making
+
+16:29.720 --> 16:31.640
+ the mature oligodendrocytes,
+
+16:31.640 --> 16:33.400
+ which are the cells that then eventually
+
+16:33.400 --> 16:36.360
+ will wrap the axons into myelin.
+
+16:36.360 --> 16:38.720
+ And this will continue, believe it or not,
+
+16:38.720 --> 16:42.200
+ until we are about 25, 30 years old.
+
+16:42.200 --> 16:45.080
+ So there is a continuous process of maturation
+
+16:45.080 --> 16:46.600
+ and tweaking and additions,
+
+16:46.600 --> 16:51.040
+ and also in response to what we do.
+
+16:51.040 --> 16:53.960
+ I remember taking api biology in high school,
+
+16:53.960 --> 16:57.040
+ and in the textbook, it said that,
+
+16:57.040 --> 16:58.560
+ I'm going by memory here,
+
+16:58.560 --> 17:02.000
+ that scientists disagree on the purpose of myelin
+
+17:03.040 --> 17:04.720
+ in the brain.
+
+17:04.720 --> 17:06.400
+ Is that totally wrong?
+
+17:06.400 --> 17:10.000
+ So like, I guess it speeds up the,
+
+17:12.000 --> 17:13.200
+ okay, but I'd be wrong here,
+
+17:13.200 --> 17:15.280
+ but I guess it speeds up the electricity traveling
+
+17:15.280 --> 17:17.680
+ down the axon or something.
+
+17:17.680 --> 17:20.160
+ So that's the most sort of canonical,
+
+17:20.160 --> 17:21.760
+ and definitely that's the case.
+
+17:21.760 --> 17:24.880
+ So you have to imagine an axon,
+
+17:24.880 --> 17:27.680
+ and you can think about it as a cable or some type
+
+17:27.680 --> 17:29.520
+ with electricity going through.
+
+17:29.520 --> 17:34.400
+ And what myelin does by insulating the outside,
+
+17:34.400 --> 17:36.360
+ I should say there are tracts of myelin
+
+17:36.360 --> 17:39.600
+ and pieces of axons that are naked without myelin.
+
+17:39.600 --> 17:41.760
+ And so by having the insulation,
+
+17:41.760 --> 17:44.040
+ the electricity instead of going straight through the cable,
+
+17:44.040 --> 17:47.240
+ it will jump over a piece of myelin, right?
+
+17:47.240 --> 17:49.960
+ To the next naked little piece and jump again,
+
+17:49.960 --> 17:52.720
+ and therefore, that's the idea that you go faster.
+
+17:53.920 --> 17:58.720
+ And it was always thought that in order to build
+
+17:58.720 --> 18:01.840
+ a big brain, a big nervous system,
+
+18:01.840 --> 18:04.160
+ in order to have a nervous system
+
+18:04.160 --> 18:06.440
+ that can do very complex type of things,
+
+18:06.440 --> 18:09.400
+ then you need a lot of myelin because you wanna go fast
+
+18:09.400 --> 18:13.320
+ with this information from point A to point B.
+
+18:13.320 --> 18:17.960
+ Well, a few years ago, maybe five years ago or so,
+
+18:17.960 --> 18:20.680
+ we discovered that some of the most evolved,
+
+18:20.680 --> 18:24.120
+ which means the newest type of neurons that we have
+
+18:24.120 --> 18:26.520
+ as non human primates, as as human beings,
+
+18:26.520 --> 18:29.120
+ in the top of our cerebral cortex,
+
+18:29.120 --> 18:30.920
+ which should be the neurons that do some
+
+18:30.920 --> 18:33.200
+ of the most complex things that we do.
+
+18:33.200 --> 18:37.080
+ Well, those have axons that have very little myelin.
+
+18:37.080 --> 18:42.080
+ Wow. And they have very interesting ways
+
+18:42.080 --> 18:44.400
+ in which they put this myelin on their axons,
+
+18:44.400 --> 18:46.400
+ you know, a little piece here, then a long track
+
+18:46.400 --> 18:48.680
+ with no myelin, another chunk there,
+
+18:48.680 --> 18:50.600
+ and some don't have myelin at all.
+
+18:50.600 --> 18:53.120
+ So now you have to explain
+
+18:54.760 --> 18:57.960
+ where we're going with evolution.
+
+18:57.960 --> 19:01.360
+ And if you think about it, perhaps as an electrical engineer,
+
+19:02.800 --> 19:06.000
+ when I looked at it, I initially thought,
+
+19:06.000 --> 19:07.560
+ I'm a developmental neurobiology,
+
+19:07.560 --> 19:10.880
+ I thought maybe this is what we see now,
+
+19:10.880 --> 19:14.160
+ but if we give evolution another few million years,
+
+19:14.160 --> 19:16.520
+ we'll see a lot of myelin on these neurons too.
+
+19:16.520 --> 19:18.840
+ But I actually think now that that's instead
+
+19:18.840 --> 19:22.000
+ the future of the brain, less myelin,
+
+19:22.000 --> 19:24.720
+ and my allow for more flexibility
+
+19:24.720 --> 19:26.720
+ on what you do with your axons,
+
+19:26.720 --> 19:28.560
+ and therefore more complicated
+
+19:28.560 --> 19:32.200
+ and unpredictable type of functions,
+
+19:32.200 --> 19:34.320
+ which is also a bit mind blowing.
+
+19:34.320 --> 19:38.480
+ So it seems like it's controlling the timing of the signal.
+
+19:38.480 --> 19:40.160
+ So they're in the timing,
+
+19:40.160 --> 19:43.320
+ you can encode a lot of information.
+
+19:43.320 --> 19:44.680
+ And so the brain...
+
+19:44.680 --> 19:48.600
+ The timing, the chemistry of that little piece of axon,
+
+19:48.600 --> 19:52.160
+ perhaps it's a dynamic process where the myelin can move.
+
+19:52.160 --> 19:57.160
+ Now you see how many layers of variability you can add,
+
+19:57.520 --> 19:58.960
+ and that's actually really good.
+
+19:58.960 --> 20:02.320
+ If you're trying to come up with a new function
+
+20:02.320 --> 20:06.600
+ or a new capability or something unpredictable in a way.
+
+20:06.600 --> 20:08.240
+ So we're gonna jump right out a little bit,
+
+20:08.240 --> 20:12.880
+ but the old question of how much is nature
+
+20:12.880 --> 20:14.560
+ and how much is nurture,
+
+20:14.560 --> 20:17.360
+ in terms of this incredible thing
+
+20:17.360 --> 20:18.920
+ after the development is over,
+
+20:20.280 --> 20:25.160
+ we seem to be kind of somewhat smart, intelligent,
+
+20:26.160 --> 20:27.600
+ cognition, consciousness,
+
+20:27.600 --> 20:30.680
+ all these things are just incredible ability of reason
+
+20:30.680 --> 20:32.080
+ and so on emerge.
+
+20:32.080 --> 20:34.960
+ In your sense, how much is in the hardware,
+
+20:34.960 --> 20:39.320
+ in the nature and how much is in the nurtures
+
+20:39.320 --> 20:41.040
+ learned through with our parents
+
+20:41.040 --> 20:42.480
+ through interacting with the environment, so on.
+
+20:42.480 --> 20:43.800
+ It's really both, right?
+
+20:43.800 --> 20:45.040
+ If you think about it.
+
+20:45.040 --> 20:48.040
+ So we are born with a brain as babies
+
+20:48.040 --> 20:53.040
+ that has most of its cells and most of its structures
+
+20:53.640 --> 20:57.920
+ and that will take a few years to grow,
+
+20:57.920 --> 21:00.640
+ to add more, to be better.
+
+21:00.640 --> 21:04.160
+ But really then we have this 20 years
+
+21:04.160 --> 21:07.080
+ of interacting with the environment around us.
+
+21:07.080 --> 21:10.800
+ And so what that brain that was so perfectly built
+
+21:10.800 --> 21:15.800
+ or imperfectly built due to our genetic cues
+
+21:16.480 --> 21:20.200
+ will then be used to incorporate the environment
+
+21:20.200 --> 21:22.760
+ in its farther maturation and development.
+
+21:22.760 --> 21:27.000
+ And so your experiences do shape your brain.
+
+21:27.000 --> 21:29.480
+ I mean, we know that like if you and I
+
+21:29.480 --> 21:33.000
+ may have had a different childhood or a different,
+
+21:33.000 --> 21:35.080
+ we have been going to different schools,
+
+21:35.080 --> 21:36.480
+ we have been learning different things
+
+21:36.480 --> 21:38.080
+ and our brain is a little bit different
+
+21:38.080 --> 21:41.200
+ because of that we behave differently because of that.
+
+21:41.200 --> 21:44.080
+ And so especially postnatally,
+
+21:44.080 --> 21:46.040
+ experience is extremely important.
+
+21:46.040 --> 21:48.800
+ We are born with a plastic brain.
+
+21:48.800 --> 21:51.480
+ What that means is a brain that is able to change
+
+21:51.480 --> 21:54.280
+ in response to stimuli.
+
+21:54.280 --> 21:56.400
+ They can be sensory.
+
+21:56.400 --> 22:01.040
+ So perhaps some of the most illuminating studies
+
+22:01.040 --> 22:03.440
+ that were done were studies in which
+
+22:03.440 --> 22:07.000
+ the sensory organs were not working, right?
+
+22:07.000 --> 22:09.520
+ If you are born with eyes that don't work,
+
+22:09.520 --> 22:12.520
+ then your very brain, the piece of the brain
+
+22:12.520 --> 22:16.000
+ that normally would process vision, the visual cortex
+
+22:17.240 --> 22:19.840
+ develops postnatally differently
+
+22:19.840 --> 22:23.520
+ and it might be used to do something different, right?
+
+22:23.520 --> 22:25.600
+ So that's the most extreme.
+
+22:25.600 --> 22:27.480
+ The plasticity of the brain, I guess,
+
+22:27.480 --> 22:29.480
+ is the magic hardware that it,
+
+22:29.480 --> 22:32.960
+ and then its flexibility in all forms
+
+22:32.960 --> 22:36.320
+ is what enables the learning postnatally.
+
+22:36.320 --> 22:39.280
+ Can you talk about organoids?
+
+22:39.280 --> 22:40.920
+ What are they?
+
+22:40.920 --> 22:43.760
+ And how can you use them to help us understand
+
+22:43.760 --> 22:45.760
+ the brain and the development of the brain?
+
+22:45.760 --> 22:47.360
+ This is very, very important.
+
+22:47.360 --> 22:49.920
+ So the first thing I'd like to say,
+
+22:49.920 --> 22:51.440
+ please keep this in the video.
+
+22:51.440 --> 22:56.040
+ The first thing I'd like to say is that an organoid,
+
+22:56.040 --> 23:01.040
+ a brain organoid, is not the same as a brain, okay?
+
+23:01.600 --> 23:03.640
+ It's a fundamental distinction.
+
+23:03.640 --> 23:08.560
+ It's a system, a cellular system,
+
+23:08.560 --> 23:12.200
+ that one can develop in the culture dish
+
+23:12.200 --> 23:17.200
+ starting from stem cells that will mimic some aspects
+
+23:17.200 --> 23:21.400
+ of the development of the brain, but not all of it.
+
+23:21.400 --> 23:23.760
+ They are very small, maximum,
+
+23:23.760 --> 23:27.920
+ they become about four to five millimeters in diameters.
+
+23:27.920 --> 23:32.920
+ They are much simpler than our brain, of course,
+
+23:33.400 --> 23:36.480
+ but yet they are the only system
+
+23:36.480 --> 23:39.520
+ where we can literally watch a process
+
+23:39.520 --> 23:42.560
+ of human brain development unfold.
+
+23:42.560 --> 23:45.080
+ And by watch, I mean study it.
+
+23:45.080 --> 23:48.000
+ Remember when I told you that we can't understand
+
+23:48.000 --> 23:50.040
+ everything about development in our own brain
+
+23:50.040 --> 23:51.560
+ by studying a mouse?
+
+23:51.560 --> 23:53.600
+ Well, we can't study the actual process
+
+23:53.600 --> 23:54.840
+ of development of the human brain
+
+23:54.840 --> 23:56.320
+ because it all happens in utero.
+
+23:56.320 --> 23:59.320
+ So we will never have access to that process ever.
+
+24:00.400 --> 24:04.320
+ And therefore, this is our next best thing,
+
+24:04.320 --> 24:08.400
+ like a bunch of stem cells that can be coaxed
+
+24:08.400 --> 24:11.720
+ into starting a process of neural tube formation.
+
+24:11.720 --> 24:14.680
+ Remember that tube that is made by the embryo rion?
+
+24:14.680 --> 24:17.160
+ And from there, a lot of the cell types
+
+24:17.160 --> 24:20.680
+ that are present within the brain
+
+24:20.680 --> 24:24.960
+ and you can simply watch it and study,
+
+24:24.960 --> 24:28.680
+ but you can also think about diseases
+
+24:28.680 --> 24:30.880
+ where development of the brain
+
+24:30.880 --> 24:34.200
+ does not proceed normally, right, properly.
+
+24:34.200 --> 24:35.920
+ Think about neurodevelopmental diseases
+
+24:35.920 --> 24:38.280
+ that are many, many different types.
+
+24:38.280 --> 24:40.200
+ Think about autism spectrum disorders,
+
+24:40.200 --> 24:42.640
+ there are also many different types of autism.
+
+24:42.640 --> 24:45.320
+ So there you could take a stem cell
+
+24:45.320 --> 24:47.520
+ which really means either a sample of blood
+
+24:47.520 --> 24:50.960
+ or a sample of skin from the patient,
+
+24:50.960 --> 24:54.360
+ make a stem cell, and then with that stem cell,
+
+24:54.360 --> 24:57.480
+ watch a process of formation of a brain organoid
+
+24:57.480 --> 25:00.640
+ of that person, with that genetics,
+
+25:00.640 --> 25:02.240
+ with that genetic code in it.
+
+25:02.240 --> 25:05.840
+ And you can ask, what is this genetic code doing
+
+25:05.840 --> 25:08.800
+ to some aspects of development of the brain?
+
+25:08.800 --> 25:12.040
+ And for the first time, you may come to solutions
+
+25:12.040 --> 25:15.440
+ like, what cells are involved in autism?
+
+25:15.440 --> 25:17.400
+ So I have so many questions around this.
+
+25:17.400 --> 25:20.560
+ So if you take this human stem cell
+
+25:20.560 --> 25:23.400
+ for that particular person with that genetic code,
+
+25:23.400 --> 25:26.560
+ how, and you try to build an organoid,
+
+25:26.560 --> 25:28.880
+ how often will it look similar?
+
+25:28.880 --> 25:31.880
+ What's the, yeah, so.
+
+25:31.880 --> 25:33.320
+ Reproducibility.
+
+25:33.320 --> 25:37.360
+ Yes, or how much variability is the flip side of that, yeah.
+
+25:37.360 --> 25:42.360
+ So there is much more variability in building organoids
+
+25:42.360 --> 25:44.560
+ than there is in building brain.
+
+25:44.560 --> 25:47.320
+ It's really true that the majority of us,
+
+25:47.320 --> 25:49.600
+ when we are born as babies,
+
+25:49.600 --> 25:52.480
+ our brains look a lot like each other.
+
+25:52.480 --> 25:54.920
+ This is the magic that the embryo does,
+
+25:54.920 --> 25:57.680
+ where it builds a brain in the context of a body
+
+25:57.680 --> 26:01.280
+ and there is very little variability there.
+
+26:01.280 --> 26:02.320
+ There is disease, of course,
+
+26:02.320 --> 26:04.000
+ but in general, little variability.
+
+26:04.000 --> 26:08.400
+ When you build an organoid, we don't have the full code
+
+26:08.400 --> 26:09.520
+ for how this is done.
+
+26:09.520 --> 26:13.440
+ And so in part, the organoid somewhat builds itself
+
+26:13.440 --> 26:15.560
+ because there are some structures of the brain
+
+26:15.560 --> 26:17.280
+ that the cells know how to make.
+
+26:18.160 --> 26:21.840
+ And another part comes from the investigator,
+
+26:21.840 --> 26:26.160
+ the scientist, adding to the media factors
+
+26:26.160 --> 26:28.040
+ that we know in the mouse, for example,
+
+26:28.040 --> 26:30.760
+ would foster a certain step of development.
+
+26:30.760 --> 26:33.240
+ But it's very limited.
+
+26:33.240 --> 26:36.160
+ And so as a result,
+
+26:36.160 --> 26:38.200
+ the kind of product you get in the end
+
+26:38.200 --> 26:39.720
+ is much more reductionist.
+
+26:39.720 --> 26:42.680
+ It's much more simple than what you get in vivo.
+
+26:42.680 --> 26:46.200
+ It mimics early events of development as of today.
+
+26:46.200 --> 26:49.080
+ And it doesn't build very complex type of anatomy
+
+26:49.080 --> 26:51.440
+ and structure does not as of today,
+
+26:52.600 --> 26:54.920
+ which happens instead in vivo.
+
+26:54.920 --> 26:59.120
+ And also the variability that you see
+
+26:59.120 --> 27:02.800
+ one organoid to the next tends to be higher
+
+27:02.800 --> 27:05.560
+ than when you compare an embryo to the next.
+
+27:05.560 --> 27:08.960
+ So, okay, then the next question is how hard
+
+27:08.960 --> 27:11.120
+ and maybe another flip side of that expensive
+
+27:11.120 --> 27:14.960
+ is it to go from one stem cell to an organoid?
+
+27:14.960 --> 27:16.760
+ How many can you build in like,
+
+27:16.760 --> 27:18.480
+ because it sounds very complicated.
+
+27:18.480 --> 27:23.480
+ It's work, definitely, and it's money, definitely.
+
+27:23.480 --> 27:28.080
+ But you can really grow a very high number
+
+27:28.080 --> 27:31.640
+ of these organoids, you know, can go perhaps,
+
+27:31.640 --> 27:33.160
+ I told you the maximum they become
+
+27:33.160 --> 27:34.800
+ about five millimeters in diameter.
+
+27:34.800 --> 27:39.800
+ So this is about the size of a tiny, tiny, you know, raising
+
+27:40.800 --> 27:43.160
+ or perhaps the seed of an apple.
+
+27:43.160 --> 27:47.560
+ And so you can grow 50 to 100 of those
+
+27:47.560 --> 27:51.360
+ inside one big bioreactors, which are these flasks
+
+27:51.360 --> 27:55.520
+ where the media provides nutrients for the organoids.
+
+27:55.520 --> 28:00.520
+ So the problem is not to grow more or less of them.
+
+28:01.760 --> 28:06.480
+ It's really to figure out how to grow them in a way
+
+28:06.480 --> 28:08.440
+ that they are more and more reproducible.
+
+28:08.440 --> 28:10.000
+ For example, organoid to organoid,
+
+28:10.000 --> 28:13.200
+ so they can be used to study a biological process
+
+28:13.200 --> 28:15.640
+ because if you have too much of variability,
+
+28:15.640 --> 28:17.160
+ then you never know if what you see
+
+28:17.160 --> 28:19.560
+ is just an exception or really the rule.
+
+28:19.560 --> 28:22.160
+ So what does an organoid look like?
+
+28:22.160 --> 28:25.120
+ Are there different neurons already emerging?
+
+28:25.120 --> 28:27.520
+ Is there, you know, well, first,
+
+28:27.520 --> 28:29.920
+ can you tell me what kind of neurons are there?
+
+28:29.920 --> 28:30.920
+ Yes.
+
+28:30.920 --> 28:35.560
+ Are they sort of all the same?
+
+28:35.560 --> 28:37.480
+ Are they not all the same?
+
+28:37.480 --> 28:39.560
+ Is how much do we understand
+
+28:39.560 --> 28:42.440
+ and how much of that variance
+
+28:42.440 --> 28:45.800
+ if any can exist in organoids?
+
+28:45.800 --> 28:49.360
+ Yes, so you could grow,
+
+28:49.360 --> 28:52.440
+ I told you that the brain has different parts.
+
+28:52.440 --> 28:56.000
+ So the cerebral cortex is on the top part of the brain,
+
+28:56.000 --> 28:57.960
+ but there is another region called the striatum
+
+28:57.960 --> 28:59.960
+ that is below the cortex and so on and so forth.
+
+28:59.960 --> 29:03.760
+ All of these regions have different types of cells
+
+29:03.760 --> 29:05.040
+ in the actual brain.
+
+29:05.040 --> 29:05.880
+ Okay.
+
+29:05.880 --> 29:08.880
+ And so scientists have been able to grow organoids
+
+29:08.880 --> 29:11.440
+ that may mimic some aspects of development
+
+29:11.440 --> 29:13.920
+ of these different regions of the brain.
+
+29:13.920 --> 29:16.480
+ And so we are very interested in the cerebral cortex.
+
+29:16.480 --> 29:17.720
+ That's the coolest part, right?
+
+29:17.720 --> 29:18.560
+ Very cool.
+
+29:18.560 --> 29:20.880
+ I agree with you.
+
+29:20.880 --> 29:23.880
+ We wouldn't be here talking if we didn't have a cerebral cortex.
+
+29:23.880 --> 29:25.200
+ It's also, I like to think,
+
+29:25.200 --> 29:27.600
+ the part of the brain that really truly makes us human,
+
+29:27.600 --> 29:30.200
+ the most evolved in recent evolution.
+
+29:30.200 --> 29:33.600
+ And so in the attempt to make the cerebral cortex
+
+29:33.600 --> 29:37.200
+ and by figuring out a way to have these organoids
+
+29:37.200 --> 29:40.240
+ continue to grow and develop for extended periods of time,
+
+29:40.240 --> 29:42.400
+ much like it happens in the real embryo,
+
+29:42.400 --> 29:44.240
+ months and months in culture,
+
+29:44.240 --> 29:47.920
+ then you can see that many different types
+
+29:47.920 --> 29:50.200
+ of neurons of the cortex appear
+
+29:50.200 --> 29:52.200
+ and at some point also the astrocytes,
+
+29:52.200 --> 29:57.200
+ so the glia cells of the cerebral cortex also appear.
+
+29:57.640 --> 29:59.000
+ What are these?
+
+29:59.000 --> 29:59.840
+ Astrocytes.
+
+29:59.840 --> 30:00.680
+ Astrocytes.
+
+30:00.680 --> 30:02.080
+ The astrocytes are not neurons,
+
+30:02.080 --> 30:03.440
+ so they're not nerve cells,
+
+30:03.440 --> 30:06.160
+ but they play very important roles.
+
+30:06.160 --> 30:09.000
+ One important role is to support the neuron,
+
+30:09.000 --> 30:11.880
+ but of course they have much more active type of roles.
+
+30:11.880 --> 30:13.280
+ They're very important, for example,
+
+30:13.280 --> 30:14.560
+ to make the synapses,
+
+30:14.560 --> 30:17.640
+ which are the point of contact and communication
+
+30:17.640 --> 30:21.480
+ between two neurons, they...
+
+30:21.480 --> 30:25.680
+ So all that chemistry fun happens in the synapses
+
+30:25.680 --> 30:28.160
+ happens because of these cells?
+
+30:28.160 --> 30:29.680
+ Are they the medium in which?
+
+30:29.680 --> 30:32.000
+ Happens because of the interactions,
+
+30:32.000 --> 30:34.760
+ happens because you are making the cells
+
+30:34.760 --> 30:36.320
+ and they have certain properties,
+
+30:36.320 --> 30:40.360
+ including the ability to make neurotransmitters,
+
+30:40.360 --> 30:43.320
+ which are the chemicals that are secreted to the synapses,
+
+30:43.320 --> 30:46.480
+ including the ability of making these axons grow
+
+30:46.480 --> 30:49.240
+ with their growth cones and so on and so forth.
+
+30:49.240 --> 30:51.400
+ And then you have other cells around there
+
+30:51.400 --> 30:55.200
+ that release chemicals or touch the neurons
+
+30:55.200 --> 30:57.200
+ or interact with them in different ways
+
+30:57.200 --> 30:59.880
+ to really foster this perfect process,
+
+30:59.880 --> 31:02.480
+ in this case of synaptogenesis.
+
+31:02.480 --> 31:05.680
+ And this does happen within organoids.
+
+31:05.680 --> 31:06.520
+ Or with organoids.
+
+31:06.520 --> 31:09.760
+ So the mechanical and the chemical stuff happens.
+
+31:09.760 --> 31:11.640
+ The connectivity between neurons.
+
+31:11.640 --> 31:13.320
+ This, in a way, is not surprising
+
+31:13.320 --> 31:18.160
+ because scientists have been culturing neurons forever.
+
+31:18.160 --> 31:20.760
+ And when you take a neuron, even a very young one,
+
+31:20.760 --> 31:23.480
+ and you culture it, eventually finds another cell
+
+31:23.480 --> 31:26.960
+ or another neuron to talk to, it will form a synapse.
+
+31:26.960 --> 31:28.520
+ Are we talking about mice neurons?
+
+31:28.520 --> 31:29.600
+ Are we talking about human neurons?
+
+31:29.600 --> 31:30.600
+ It doesn't matter, both.
+
+31:30.600 --> 31:33.280
+ So you can culture a neuron like a single neuron
+
+31:33.280 --> 31:37.920
+ and give it a little friend and it starts interacting?
+
+31:37.920 --> 31:40.240
+ Yes. So neurons are able to...
+
+31:40.240 --> 31:44.560
+ It sounds... It's more simple than what it may sound to you.
+
+31:44.560 --> 31:48.320
+ Neurons have molecular properties and structural properties
+
+31:48.320 --> 31:51.120
+ that allow them to really communicate with other cells.
+
+31:51.120 --> 31:53.160
+ And so if you put not one neuron,
+
+31:53.160 --> 31:55.120
+ but if you put several neurons together,
+
+31:55.120 --> 32:00.240
+ chances are that they will form synapses with each other.
+
+32:00.240 --> 32:01.120
+ Okay, great.
+
+32:01.120 --> 32:03.360
+ So an organoid is not a brain.
+
+32:03.360 --> 32:03.880
+ No.
+
+32:03.880 --> 32:07.600
+ But there's some...
+
+32:07.600 --> 32:10.440
+ It's able to, especially what you're talking about,
+
+32:10.440 --> 32:15.120
+ mimic some properties of the cerebral cortex, for example.
+
+32:15.120 --> 32:17.960
+ So what can you understand about the brain
+
+32:17.960 --> 32:21.040
+ by studying an organoid of the cerebral cortex?
+
+32:21.040 --> 32:26.400
+ I can literally study all this incredible diversity of cell type,
+
+32:26.400 --> 32:29.040
+ all these many, many different classes of cells.
+
+32:29.040 --> 32:30.760
+ How are they made?
+
+32:30.760 --> 32:32.520
+ How do they look like?
+
+32:32.520 --> 32:34.920
+ What do they need to be made properly?
+
+32:34.920 --> 32:36.280
+ And what goes wrong?
+
+32:36.280 --> 32:39.680
+ If now the genetics of that stem cell
+
+32:39.680 --> 32:42.800
+ that I used to make the organoid came from a patient
+
+32:42.800 --> 32:44.320
+ with a neurodevelopmental disease,
+
+32:44.320 --> 32:47.600
+ can I actually watch for the very first time
+
+32:47.600 --> 32:51.400
+ what may have gone wrong years before in this kid
+
+32:51.400 --> 32:53.480
+ when its own brain was being made?
+
+32:53.480 --> 32:54.720
+ Think about that loop.
+
+32:54.720 --> 32:59.600
+ In a way, it's a little tiny rudimentary window
+
+32:59.600 --> 33:04.240
+ into the past, into the time when that brain,
+
+33:04.240 --> 33:07.680
+ in a kid that had this neurodevelopmental disease,
+
+33:07.680 --> 33:10.120
+ was being made.
+
+33:10.120 --> 33:12.880
+ And I think that's unbelievably powerful
+
+33:12.880 --> 33:16.800
+ because today we have no idea of what cell types,
+
+33:16.800 --> 33:20.880
+ we barely know what brain regions are affected in these diseases.
+
+33:20.880 --> 33:23.720
+ Now we have an experimental system
+
+33:23.720 --> 33:25.440
+ that we can study in the lab
+
+33:25.440 --> 33:28.440
+ and we can ask what are the cells affected?
+
+33:28.440 --> 33:31.840
+ When, during development, things went wrong.
+
+33:31.840 --> 33:35.200
+ What are the molecules among the many, many different molecules
+
+33:35.200 --> 33:36.600
+ that control brain development?
+
+33:36.600 --> 33:39.720
+ Which ones are the ones that really messed up here
+
+33:39.720 --> 33:42.160
+ and we want perhaps to fix?
+
+33:42.160 --> 33:44.520
+ And what is really the final product?
+
+33:44.520 --> 33:48.560
+ Is it a less strong kind of circuit and brain?
+
+33:48.560 --> 33:50.560
+ Is it a brain that lacks a cell type?
+
+33:50.560 --> 33:52.040
+ Is it a, what is it?
+
+33:52.040 --> 33:54.920
+ Because then we can think about treatment
+
+33:54.920 --> 33:59.360
+ and care for these patients that is informed
+
+33:59.360 --> 34:02.040
+ rather than just based on current diagnostics.
+
+34:02.040 --> 34:06.240
+ So how hard is it to detect through the developmental process?
+
+34:06.240 --> 34:09.920
+ It's a super exciting tool
+
+34:09.920 --> 34:15.160
+ to see how different conditions develop.
+
+34:15.160 --> 34:17.640
+ How hard is it to detect that, wait a minute,
+
+34:17.640 --> 34:20.760
+ this is abnormal development.
+
+34:20.760 --> 34:22.080
+ Yeah.
+
+34:22.080 --> 34:24.840
+ That's how hard it, how much signals there,
+
+34:24.840 --> 34:26.520
+ how much of it is it a mess?
+
+34:26.520 --> 34:29.520
+ Because things can go wrong at multiple levels, right?
+
+34:29.520 --> 34:34.360
+ You could have a cell that is born and built
+
+34:34.360 --> 34:36.280
+ but then doesn't work properly
+
+34:36.280 --> 34:38.360
+ or a cell that is not even born
+
+34:38.360 --> 34:40.760
+ or a cell that doesn't interact with other cells differently
+
+34:40.760 --> 34:42.160
+ and so on and so forth.
+
+34:42.160 --> 34:44.440
+ So today we have technology
+
+34:44.440 --> 34:47.800
+ that we did not have even five years ago
+
+34:47.800 --> 34:49.800
+ that allows us to look, for example,
+
+34:49.800 --> 34:52.160
+ at the molecular picture of a cell,
+
+34:52.160 --> 34:56.840
+ of a single cell in a sea of cells with high precision.
+
+34:56.840 --> 34:58.920
+ And so that molecular information
+
+34:58.920 --> 35:01.840
+ where you compare many, many single cells
+
+35:01.840 --> 35:03.720
+ for the genes that they produce
+
+35:03.720 --> 35:06.240
+ between a control individual
+
+35:06.240 --> 35:10.200
+ and an individual with a neurodevelopmental disease,
+
+35:10.200 --> 35:13.880
+ that may tell you what is different, molecularly.
+
+35:13.880 --> 35:18.640
+ Or you could see that some cells are not even made,
+
+35:18.640 --> 35:20.840
+ for example, or that the process of maturation
+
+35:20.840 --> 35:22.680
+ of the cells may be wrong.
+
+35:22.680 --> 35:25.080
+ There are many different levels here
+
+35:26.040 --> 35:29.640
+ and we can study the cells at the molecular level
+
+35:29.640 --> 35:33.440
+ but also we can use the organoids to ask questions
+
+35:33.440 --> 35:35.360
+ about the properties of the neurons,
+
+35:35.360 --> 35:37.400
+ the functional properties,
+
+35:37.400 --> 35:39.000
+ how they communicate with each other,
+
+35:39.000 --> 35:41.440
+ how they respond to a stimulus and so on and so forth
+
+35:41.440 --> 35:46.440
+ and we may get abnormalities there, right?
+
+35:46.440 --> 35:51.440
+ And detect those, so how early is this work in the,
+
+35:51.920 --> 35:54.400
+ maybe in the history of science?
+
+35:54.400 --> 35:59.400
+ So, so, I mean, like, so if you were to,
+
+35:59.840 --> 36:04.840
+ if you and I time travel a thousand years into the future,
+
+36:05.280 --> 36:10.040
+ organoids seem to be, maybe I'm romanticizing the notion
+
+36:10.040 --> 36:12.880
+ but you're building not a brain
+
+36:12.880 --> 36:15.800
+ but something that has properties of a brain.
+
+36:15.800 --> 36:19.120
+ So it feels like you might be getting close to,
+
+36:19.120 --> 36:23.320
+ in the building process, to build us to understand.
+
+36:23.320 --> 36:28.320
+ So how far are we in this understanding
+
+36:29.160 --> 36:30.360
+ process of development?
+
+36:31.520 --> 36:34.320
+ A thousand years from now, it's a long time from now.
+
+36:34.320 --> 36:36.560
+ So if this planet is still gonna be here,
+
+36:36.560 --> 36:38.280
+ a thousand years from now.
+
+36:38.280 --> 36:42.080
+ So I mean, if, you know, like they write a book,
+
+36:42.080 --> 36:44.120
+ obviously there'll be a chapter about you.
+
+36:44.120 --> 36:47.320
+ That's probably the science fiction book today.
+
+36:47.320 --> 36:48.160
+ Yeah, today.
+
+36:48.160 --> 36:50.840
+ But I mean, I guess where we really understood
+
+36:50.840 --> 36:53.400
+ very little about the brain a century ago,
+
+36:53.400 --> 36:55.920
+ where I was a big fan in high school,
+
+36:55.920 --> 36:58.760
+ reading Freud and so on, still am of psychiatry.
+
+36:59.680 --> 37:01.480
+ I would say we still understand very little
+
+37:01.480 --> 37:04.720
+ about the functional aspect of just,
+
+37:04.720 --> 37:07.760
+ but how in the history of understanding
+
+37:07.760 --> 37:09.640
+ the biology of the brain, the development,
+
+37:09.640 --> 37:11.240
+ how far are we along?
+
+37:11.240 --> 37:12.960
+ It's a very good question.
+
+37:12.960 --> 37:15.520
+ And so this is just, of course, my opinion.
+
+37:15.520 --> 37:19.720
+ I think that we did not have technology,
+
+37:19.720 --> 37:23.160
+ even 10 years ago or certainly not 20 years ago,
+
+37:23.160 --> 37:27.760
+ to even think about experimentally investigating
+
+37:27.760 --> 37:30.160
+ the development of the human brain.
+
+37:30.160 --> 37:32.200
+ So we've done a lot of work in science
+
+37:32.200 --> 37:35.480
+ to study the brain on many other organisms.
+
+37:35.480 --> 37:39.600
+ Now we have some technologies which I'll spell out
+
+37:39.600 --> 37:43.120
+ that allow us to actually look at the real thing
+
+37:43.120 --> 37:45.040
+ and look at the brain, at the human brain.
+
+37:45.040 --> 37:46.840
+ So what are these technologies?
+
+37:46.840 --> 37:50.440
+ There has been huge progress in stem cell biology.
+
+37:50.440 --> 37:54.080
+ The moment someone figured out how to turn a skin cell
+
+37:54.080 --> 37:57.760
+ into an embryonic stem cell, basically,
+
+37:57.760 --> 38:00.160
+ and that how that embryonic stem cell
+
+38:00.160 --> 38:02.480
+ could begin a process of development again
+
+38:02.480 --> 38:04.000
+ to, for example, make a brain,
+
+38:04.000 --> 38:06.040
+ there was a huge, you know, advance.
+
+38:06.040 --> 38:08.200
+ And in fact, there was a Nobel Prize for that.
+
+38:08.200 --> 38:10.440
+ That started the field, really,
+
+38:10.440 --> 38:14.240
+ of using stem cells to build organs.
+
+38:14.240 --> 38:17.080
+ Now we can build on all the knowledge of development
+
+38:17.080 --> 38:18.560
+ that we build over the many, many, many years
+
+38:18.560 --> 38:20.720
+ to say, how do we make these stem cells?
+
+38:20.720 --> 38:22.680
+ Now make more and more complex aspects
+
+38:22.680 --> 38:25.280
+ of development of the human brain.
+
+38:25.280 --> 38:28.480
+ So this field is young, the field of brain organoids,
+
+38:28.480 --> 38:30.120
+ but it's moving fast.
+
+38:30.120 --> 38:32.560
+ And it's moving fast in a very serious way
+
+38:32.560 --> 38:35.960
+ that is rooted in labs with the right ethical framework
+
+38:35.960 --> 38:39.720
+ and really building on, you know,
+
+38:39.720 --> 38:43.520
+ solid science for what reality is and what is not.
+
+38:43.520 --> 38:46.120
+ And, but it will go fast
+
+38:46.120 --> 38:49.120
+ and it will be more and more powerful.
+
+38:49.120 --> 38:52.480
+ We also have technology that allows us to basically study
+
+38:52.480 --> 38:54.640
+ the properties of single cells
+
+38:54.640 --> 38:59.240
+ across many, many millions of single cells,
+
+38:59.240 --> 39:02.160
+ which we didn't have perhaps five years ago.
+
+39:02.160 --> 39:04.840
+ So now with that, even an organoid
+
+39:04.840 --> 39:08.480
+ that has millions of cells can be profiled in a way,
+
+39:08.480 --> 39:11.320
+ looked at with very, very high resolution,
+
+39:11.320 --> 39:14.960
+ the single cell level to really understand what is going on.
+
+39:14.960 --> 39:17.520
+ And you could do it in multiple stages of development
+
+39:17.520 --> 39:20.120
+ and you can build your hypothesis and so on and so forth.
+
+39:20.120 --> 39:22.600
+ So it's not gonna be a thousand years.
+
+39:22.600 --> 39:25.240
+ It's gonna be a shorter amount of time.
+
+39:25.240 --> 39:29.480
+ And I see this as sort of an exponential growth
+
+39:29.480 --> 39:33.560
+ of this field enabled by these technologies
+
+39:33.560 --> 39:35.000
+ that we didn't have before.
+
+39:35.000 --> 39:36.960
+ And so we're gonna see something transformative
+
+39:36.960 --> 39:41.880
+ that we didn't see at all in the prior thousand years.
+
+39:41.880 --> 39:44.640
+ So I apologize for the crazy sci fi questions,
+
+39:44.640 --> 39:48.840
+ but the developmental process is fascinating to watch
+
+39:48.840 --> 39:53.360
+ and study, but how far are we away from
+
+39:53.360 --> 39:57.280
+ and maybe how difficult is it to build
+
+39:57.280 --> 40:02.280
+ not just an organoid, but a human brain from a stem cell?
+
+40:02.280 --> 40:05.640
+ Yeah, first of all, that's not the goal
+
+40:05.640 --> 40:09.400
+ for the majority of the serial scientists that work on this
+
+40:09.400 --> 40:14.160
+ because you don't have to build the whole human brain
+
+40:14.160 --> 40:17.000
+ to make this model useful for understanding
+
+40:17.000 --> 40:20.440
+ how the brain develops or understanding disease.
+
+40:20.440 --> 40:22.440
+ You don't have to build the whole thing.
+
+40:22.440 --> 40:25.200
+ So let me just comment on that, it's fascinating.
+
+40:25.200 --> 40:29.200
+ It shows to me the difference between you and I
+
+40:29.200 --> 40:32.240
+ is you're actually trying to understand the beauty
+
+40:32.240 --> 40:35.520
+ of the human brain and to use it to really help
+
+40:35.520 --> 40:38.800
+ thousands or millions of people with disease and so on, right?
+
+40:38.800 --> 40:41.480
+ From an artificial intelligence perspective,
+
+40:41.480 --> 40:45.600
+ we're trying to build systems that we can put in robots
+
+40:45.600 --> 40:49.080
+ and try to create systems that have echoes
+
+40:49.080 --> 40:52.360
+ of the intelligence about reasoning about the world,
+
+40:52.360 --> 40:53.600
+ navigating the world.
+
+40:53.600 --> 40:56.040
+ It's different objectives, I think.
+
+40:56.040 --> 40:57.520
+ Yeah, that's very much science fiction.
+
+40:57.520 --> 41:00.280
+ Science fiction, but we operate in science fiction a little bit.
+
+41:00.280 --> 41:03.440
+ But so on that point of building a brain,
+
+41:03.440 --> 41:05.800
+ even though that is not the focus or interest,
+
+41:05.800 --> 41:08.520
+ perhaps, of the community, how difficult is it?
+
+41:08.520 --> 41:11.200
+ Is it truly science fiction at this point?
+
+41:11.200 --> 41:13.960
+ I think the field will progress, like I said,
+
+41:13.960 --> 41:17.960
+ and that the system will be more and more complex in a way,
+
+41:17.960 --> 41:18.720
+ right?
+
+41:18.720 --> 41:23.880
+ But there are properties that emerge from the human brain
+
+41:23.880 --> 41:26.640
+ that have to do with the mind, that may have to do with consciousness,
+
+41:26.640 --> 41:29.840
+ that may have to do with intelligence or whatever.
+
+41:29.840 --> 41:33.720
+ We really don't understand even how they can emerge
+
+41:33.720 --> 41:36.880
+ from an actual real brain, and therefore, we cannot measure
+
+41:36.880 --> 41:40.160
+ or study in an organoid.
+
+41:40.160 --> 41:43.040
+ So I think that this field, many, many years from now,
+
+41:43.040 --> 41:48.240
+ may lead to the building of better neural circuits
+
+41:48.240 --> 41:50.640
+ that really are built out of understanding of how
+
+41:50.640 --> 41:52.400
+ this process really works.
+
+41:52.400 --> 41:57.000
+ And it's hard to predict how complex this really will be.
+
+41:57.000 --> 42:01.200
+ I really don't think we're so far from, it makes me laugh, really.
+
+42:01.200 --> 42:05.120
+ It's really that far from building the human brain.
+
+42:05.120 --> 42:10.040
+ But you're going to be building something that is always
+
+42:10.040 --> 42:14.800
+ a bad version of it, but that may have really powerful properties
+
+42:14.800 --> 42:18.560
+ and might be able to respond to stimuli
+
+42:18.560 --> 42:21.880
+ or be used in certain contexts.
+
+42:21.880 --> 42:24.800
+ And this is why I really think that there is no other way
+
+42:24.800 --> 42:28.200
+ to do this science, but within the right ethical framework.
+
+42:28.200 --> 42:31.440
+ Because where you're going with this is also,
+
+42:31.440 --> 42:34.160
+ we can talk about science fiction and write that book,
+
+42:34.160 --> 42:36.600
+ and we could today.
+
+42:36.600 --> 42:41.520
+ But this work happens in a specific ethical framework
+
+42:41.520 --> 42:44.880
+ that we don't decide just as scientists, but also as a society.
+
+42:44.880 --> 42:48.560
+ So the ethical framework here is a fascinating one,
+
+42:48.560 --> 42:51.120
+ is a complicated one.
+
+42:51.120 --> 42:55.680
+ Do you have a sense, a grasp of how we think about ethically,
+
+42:55.680 --> 43:04.160
+ of building organoids from human stem cells to understand the brain?
+
+43:04.160 --> 43:09.720
+ It seems like a tool for helping potentially millions of people
+
+43:09.720 --> 43:14.960
+ cure diseases, or at least start to cure by understanding it.
+
+43:14.960 --> 43:20.560
+ But is there more, is there gray areas that are ethical,
+
+43:20.560 --> 43:22.320
+ that we have to think about ethically?
+
+43:22.320 --> 43:23.160
+ Absolutely.
+
+43:23.160 --> 43:25.520
+ We must think about that.
+
+43:25.520 --> 43:29.560
+ Every discussion about the ethics of this
+
+43:29.560 --> 43:34.480
+ needs to be based on actual data from the models that we have today
+
+43:34.480 --> 43:36.280
+ and from the ones that we will have tomorrow.
+
+43:36.280 --> 43:37.800
+ So it's a continuous conversation.
+
+43:37.800 --> 43:39.840
+ It's not something that you decide now.
+
+43:39.840 --> 43:42.000
+ Today, there is no issue, really.
+
+43:42.000 --> 43:47.240
+ Very simple models that clearly can help you in many ways
+
+43:47.240 --> 43:49.880
+ without much to think about.
+
+43:49.880 --> 43:52.200
+ But tomorrow, we need to have another conversation,
+
+43:52.200 --> 43:53.160
+ and so on and so forth.
+
+43:53.160 --> 43:57.120
+ And so the way we do this is to actually really bring together
+
+43:57.120 --> 44:00.440
+ constantly a group of people that are not only scientists,
+
+44:00.440 --> 44:04.160
+ but also bioethicists, lawyers, philosophers, psychiatrists,
+
+44:04.160 --> 44:06.680
+ and psychologists, and so on and so forth,
+
+44:06.680 --> 44:13.040
+ to decide as a society, really, what we should
+
+44:13.040 --> 44:15.320
+ and what we should not do.
+
+44:15.320 --> 44:17.600
+ So that's the way to think about the ethics.
+
+44:17.600 --> 44:21.440
+ Now, I also think, though, that as a scientist,
+
+44:21.440 --> 44:23.840
+ I have a moral responsibility.
+
+44:23.840 --> 44:28.360
+ So if you think about how transformative
+
+44:28.360 --> 44:32.640
+ it could be for understanding and curing a neuropsychiatric
+
+44:32.640 --> 44:37.320
+ disease, to be able to actually watch and study
+
+44:37.320 --> 44:41.480
+ and treat with drugs the very brain of the patient
+
+44:41.480 --> 44:44.720
+ that you are trying to study, how transformative
+
+44:44.720 --> 44:47.200
+ at this moment in time this could be.
+
+44:47.200 --> 44:47.960
+ We couldn't do it.
+
+44:47.960 --> 44:50.800
+ Five years ago, we could do it now.
+
+44:50.800 --> 44:53.440
+ Taking a stem cell of a particular patient
+
+44:53.440 --> 44:57.480
+ and make an organoid for a simple and different
+
+44:57.480 --> 45:01.160
+ from the human brain, it still is his process
+
+45:01.160 --> 45:04.720
+ of brain development with his or her genetics.
+
+45:04.720 --> 45:08.280
+ And we could understand perhaps what is going wrong.
+
+45:08.280 --> 45:10.960
+ Perhaps we could use as a platform, as a cellular platform,
+
+45:10.960 --> 45:13.720
+ to screen for drugs, to fix a process,
+
+45:13.720 --> 45:15.280
+ and so on and so forth.
+
+45:15.280 --> 45:18.840
+ So we could do it now, we couldn't do it five years ago.
+
+45:18.840 --> 45:20.480
+ Should we not do it?
+
+45:20.480 --> 45:24.760
+ What is the downside of doing it?
+
+45:24.760 --> 45:27.320
+ I don't see a downside at this very moment.
+
+45:27.320 --> 45:30.880
+ If we invited a lot of people, I'm sure there would be
+
+45:30.880 --> 45:33.440
+ somebody who would argue against it,
+
+45:33.440 --> 45:37.680
+ what would be the devil's advocate argument?
+
+45:39.680 --> 45:42.960
+ So it's exactly perhaps what you alluded at
+
+45:42.960 --> 45:47.120
+ with your question, that you are making a,
+
+45:47.120 --> 45:51.680
+ enabling some process of formation of the brain
+
+45:51.680 --> 45:54.440
+ that could be misused at some point,
+
+45:54.440 --> 45:59.080
+ or that could be showing properties
+
+45:59.080 --> 46:03.960
+ that ethically we don't wanna see in a tissue.
+
+46:03.960 --> 46:07.760
+ So today, I repeat, today this is not an issue.
+
+46:07.760 --> 46:11.280
+ And so you just gain dramatically from the science
+
+46:11.280 --> 46:13.720
+ without, because the system is so simple
+
+46:13.720 --> 46:17.840
+ and so different in a way from the actual brain.
+
+46:17.840 --> 46:20.000
+ But because it is the brain,
+
+46:20.000 --> 46:23.960
+ we have an obligation to really consider all of this, right?
+
+46:23.960 --> 46:27.160
+ And again, it's a balanced conversation
+
+46:27.160 --> 46:30.360
+ where we should put disease and betterment of humanity
+
+46:30.360 --> 46:32.440
+ also on that plate.
+
+46:32.440 --> 46:35.440
+ What do you think, at least historically,
+
+46:35.440 --> 46:37.280
+ there was some politicization,
+
+46:37.280 --> 46:42.280
+ politicization of embryonic stem cells,
+
+46:44.360 --> 46:45.960
+ a stem cell research.
+
+46:47.160 --> 46:49.160
+ Do you still see that out there?
+
+46:49.160 --> 46:53.600
+ Is that still a force that we have to think about,
+
+46:53.600 --> 46:55.600
+ especially in this larger discourse
+
+46:55.600 --> 46:57.600
+ that we're having about the role of science
+
+46:57.600 --> 47:00.640
+ in at least American society?
+
+47:00.640 --> 47:03.520
+ Yeah, this is a very good question.
+
+47:03.520 --> 47:05.040
+ It's very, very important.
+
+47:05.040 --> 47:08.480
+ I see a very central role for scientists
+
+47:08.480 --> 47:12.040
+ to inform decisions about what we should
+
+47:12.040 --> 47:14.440
+ or should not do in society.
+
+47:14.440 --> 47:16.400
+ And this is because the scientists
+
+47:16.400 --> 47:20.440
+ have the firsthand look and understanding
+
+47:20.440 --> 47:23.520
+ of really the work that they are doing.
+
+47:23.520 --> 47:26.080
+ And again, this varies depending on
+
+47:26.080 --> 47:27.480
+ what we're talking about here.
+
+47:27.480 --> 47:30.800
+ So now we're talking about brain organoids.
+
+47:30.800 --> 47:33.800
+ I think that the scientists need to be part
+
+47:33.800 --> 47:36.520
+ of that conversation about what is,
+
+47:36.520 --> 47:38.040
+ will be allowed in the future
+
+47:38.040 --> 47:40.840
+ or not allowed in the future to do with the system.
+
+47:40.840 --> 47:43.400
+ And I think that is very, very important
+
+47:43.400 --> 47:47.880
+ because they bring reality of data to the conversation.
+
+47:48.880 --> 47:51.720
+ And so they should have a voice.
+
+47:51.720 --> 47:53.400
+ So data should have a voice.
+
+47:53.400 --> 47:55.200
+ Data needs to have a voice.
+
+47:55.200 --> 47:59.360
+ Because in not only data, we should also be good
+
+47:59.360 --> 48:04.240
+ at communicating with non scientists the data.
+
+48:04.240 --> 48:06.840
+ So there has been, often time,
+
+48:06.840 --> 48:11.840
+ there is a lot of discussion and excitement
+
+48:12.280 --> 48:16.320
+ and fights about certain topics
+
+48:16.320 --> 48:19.320
+ just because of the way they are described.
+
+48:19.320 --> 48:21.000
+ I'll give you an example.
+
+48:21.000 --> 48:23.400
+ If I called the same cellular system,
+
+48:23.400 --> 48:27.080
+ we just talked about a brain organoid.
+
+48:27.080 --> 48:30.320
+ Or if I called it a human mini brain,
+
+48:30.320 --> 48:34.600
+ your reaction is gonna be very different to this.
+
+48:34.600 --> 48:37.760
+ And so the way the systems are described,
+
+48:37.760 --> 48:40.720
+ I mean, we and journalists alike
+
+48:40.720 --> 48:43.720
+ need to be a bit careful that this debate
+
+48:43.720 --> 48:46.080
+ is a real debate and informed by real data.
+
+48:46.080 --> 48:47.960
+ That's all I'm asking.
+
+48:47.960 --> 48:49.600
+ And yeah, the language matters here.
+
+48:49.600 --> 48:51.280
+ So I work on autonomous vehicles
+
+48:51.280 --> 48:54.960
+ and there the use of language could drastically
+
+48:54.960 --> 48:57.480
+ change the interpretation and the way people feel
+
+48:57.480 --> 49:01.520
+ about what is the right way to proceed forward.
+
+49:01.520 --> 49:04.720
+ You are, as I've seen from a presentation,
+
+49:04.720 --> 49:06.240
+ you're a parent.
+
+49:06.240 --> 49:09.840
+ I saw you show a couple of pictures of your son.
+
+49:09.840 --> 49:11.440
+ Is it just the one?
+
+49:11.440 --> 49:12.280
+ Two.
+
+49:12.280 --> 49:13.120
+ Two.
+
+49:13.120 --> 49:13.960
+ Son and a daughter.
+
+49:13.960 --> 49:14.800
+ Son and a daughter.
+
+49:14.800 --> 49:17.360
+ So what have you learned from the human brain
+
+49:17.360 --> 49:20.120
+ by raising two of them?
+
+49:20.120 --> 49:22.800
+ More than I could ever learn in a lab.
+
+49:22.800 --> 49:25.600
+ What have I learned?
+
+49:26.840 --> 49:28.640
+ I've learned that children really have
+
+49:28.640 --> 49:31.520
+ these amazing plastic minds, right?
+
+49:31.520 --> 49:35.880
+ That we have a responsibility to, you know,
+
+49:35.880 --> 49:38.440
+ foster their growth in good, healthy ways
+
+49:39.360 --> 49:42.320
+ that keep them curious, that keep some adventures,
+
+49:42.320 --> 49:45.920
+ that doesn't raise them in fear of things.
+
+49:46.840 --> 49:48.920
+ But also respecting who they are,
+
+49:48.920 --> 49:51.320
+ which is in part, you know, coming from the genetics
+
+49:51.320 --> 49:53.840
+ we talked about, my children are very different
+
+49:53.840 --> 49:55.240
+ from each other despite the fact
+
+49:55.240 --> 49:57.840
+ that they're the product of the same two parents.
+
+49:59.320 --> 50:03.080
+ I also learned that what you do for them
+
+50:03.080 --> 50:04.280
+ comes back to you.
+
+50:04.280 --> 50:05.840
+ Like, you know, if you're a good parent,
+
+50:05.840 --> 50:09.800
+ you're gonna, most of the time have, you know,
+
+50:09.800 --> 50:12.200
+ perhaps decent kids at the end.
+
+50:12.200 --> 50:13.760
+ So what do you think, just a quick comment,
+
+50:13.760 --> 50:17.760
+ what do you think is the source of that difference?
+
+50:17.760 --> 50:20.960
+ It's often the surprising thing for parents.
+
+50:20.960 --> 50:24.000
+ I can't believe that our kids,
+
+50:25.640 --> 50:28.080
+ they're so different, yet they came from the same parents.
+
+50:28.080 --> 50:29.640
+ Well, they are genetically different.
+
+50:29.640 --> 50:31.920
+ Even they came from the same two parents
+
+50:31.920 --> 50:33.640
+ because the mixing of gametes,
+
+50:33.640 --> 50:35.720
+ so when we know these genetics,
+
+50:35.720 --> 50:39.800
+ creates every time a genetically different individual
+
+50:39.800 --> 50:43.760
+ which will have a specific mix of genes
+
+50:43.760 --> 50:46.560
+ that is a different mix every time from the two parents.
+
+50:46.560 --> 50:50.320
+ And so they're not twins.
+
+50:50.320 --> 50:52.960
+ They're genetically different.
+
+50:52.960 --> 50:55.320
+ Just that little bit of variation.
+
+50:55.320 --> 50:58.320
+ As you said, really from a biological perspective,
+
+50:58.320 --> 51:00.600
+ the brains look pretty similar.
+
+51:00.600 --> 51:02.400
+ Well, so let me clarify that.
+
+51:02.400 --> 51:05.400
+ So the genetics you have, the genes that you have,
+
+51:05.400 --> 51:08.680
+ that play that beautiful orchestrated symphony
+
+51:08.680 --> 51:12.040
+ of development, different genes
+
+51:12.040 --> 51:13.920
+ will play it slightly differently.
+
+51:13.920 --> 51:16.120
+ It's like playing the same piece of music
+
+51:16.120 --> 51:17.960
+ but with the different orchestra
+
+51:17.960 --> 51:20.000
+ and a different director, right?
+
+51:20.000 --> 51:21.440
+ The music will not come out.
+
+51:21.440 --> 51:25.400
+ It will be still a piece by the same author
+
+51:25.400 --> 51:27.040
+ but it will come out differently
+
+51:27.040 --> 51:28.920
+ if it's played by the high school orchestra
+
+51:28.920 --> 51:33.440
+ instead of the, instead of the Scala in Milan.
+
+51:34.680 --> 51:39.160
+ And so you are born superficially with the same brain.
+
+51:39.160 --> 51:43.440
+ It has the same cell types, similar patterns of connectivity
+
+51:43.440 --> 51:45.200
+ but the properties of the cells
+
+51:45.200 --> 51:47.600
+ and how the cells will then react to the environment
+
+51:47.600 --> 51:51.320
+ as you experience your world will be also shaped
+
+51:51.320 --> 51:53.680
+ by who genetically you are.
+
+51:53.680 --> 51:55.120
+ Speaking just as a parent,
+
+51:55.120 --> 51:56.880
+ this is not something that comes from my work.
+
+51:56.880 --> 51:58.840
+ I think you can tell at birth
+
+51:58.840 --> 52:01.120
+ that these kids are different
+
+52:01.120 --> 52:04.560
+ and that they have a different personality in a way, right?
+
+52:05.560 --> 52:07.600
+ So both is needed.
+
+52:07.600 --> 52:10.800
+ The genetics as well as the nurturing afterwards.
+
+52:11.760 --> 52:14.600
+ So you are one human with a brain
+
+52:14.600 --> 52:17.200
+ sort of living through the whole mess of it.
+
+52:17.200 --> 52:21.000
+ The human condition, full of love, maybe fear,
+
+52:21.000 --> 52:22.880
+ ultimately mortal.
+
+52:23.880 --> 52:27.080
+ How has studying the brain changed the way you see yourself?
+
+52:27.080 --> 52:29.880
+ When you look in the mirror, when you think about your life,
+
+52:29.880 --> 52:31.880
+ the fears, the love.
+
+52:31.880 --> 52:34.040
+ When you see your own life, your own mortality.
+
+52:34.040 --> 52:36.840
+ Yeah, that's a very good question.
+
+52:37.960 --> 52:43.160
+ It's almost impossible to dissociate some time for me.
+
+52:43.160 --> 52:45.880
+ Some of the things we do or some of the things
+
+52:45.880 --> 52:48.200
+ that other people do from,
+
+52:48.200 --> 52:51.080
+ oh, that's because that part of the brain
+
+52:51.960 --> 52:54.080
+ is working in a certain way.
+
+52:54.080 --> 52:57.840
+ Or thinking about a teenager,
+
+52:59.080 --> 53:01.800
+ going through teenage years and being a time funny
+
+53:01.800 --> 53:03.560
+ in the way they think.
+
+53:03.560 --> 53:07.200
+ And impossible for me not to think it's because
+
+53:07.200 --> 53:10.480
+ they're going through this period of time called
+
+53:10.480 --> 53:12.640
+ critical periods of plasticity.
+
+53:12.640 --> 53:13.480
+ Yeah.
+
+53:13.480 --> 53:16.400
+ Where their synapses are being eliminated here and there
+
+53:16.400 --> 53:17.760
+ and they're just confused.
+
+53:17.760 --> 53:22.280
+ And so from that comes perhaps a different take
+
+53:22.280 --> 53:27.280
+ on that behavior or maybe I can justify scientifically
+
+53:28.080 --> 53:30.120
+ in some sort of way.
+
+53:30.120 --> 53:32.280
+ I also look at humanity in general
+
+53:32.280 --> 53:37.040
+ and I am amazed by what we can do
+
+53:37.040 --> 53:39.960
+ and the kind of ideas that we can come up with.
+
+53:39.960 --> 53:42.840
+ And I cannot stop thinking about
+
+53:42.840 --> 53:46.440
+ how the brain is continuing to evolve.
+
+53:46.440 --> 53:47.360
+ I don't know if you do this,
+
+53:47.360 --> 53:49.720
+ but I think about the next brain sometimes.
+
+53:49.720 --> 53:51.080
+ Where are we going with this?
+
+53:51.080 --> 53:53.920
+ Like what are the features of this brain
+
+53:53.920 --> 53:57.920
+ that evolution is really playing with
+
+53:57.920 --> 54:02.920
+ to get us in the future, the new brain?
+
+54:03.080 --> 54:04.280
+ It's not over, right?
+
+54:04.280 --> 54:07.200
+ It's a work in progress.
+
+54:07.200 --> 54:09.280
+ So let me just a quick comment on that.
+
+54:09.280 --> 54:14.280
+ Do you see, do you think there's a lot of fascination
+
+54:14.520 --> 54:16.240
+ and hope for artificial intelligence
+
+54:16.240 --> 54:17.960
+ of creating artificial brains?
+
+54:17.960 --> 54:20.320
+ You said the next brain.
+
+54:20.320 --> 54:23.600
+ When you imagine over a period of a thousand years
+
+54:23.600 --> 54:25.760
+ the evolution of the human brain,
+
+54:25.760 --> 54:28.920
+ do you sometimes envisioning that future
+
+54:28.920 --> 54:31.440
+ see an artificial one?
+
+54:31.440 --> 54:34.280
+ Artificial intelligence as it is hoped by many,
+
+54:34.280 --> 54:36.840
+ not hoped, thought by many people
+
+54:36.840 --> 54:39.120
+ would be actually the next evolutionary step
+
+54:39.120 --> 54:40.680
+ in the development of humans.
+
+54:40.680 --> 54:45.480
+ Yeah, I think in a way that will happen, right?
+
+54:45.480 --> 54:48.760
+ It's almost like a part of the way we evolve.
+
+54:48.760 --> 54:51.400
+ We evolve in the world that we created,
+
+54:51.400 --> 54:55.520
+ that we interact with, that shape us as we grow up
+
+54:55.520 --> 54:56.840
+ and so on and so forth.
+
+54:58.440 --> 55:01.120
+ Sometime I think about something that may sound silly,
+
+55:01.120 --> 55:04.720
+ but think about the use of cell phones.
+
+55:04.720 --> 55:07.240
+ Part of me thinks that somehow in their brain
+
+55:07.240 --> 55:09.160
+ there will be a region of the cortex
+
+55:09.160 --> 55:13.720
+ that is attuned to that tool.
+
+55:13.720 --> 55:16.600
+ And this comes from a lot of studies
+
+55:16.600 --> 55:21.000
+ in model organisms where really the cortex
+
+55:21.000 --> 55:24.280
+ especially adapts to the kind of things you have to do.
+
+55:24.280 --> 55:28.640
+ So if we need to move our fingers in a very specific way,
+
+55:28.640 --> 55:31.040
+ we have a part of our cortex that allows us to do
+
+55:31.040 --> 55:33.080
+ this kind of very precise movement.
+
+55:34.440 --> 55:37.000
+ An owl that has to see very, very far away
+
+55:37.000 --> 55:39.440
+ with big eyes, the visual cortex, very big.
+
+55:39.440 --> 55:43.280
+ It's the brain attunes to your environment.
+
+55:43.280 --> 55:47.600
+ So the brain will attune to the technologies
+
+55:47.600 --> 55:51.200
+ that we will have and will be shaped by it.
+
+55:51.200 --> 55:53.000
+ So the cortex very well may be.
+
+55:53.000 --> 55:54.640
+ Will be shaped by it.
+
+55:54.640 --> 55:57.360
+ In artificial intelligence, it may merge with it,
+
+55:57.360 --> 56:01.240
+ it may get enveloped and adjusted.
+
+56:01.240 --> 56:04.200
+ Even if it's not a merge of the kind of,
+
+56:04.200 --> 56:07.000
+ oh, let's have a synthetic element together
+
+56:07.000 --> 56:08.800
+ with a biological one.
+
+56:08.800 --> 56:11.840
+ The very space around us, the fact, for example,
+
+56:11.840 --> 56:15.280
+ think about we put on some goggles of virtual reality
+
+56:15.280 --> 56:18.840
+ and we physically are surfing the ocean, right?
+
+56:18.840 --> 56:21.800
+ Like I've done it and you have all these emotions
+
+56:21.800 --> 56:26.800
+ that come to you, your brain placed you in that reality.
+
+56:27.200 --> 56:29.720
+ And it was able to do it like that
+
+56:29.720 --> 56:31.160
+ just by putting the goggles on.
+
+56:31.160 --> 56:36.040
+ I didn't take thousands of years of adapting to this.
+
+56:36.040 --> 56:39.360
+ The brain is plastic, so adapts to new technology.
+
+56:39.360 --> 56:41.840
+ So you could do it from the outside
+
+56:41.840 --> 56:46.840
+ by simply hijacking some sensory capacities that we have.
+
+56:47.680 --> 56:51.640
+ So clearly over recent evolution,
+
+56:51.640 --> 56:54.040
+ the cerebral cortex has been a part of the brain
+
+56:54.040 --> 56:56.040
+ that has known the most evolution.
+
+56:56.040 --> 57:00.840
+ So we have put a lot of chips on evolving
+
+57:00.840 --> 57:02.640
+ this specific part of the brain
+
+57:02.640 --> 57:06.000
+ and the evolution of cortex is plasticity.
+
+57:06.000 --> 57:10.320
+ It's this ability to change in response to things.
+
+57:10.320 --> 57:13.840
+ So yes, they will integrate that we want it or not.
+
+57:15.000 --> 57:18.200
+ Well, there's no better way to end it, Paola.
+
+57:18.200 --> 57:19.520
+ Thank you so much for talking to me.
+
+57:19.520 --> 57:20.360
+ You're very welcome.
+
+57:20.360 --> 57:21.200
+ That's great.
+
+57:21.200 --> 57:31.200
+ Thank you.
+

vtt/episode_034_small.vtt

@@ -0,0 +1,1529 @@
+WEBVTT
+
+00:00.000 --> 00:04.640
+ The following is a conversation with Pamela McCordick. She's an author who has written
+
+00:04.640 --> 00:08.400
+ on the history and the philosophical significance of artificial intelligence.
+
+00:09.040 --> 00:17.440
+ Her books include Machines Who Think in 1979, The Fifth Generation in 1983, with Ed Fangenbaum,
+
+00:17.440 --> 00:22.960
+ who's considered to be the father of expert systems, The Edge of Chaos, The Features of Women,
+
+00:22.960 --> 00:28.960
+ and many more books. I came across her work in an unusual way by stumbling in a quote from
+
+00:28.960 --> 00:35.280
+ Machines Who Think that is something like, artificial intelligence began with the ancient
+
+00:35.280 --> 00:41.920
+ wish to forge the gods. That was a beautiful way to draw a connecting line between our societal
+
+00:41.920 --> 00:48.720
+ relationship with AI from the grounded day to day science, math, and engineering to popular stories
+
+00:48.720 --> 00:55.520
+ and science fiction and myths of automatons that go back for centuries. Through her literary work,
+
+00:55.520 --> 01:00.400
+ she has spent a lot of time with the seminal figures of artificial intelligence,
+
+01:00.400 --> 01:07.760
+ including the founding fathers of AI from the 1956 Dartmouth summer workshop where the field
+
+01:07.760 --> 01:13.600
+ was launched. I reached out to Pamela for a conversation in hopes of getting a sense of
+
+01:13.600 --> 01:18.400
+ what those early days were like and how their dreams continued to reverberate
+
+01:18.400 --> 01:23.840
+ through the work of our community today. I often don't know where the conversation may take us,
+
+01:23.840 --> 01:29.600
+ but I jump in and see. Having no constraints, rules, or goals is a wonderful way to discover new
+
+01:29.600 --> 01:36.320
+ ideas. This is the Artificial Intelligence Podcast. If you enjoy it, subscribe on YouTube,
+
+01:36.320 --> 01:41.600
+ give it five stars on iTunes, support it on Patreon, or simply connect with me on Twitter
+
+01:41.600 --> 01:49.680
+ at Lex Freedman, spelled F R I D M A N. And now here's my conversation with Pamela McCordick.
+
+01:49.680 --> 01:58.640
+ In 1979, your book, Machines Who Think, was published. In it, you interview some of the early
+
+01:58.640 --> 02:06.320
+ AI pioneers and explore the idea that AI was born not out of maybe math and computer science,
+
+02:06.320 --> 02:14.960
+ but out of myth and legend. So tell me if you could the story of how you first arrived at the
+
+02:14.960 --> 02:22.400
+ book, the journey of beginning to write it. I had been a novelist. I'd published two novels.
+
+02:23.120 --> 02:31.760
+ And I was sitting under the portal at Stanford one day in the house we were renting for the
+
+02:31.760 --> 02:37.760
+ summer. And I thought, I should write a novel about these weird people in AI, I know. And then I
+
+02:37.760 --> 02:44.240
+ thought, ah, don't write a novel, write a history. Simple. Just go around, you know, interview them,
+
+02:44.240 --> 02:50.400
+ splice it together. Voila, instant book. Ha, ha, ha. It was much harder than that.
+
+02:50.400 --> 02:58.320
+ But nobody else was doing it. And so I thought, well, this is a great opportunity. And there were
+
+02:59.760 --> 03:06.240
+ people who, John McCarthy, for example, thought it was a nutty idea. There were much, you know,
+
+03:06.240 --> 03:11.920
+ the field had not evolved yet, so on. And he had some mathematical thing he thought I should write
+
+03:11.920 --> 03:18.480
+ instead. And I said, no, John, I am not a woman in search of a project. I'm, this is what I want
+
+03:18.480 --> 03:24.000
+ to do. I hope you'll cooperate. And he said, oh, mother, mother, well, okay, it's your, your time.
+
+03:24.960 --> 03:31.280
+ What was the pitch for the, I mean, such a young field at that point. How do you write
+
+03:31.280 --> 03:37.520
+ a personal history of a field that's so young? I said, this is wonderful. The founders of the
+
+03:37.520 --> 03:43.120
+ field are alive and kicking and able to talk about what they're doing. Did they sound or feel like
+
+03:43.120 --> 03:48.560
+ founders at the time? Did they know that they've been found, that they've founded something? Oh,
+
+03:48.560 --> 03:55.120
+ yeah, they knew what they were doing was very important, very. What they, what I now see in
+
+03:55.120 --> 04:04.080
+ retrospect is that they were at the height of their research careers. And it's humbling to me
+
+04:04.080 --> 04:09.280
+ that they took time out from all the things that they had to do as a consequence of being there.
+
+04:10.320 --> 04:14.560
+ And to talk to this woman who said, I think I'm going to write a book about you.
+
+04:14.560 --> 04:23.840
+ No, it was amazing, just amazing. So who, who stands out to you? Maybe looking 63 years ago,
+
+04:23.840 --> 04:31.040
+ the Dartmouth conference. So Marvin Minsky was there. McCarthy was there. Claude Shannon,
+
+04:31.040 --> 04:36.960
+ Alan Newell, Herb Simon, some of the folks you've mentioned. Right. Then there's other characters,
+
+04:36.960 --> 04:44.720
+ right? One of your coauthors. He wasn't at Dartmouth. He wasn't at Dartmouth, but I mean.
+
+04:44.720 --> 04:50.880
+ He was a, I think an undergraduate then. And, and of course, Joe Traub. I mean,
+
+04:50.880 --> 04:58.720
+ all of these are players, not at Dartmouth them, but in that era. Right. It's same you and so on.
+
+04:58.720 --> 05:03.680
+ So who are the characters, if you could paint a picture that stand out to you from memory,
+
+05:03.680 --> 05:07.200
+ those people you've interviewed and maybe not people that were just in the,
+
+05:08.400 --> 05:13.760
+ in the, the atmosphere, in the atmosphere. Of course, the four founding fathers were
+
+05:13.760 --> 05:17.040
+ extraordinary guys. They really were. Who are the founding fathers?
+
+05:18.560 --> 05:22.480
+ Alan Newell, Herbert Simon, Marvin Minsky, John McCarthy,
+
+05:22.480 --> 05:26.240
+ they were the four who were not only at the Dartmouth conference,
+
+05:26.240 --> 05:31.200
+ but Newell and Simon arrived there with a working program called the logic theorist.
+
+05:31.200 --> 05:38.400
+ Everybody else had great ideas about how they might do it, but they weren't going to do it yet.
+
+05:41.040 --> 05:48.720
+ And you mentioned Joe Traub, my husband. I was immersed in AI before I met Joe,
+
+05:50.080 --> 05:54.960
+ because I had been Ed Feigenbaum's assistant at Stanford. And before that,
+
+05:54.960 --> 06:01.520
+ I had worked on a book by edited by Feigenbaum and Julian Feldman called
+
+06:02.800 --> 06:09.200
+ Computers and Thought. It was the first textbook of readings of AI. And they, they only did it
+
+06:09.200 --> 06:13.120
+ because they were trying to teach AI to people at Berkeley. And there was nothing, you know,
+
+06:13.120 --> 06:17.600
+ you'd have to send them to this journal and that journal. This was not the internet where you could
+
+06:17.600 --> 06:26.080
+ go look at an article. So I was fascinated from the get go by AI. I was an English major, you know,
+
+06:26.080 --> 06:33.200
+ what did I know? And yet I was fascinated. And that's why you saw that historical,
+
+06:33.200 --> 06:40.320
+ that literary background, which I think is very much a part of the continuum of AI that
+
+06:40.320 --> 06:48.000
+ the AI grew out of that same impulse. Was that, yeah, that traditional? What, what was, what drew
+
+06:48.000 --> 06:54.800
+ you to AI? How did you even think of it back, back then? What, what was the possibilities,
+
+06:54.800 --> 07:03.200
+ the dreams? What was interesting to you? The idea of intelligence outside the human cranium,
+
+07:03.200 --> 07:08.000
+ this was a phenomenal idea. And even when I finished machines who think,
+
+07:08.000 --> 07:15.040
+ I didn't know if they were going to succeed. In fact, the final chapter is very wishy washy,
+
+07:15.040 --> 07:25.200
+ frankly. I don't succeed the field did. Yeah. Yeah. So was there the idea that AI began with
+
+07:25.200 --> 07:32.000
+ the wish to forge the God? So the spiritual component that we crave to create this other
+
+07:32.000 --> 07:40.880
+ thing greater than ourselves? For those guys, I don't think so. Newell and Simon were cognitive
+
+07:40.880 --> 07:49.840
+ psychologists. What they wanted was to simulate aspects of human intelligence. And they found
+
+07:49.840 --> 07:57.600
+ they could do it on the computer. Minsky just thought it was a really cool thing to do.
+
+07:57.600 --> 08:07.120
+ Likewise, McCarthy. McCarthy had got the idea in 1949 when, when he was a Caltech student. And
+
+08:08.560 --> 08:15.520
+ he listened to somebody's lecture. It's in my book, I forget who it was. And he thought,
+
+08:15.520 --> 08:20.480
+ oh, that would be fun to do. How do we do that? And he took a very mathematical approach.
+
+08:20.480 --> 08:28.800
+ Minsky was hybrid. And Newell and Simon were very much cognitive psychology. How can we
+
+08:28.800 --> 08:37.280
+ simulate various things about human cognition? What happened over the many years is, of course,
+
+08:37.280 --> 08:42.480
+ our definition of intelligence expanded tremendously. I mean, these days,
+
+08:43.920 --> 08:48.960
+ biologists are comfortable talking about the intelligence of cell, the intelligence of the
+
+08:48.960 --> 08:57.840
+ brain, not just human brain, but the intelligence of any kind of brain, cephalopause. I mean,
+
+08:59.520 --> 09:05.840
+ an octopus is really intelligent by any, we wouldn't have thought of that in the 60s,
+
+09:05.840 --> 09:12.560
+ even the 70s. So all these things have worked in. And I did hear one behavioral
+
+09:12.560 --> 09:20.640
+ primatologist, Franz Duval, say AI taught us the questions to ask.
+
+09:22.800 --> 09:27.760
+ Yeah, this is what happens, right? It's when you try to build it, is when you start to actually
+
+09:27.760 --> 09:35.360
+ ask questions, if it puts a mirror to ourselves. So you were there in the middle of it. It seems
+
+09:35.360 --> 09:41.920
+ like not many people were asking the questions that you were trying to look at this field,
+
+09:41.920 --> 09:48.480
+ the way you were. I was solo. When I went to get funding for this, because I needed somebody to
+
+09:48.480 --> 09:59.840
+ transcribe the interviews and I needed travel expenses, I went to every thing you could think of,
+
+09:59.840 --> 10:11.280
+ the NSF, the DARPA. There was an Air Force place that doled out money. And each of them said,
+
+10:11.840 --> 10:18.640
+ well, that was very interesting. That's a very interesting idea. But we'll think about it.
+
+10:19.200 --> 10:24.320
+ And the National Science Foundation actually said to me in plain English,
+
+10:24.320 --> 10:30.960
+ hey, you're only a writer. You're not an historian of science. And I said, yeah, that's true. But
+
+10:30.960 --> 10:35.360
+ the historians of science will be crawling all over this field. I'm writing for the general
+
+10:35.360 --> 10:44.000
+ audience. So I thought, and they still wouldn't budge. I finally got a private grant without
+
+10:44.000 --> 10:51.440
+ knowing who it was from from Ed Fredkin at MIT. He was a wealthy man, and he liked what he called
+
+10:51.440 --> 10:56.880
+ crackpot ideas. And he considered this a crackpot idea. This a crackpot idea. And he was willing to
+
+10:56.880 --> 11:04.240
+ support it. I am ever grateful. Let me say that. You know, some would say that a history of science
+
+11:04.240 --> 11:09.360
+ approach to AI, or even just a history or anything like the book that you've written,
+
+11:09.360 --> 11:16.640
+ hasn't been written since. Maybe I'm not familiar. But it's certainly not many.
+
+11:16.640 --> 11:24.000
+ If we think about bigger than just these couple of decades, a few decades, what are the roots
+
+11:25.120 --> 11:32.160
+ of AI? Oh, they go back so far. Yes, of course, there's all the legendary stuff, the
+
+11:32.800 --> 11:42.160
+ Golem and the early robots of the 20th century. But they go back much further than that. If
+
+11:42.160 --> 11:50.320
+ you read Homer, Homer has robots in the Iliad. And a classical scholar was pointing out to me
+
+11:50.320 --> 11:55.440
+ just a few months ago. Well, you said you just read the Odyssey. The Odyssey is full of robots.
+
+11:55.440 --> 12:01.520
+ It is, I said. Yeah, how do you think Odysseus's ship gets from place one place to another? He
+
+12:01.520 --> 12:09.040
+ doesn't have the crew people to do that, the crew men. Yeah, it's magic. It's robots. Oh, I thought.
+
+12:09.040 --> 12:17.680
+ How interesting. So we've had this notion of AI for a long time. And then toward the end of the
+
+12:17.680 --> 12:24.000
+ 19th century, the beginning of the 20th century, there were scientists who actually tried to
+
+12:24.000 --> 12:29.280
+ make this happen some way or another, not successfully, they didn't have the technology
+
+12:29.280 --> 12:40.160
+ for it. And of course, Babbage, in the 1850s and 60s, he saw that what he was building was capable
+
+12:40.160 --> 12:47.040
+ of intelligent behavior. And he, when he ran out of funding, the British government finally said,
+
+12:47.040 --> 12:53.360
+ that's enough. He and Lady Lovelace decided, oh, well, why don't we make, you know, why don't we
+
+12:53.360 --> 13:00.560
+ play the ponies with this? He had other ideas for raising money too. But if we actually reach back
+
+13:00.560 --> 13:07.280
+ once again, I think people don't actually really know that robots do appear or ideas of robots.
+
+13:07.280 --> 13:14.240
+ You talk about the Hellenic and the Hebraic points of view. Oh, yes. Can you tell me about each?
+
+13:15.040 --> 13:22.480
+ I defined it this way, the Hellenic point of view is robots are great. You know, they're party help,
+
+13:22.480 --> 13:30.320
+ they help this guy, Hephaestus, this God Hephaestus in his forge. I presume he made them to help him,
+
+13:31.200 --> 13:38.560
+ and so on and so forth. And they welcome the whole idea of robots. The Hebraic view has to do with,
+
+13:39.360 --> 13:46.800
+ I think it's the second commandment, thou shalt not make any graven image. In other words, you
+
+13:46.800 --> 13:54.480
+ better not start imitating humans, because that's just forbidden. It's the second commandment.
+
+13:55.520 --> 14:05.840
+ And a lot of the reaction to artificial intelligence has been a sense that this is
+
+14:05.840 --> 14:16.800
+ this is somehow wicked. This is somehow blasphemous. We shouldn't be going there. Now, you can say,
+
+14:16.800 --> 14:21.600
+ yeah, but there're going to be some downsides. And I say, yes, there are. But blasphemy is not one of
+
+14:21.600 --> 14:29.520
+ them. You know, there's a kind of fear that feels to be almost primal. Is there religious roots to
+
+14:29.520 --> 14:36.160
+ that? Because so much of our society has religious roots. And so there is a feeling of, like you
+
+14:36.160 --> 14:44.080
+ said, blasphemy of creating the other, of creating something, you know, it doesn't have to be artificial
+
+14:44.080 --> 14:50.480
+ intelligence. It's creating life in general. It's the Frankenstein idea. There's the annotated
+
+14:50.480 --> 14:58.080
+ Frankenstein on my coffee table. It's a tremendous novel. It really is just beautifully perceptive.
+
+14:58.080 --> 15:06.800
+ Yes, we do fear this and we have good reason to fear it, but because it can get out of hand.
+
+15:06.800 --> 15:11.360
+ Maybe you can speak to that fear, the psychology, if you thought about it, you know,
+
+15:11.360 --> 15:16.160
+ there's a practical set of fears, concerns in the short term, you can think of, if we actually
+
+15:16.160 --> 15:22.720
+ think about artificial intelligence systems, you can think about bias of discrimination in
+
+15:22.720 --> 15:32.160
+ algorithms or you can think about their social networks, have algorithms that recommend the
+
+15:32.160 --> 15:37.680
+ content you see, thereby these algorithms control the behavior of the masses. There's these concerns.
+
+15:38.240 --> 15:45.040
+ But to me, it feels like the fear that people have is deeper than that. So have you thought about
+
+15:45.040 --> 15:53.440
+ the psychology of it? I think in a superficial way I have. There is this notion that if we
+
+15:55.680 --> 16:01.200
+ produce a machine that can think, it will outthink us and therefore replace us.
+
+16:02.000 --> 16:11.840
+ I guess that's a primal fear of almost kind of a kind of mortality. So around the time you said
+
+16:11.840 --> 16:21.920
+ you worked with Ed Stamford with Ed Faganbaum. So let's look at that one person throughout his
+
+16:21.920 --> 16:31.600
+ history, clearly a key person, one of the many in the history of AI. How has he changed in general
+
+16:31.600 --> 16:36.480
+ around him? How has Stamford changed in the last, how many years are we talking about here?
+
+16:36.480 --> 16:44.720
+ Oh, since 65. So maybe it doesn't have to be about him. It could be bigger, but because he was a
+
+16:44.720 --> 16:51.360
+ key person in expert systems, for example, how are these folks who you've interviewed
+
+16:53.040 --> 16:58.400
+ in the 70s, 79, changed through the decades?
+
+16:58.400 --> 17:10.720
+ In Ed's case, I know him well. We are dear friends. We see each other every month or so.
+
+17:11.520 --> 17:16.160
+ He told me that when machines who think first came out, he really thought all the front
+
+17:16.160 --> 17:26.000
+ matter was kind of baloney. And 10 years later, he said, no, I see what you're getting at. Yes,
+
+17:26.000 --> 17:31.520
+ this is an impulse that has been, this has been a human impulse for thousands of years
+
+17:32.160 --> 17:36.640
+ to create something outside the human cranium that has intelligence.
+
+17:41.120 --> 17:47.520
+ I think it's very hard when you're down at the algorithmic level, and you're just trying to
+
+17:47.520 --> 17:53.840
+ make something work, which is hard enough to step back and think of the big picture.
+
+17:53.840 --> 18:02.000
+ It reminds me of when I was in Santa Fe, I knew a lot of archaeologists, which was a hobby of mine,
+
+18:02.800 --> 18:08.000
+ and I would say, yeah, yeah, well, you can look at the shards and say, oh,
+
+18:08.000 --> 18:14.160
+ this came from this tribe and this came from this trade route and so on. But what about the big
+
+18:14.160 --> 18:21.600
+ picture? And a very distinguished archaeologist said to me, they don't think that way. You do
+
+18:21.600 --> 18:27.920
+ know they're trying to match the shard to the to where it came from. That's, you know, where did
+
+18:27.920 --> 18:34.480
+ this corn, the remainder of this corn come from? Was it grown here? Was it grown elsewhere? And I
+
+18:34.480 --> 18:44.560
+ think this is part of the AI, any scientific field. You're so busy doing the hard work. And it is
+
+18:44.560 --> 18:49.920
+ hard work that you don't step back and say, oh, well, now let's talk about the, you know,
+
+18:49.920 --> 18:56.720
+ the general meaning of all this. Yes. So none of the, even Minsky and McCarthy,
+
+18:58.080 --> 19:04.880
+ they, oh, those guys did. Yeah. The founding fathers did early on or pretty early on. Well,
+
+19:04.880 --> 19:11.200
+ they had, but in a different way from how I looked at it, the two cognitive psychologists,
+
+19:11.200 --> 19:20.960
+ Newell and Simon, they wanted to imagine reforming cognitive psychology so that we would really,
+
+19:20.960 --> 19:31.520
+ really understand the brain. Yeah. Minsky was more speculative. And John McCarthy saw it as,
+
+19:32.960 --> 19:40.080
+ I think I'm doing, doing him right by this. He really saw it as a great boon for human beings to
+
+19:40.080 --> 19:49.440
+ have this technology. And that was reason enough to do it. And he had wonderful, wonderful fables
+
+19:50.240 --> 19:57.920
+ about how if you do the mathematics, you will see that these things are really good for human beings.
+
+19:57.920 --> 20:05.280
+ And if you had a technological objection, he had an answer, a technological answer. But here's how
+
+20:05.280 --> 20:10.480
+ we could get over that. And then blah, blah, blah, blah. And one of his favorite things was
+
+20:10.480 --> 20:15.680
+ what he called the literary problem, which of course, he presented to me several times.
+
+20:16.400 --> 20:23.680
+ That is, everything in literature, there are conventions in literature. One of the conventions
+
+20:23.680 --> 20:37.040
+ is that you have a villain and a hero. And the hero in most literature is human. And the villain
+
+20:37.040 --> 20:42.000
+ in most literature is a machine. And he said, no, that's just not the way it's going to be.
+
+20:42.560 --> 20:48.000
+ But that's the way we're used to it. So when we tell stories about AI, it's always with this
+
+20:48.000 --> 20:57.760
+ paradigm. I thought, yeah, he's right. Looking back, the classics, RUR is certainly the machines
+
+20:57.760 --> 21:07.040
+ trying to overthrow the humans. Frankenstein is different. Frankenstein is a creature.
+
+21:08.480 --> 21:14.560
+ He never has a name. Frankenstein, of course, is the guy who created him, the human Dr. Frankenstein.
+
+21:14.560 --> 21:23.120
+ And this creature wants to be loved, wants to be accepted. And it is only when Frankenstein
+
+21:24.720 --> 21:32.720
+ turns his head, in fact, runs the other way. And the creature is without love
+
+21:34.400 --> 21:38.560
+ that he becomes the monster that he later becomes.
+
+21:39.680 --> 21:43.840
+ So who's the villain in Frankenstein? It's unclear, right?
+
+21:43.840 --> 21:45.520
+ Oh, it is unclear. Yeah.
+
+21:45.520 --> 21:54.320
+ It's really the people who drive him, by driving him away, they bring out the worst.
+
+21:54.320 --> 22:00.800
+ That's right. They give him no human solace. And he is driven away, you're right.
+
+22:03.040 --> 22:10.160
+ He becomes, at one point, the friend of a blind man. And he serves this blind man,
+
+22:10.160 --> 22:16.640
+ and they become very friendly. But when the sighted people of the blind man's family come in,
+
+22:18.640 --> 22:26.000
+ you got a monster here. So it's very didactic in its way. And what I didn't know is that Mary Shelley
+
+22:26.000 --> 22:33.440
+ and Percy Shelley were great readers of the literature surrounding abolition in the United
+
+22:33.440 --> 22:41.200
+ States, the abolition of slavery. And they picked that up wholesale. You are making monsters of
+
+22:41.200 --> 22:45.680
+ these people because you won't give them the respect and love that they deserve.
+
+22:46.800 --> 22:54.880
+ Do you have, if we get philosophical for a second, do you worry that once we create
+
+22:54.880 --> 22:59.840
+ machines that are a little bit more intelligent? Let's look at Roomba, the vacuum cleaner,
+
+22:59.840 --> 23:05.360
+ that this darker part of human nature where we abuse
+
+23:07.760 --> 23:12.400
+ the other, somebody who's different, will come out?
+
+23:13.520 --> 23:22.640
+ I don't worry about it. I could imagine it happening. But I think that what AI has to offer
+
+23:22.640 --> 23:32.480
+ the human race will be so attractive that people will be won over. So you have looked deep into
+
+23:32.480 --> 23:40.080
+ these people, had deep conversations, and it's interesting to get a sense of stories of the
+
+23:40.080 --> 23:44.480
+ way they were thinking and the way it was changed, the way your own thinking about AI has changed.
+
+23:44.480 --> 23:53.360
+ As you mentioned, McCarthy, what about the years at CMU, Carnegie Mellon, with Joe?
+
+23:53.360 --> 24:02.800
+ Sure. Joe was not in AI. He was in algorithmic complexity.
+
+24:03.440 --> 24:09.040
+ Was there always a line between AI and computer science, for example? Is AI its own place of
+
+24:09.040 --> 24:15.920
+ outcasts? Was that the feeling? There was a kind of outcast period for AI.
+
+24:15.920 --> 24:28.720
+ For instance, in 1974, the new field was hardly 10 years old. The new field of computer science
+
+24:28.720 --> 24:33.200
+ was asked by the National Science Foundation, I believe, but it may have been the National
+
+24:33.200 --> 24:42.720
+ Academies, I can't remember, to tell our fellow scientists where computer science is and what
+
+24:42.720 --> 24:52.880
+ it means. And they wanted to leave out AI. And they only agreed to put it in because Don Knuth
+
+24:52.880 --> 24:59.760
+ said, hey, this is important. You can't just leave that out. Really? Don? Don Knuth, yes.
+
+24:59.760 --> 25:06.480
+ I talked to Mr. Nietzsche. Out of all the people. Yes. But you see, an AI person couldn't have made
+
+25:06.480 --> 25:10.880
+ that argument. He wouldn't have been believed, but Knuth was believed. Yes.
+
+25:10.880 --> 25:18.160
+ So Joe Trout worked on the real stuff. Joe was working on algorithmic complexity,
+
+25:18.160 --> 25:24.800
+ but he would say in plain English again and again, the smartest people I know are in AI.
+
+25:24.800 --> 25:32.320
+ Really? Oh, yes. No question. Anyway, Joe loved these guys. What happened was that
+
+25:34.080 --> 25:40.160
+ I guess it was as I started to write machines who think, Herb Simon and I became very close
+
+25:40.160 --> 25:46.000
+ friends. He would walk past our house on Northumberland Street every day after work.
+
+25:46.560 --> 25:52.160
+ And I would just be putting my cover on my typewriter and I would lean out the door and say,
+
+25:52.160 --> 25:58.800
+ Herb, would you like a sherry? And Herb almost always would like a sherry. So he'd stop in
+
+25:59.440 --> 26:06.000
+ and we'd talk for an hour, two hours. My journal says we talked this afternoon for three hours.
+
+26:06.720 --> 26:11.520
+ What was on his mind at the time in terms of on the AI side of things?
+
+26:12.160 --> 26:15.120
+ We didn't talk too much about AI. We talked about other things. Just life.
+
+26:15.120 --> 26:24.480
+ We both love literature and Herb had read Proust in the original French twice all the way through.
+
+26:25.280 --> 26:31.280
+ I can't. I read it in English in translation. So we talked about literature. We talked about
+
+26:31.280 --> 26:37.120
+ languages. We talked about music because he loved music. We talked about art because he was
+
+26:37.120 --> 26:45.840
+ he was actually enough of a painter that he had to give it up because he was afraid it was interfering
+
+26:45.840 --> 26:53.120
+ with his research and so on. So no, it was really just chat chat, but it was very warm.
+
+26:54.000 --> 26:59.840
+ So one summer I said to Herb, you know, my students have all the really interesting
+
+26:59.840 --> 27:04.480
+ conversations. I was teaching at the University of Pittsburgh then in the English department.
+
+27:04.480 --> 27:08.880
+ And, you know, they get to talk about the meaning of life and that kind of thing.
+
+27:08.880 --> 27:15.200
+ And what do I have? I have university meetings where we talk about the photocopying budget and,
+
+27:15.200 --> 27:20.160
+ you know, whether the course on romantic poetry should be one semester or two.
+
+27:21.200 --> 27:25.760
+ So Herb laughed. He said, yes, I know what you mean. He said, but, you know, you could do something
+
+27:25.760 --> 27:33.920
+ about that. Dot, that was his wife, Dot and I used to have a salon at the University of Chicago every
+
+27:33.920 --> 27:42.400
+ Sunday night. And we would have essentially an open house. And people knew it wasn't for a small
+
+27:42.400 --> 27:51.440
+ talk. It was really for some topic of depth. He said, but my advice would be that you choose
+
+27:51.440 --> 27:59.200
+ the topic ahead of time. Fine, I said. So the following, we exchanged mail over the summer.
+
+27:59.200 --> 28:09.120
+ That was US post in those days because you didn't have personal email. And I decided I would organize
+
+28:09.120 --> 28:16.880
+ it. And there would be eight of us, Alan Nolan, his wife, Herb Simon, and his wife, Dorothea.
+
+28:16.880 --> 28:27.040
+ There was a novelist in town, a man named Mark Harris. He had just arrived and his wife, Josephine.
+
+28:27.840 --> 28:33.040
+ Mark was most famous then for a novel called Bang the Drum Slowly, which was about baseball.
+
+28:34.160 --> 28:43.360
+ And Joe and me, so eight people. And we met monthly and we just sank our teeth into really
+
+28:43.360 --> 28:52.160
+ hard topics. And it was great fun. How have your own views around artificial intelligence changed
+
+28:53.200 --> 28:57.520
+ in through the process of writing machines who think and afterwards the ripple effects?
+
+28:58.240 --> 29:04.400
+ I was a little skeptical that this whole thing would work out. It didn't matter. To me, it was
+
+29:04.400 --> 29:16.160
+ so audacious. This whole thing being AI generally. And in some ways, it hasn't worked out the way I
+
+29:16.160 --> 29:25.760
+ expected so far. That is to say, there is this wonderful lot of apps, thanks to deep learning
+
+29:25.760 --> 29:35.680
+ and so on. But those are algorithmic. And in the part of symbolic processing,
+
+29:36.640 --> 29:45.600
+ there is very little yet. And that's a field that lies waiting for industrious graduate students.
+
+29:46.800 --> 29:53.040
+ Maybe you can tell me some figures that popped up in your life in the 80s with expert systems,
+
+29:53.040 --> 30:01.840
+ where there was the symbolic AI possibilities of what most people think of as AI. If you dream
+
+30:01.840 --> 30:08.000
+ of the possibilities of AI, it's really expert systems. And those hit a few walls and there
+
+30:08.000 --> 30:12.960
+ were challenges there. And I think, yes, they will reemerge again with some new breakthroughs and so
+
+30:12.960 --> 30:18.640
+ on. But what did that feel like, both the possibility and the winter that followed, the
+
+30:18.640 --> 30:25.520
+ slowdown in research? This whole thing about AI winter is, to me, a crock.
+
+30:26.160 --> 30:33.200
+ It's no winters. Because I look at the basic research that was being done in the 80s, which is
+
+30:33.200 --> 30:39.520
+ supposed to be, my God, it was really important. It was laying down things that nobody had thought
+
+30:39.520 --> 30:44.880
+ about before. But it was basic research. You couldn't monetize it. Hence the winter.
+
+30:44.880 --> 30:53.680
+ Science research goes and fits and starts. It isn't this nice, smooth,
+
+30:54.240 --> 30:59.200
+ oh, this follows this, follows this. No, it just doesn't work that way.
+
+30:59.200 --> 31:03.600
+ Well, the interesting thing, the way winters happen, it's never the fault of the researchers.
+
+31:04.480 --> 31:11.920
+ It's the some source of hype, over promising. Well, no, let me take that back. Sometimes it
+
+31:11.920 --> 31:17.200
+ is the fault of the researchers. Sometimes certain researchers might overpromise the
+
+31:17.200 --> 31:23.760
+ possibilities. They themselves believe that we're just a few years away, sort of just recently talked
+
+31:23.760 --> 31:30.240
+ to Elon Musk and he believes he'll have an autonomous vehicle in a year and he believes it.
+
+31:30.240 --> 31:33.520
+ A year? A year, yeah, would have mass deployment of a time.
+
+31:33.520 --> 31:38.640
+ For the record, this is 2019 right now. So he's talking 2020.
+
+31:38.640 --> 31:44.800
+ To do the impossible, you really have to believe it. And I think what's going to happen when you
+
+31:44.800 --> 31:49.520
+ believe it, because there's a lot of really brilliant people around him, is some good stuff
+
+31:49.520 --> 31:54.640
+ will come out of it. Some unexpected brilliant breakthroughs will come out of it. When you
+
+31:54.640 --> 31:59.520
+ really believe it, when you work that hard. I believe that and I believe autonomous vehicles
+
+31:59.520 --> 32:05.280
+ will come. I just don't believe it'll be in a year. I wish. But nevertheless, there's
+
+32:05.280 --> 32:11.680
+ autonomous vehicles is a good example. There's a feeling many companies have promised by 2021,
+
+32:11.680 --> 32:18.000
+ by 2022 for GM. Basically, every single automotive company has promised they'll
+
+32:18.000 --> 32:22.480
+ have autonomous vehicles. So that kind of overpromise is what leads to the winter.
+
+32:23.040 --> 32:28.320
+ Because we'll come to those dates, there won't be autonomous vehicles, and there'll be a feeling,
+
+32:28.960 --> 32:34.160
+ well, wait a minute, if we took your word at that time, that means we just spent billions of
+
+32:34.160 --> 32:41.600
+ dollars, had made no money. And there's a counter response to where everybody gives up on it.
+
+32:41.600 --> 32:49.600
+ Sort of intellectually, at every level, the hope just dies. And all that's left is a few basic
+
+32:49.600 --> 32:56.800
+ researchers. So you're uncomfortable with some aspects of this idea. Well, it's the difference
+
+32:56.800 --> 33:04.160
+ between science and commerce. So you think science, science goes on the way it does?
+
+33:06.480 --> 33:14.800
+ Science can really be killed by not getting proper funding or timely funding. I think
+
+33:14.800 --> 33:22.080
+ Great Britain was a perfect example of that. The Lighthill report in the 1960s.
+
+33:22.080 --> 33:27.360
+ The year essentially said, there's no use of Great Britain putting any money into this.
+
+33:27.360 --> 33:35.600
+ It's going nowhere. And this was all about social factions in Great Britain.
+
+33:36.960 --> 33:44.400
+ Edinburgh hated Cambridge, and Cambridge hated Manchester, and somebody else can write that
+
+33:44.400 --> 33:53.760
+ story. But it really did have a hard effect on research there. Now, they've come roaring back
+
+33:53.760 --> 34:01.360
+ with deep mind. But that's one guy and his visionaries around him.
+
+34:01.360 --> 34:08.320
+ But just to push on that, it's kind of interesting, you have this dislike of the idea of an AI winter.
+
+34:08.320 --> 34:15.440
+ Where's that coming from? Where were you? Oh, because I just don't think it's true.
+
+34:16.560 --> 34:21.360
+ There was a particular period of time. It's a romantic notion, certainly.
+
+34:21.360 --> 34:32.960
+ Yeah, well, I admire science, perhaps more than I admire commerce. Commerce is fine. Hey,
+
+34:32.960 --> 34:42.960
+ you know, we all got to live. But science has a much longer view than commerce,
+
+34:44.080 --> 34:54.000
+ and continues almost regardless. It can't continue totally regardless, but it almost
+
+34:54.000 --> 34:59.600
+ regardless of what's saleable and what's not, what's monetizable and what's not.
+
+34:59.600 --> 35:05.840
+ So the winter is just something that happens on the commerce side, and the science marches.
+
+35:07.200 --> 35:12.560
+ That's a beautifully optimistic inspired message. I agree with you. I think
+
+35:13.760 --> 35:19.440
+ if we look at the key people that work in AI, they work in key scientists in most disciplines,
+
+35:19.440 --> 35:25.360
+ they continue working out of the love for science. You can always scrape up some funding
+
+35:25.360 --> 35:33.120
+ to stay alive, and they continue working diligently. But there certainly is a huge
+
+35:33.120 --> 35:39.840
+ amount of funding now, and there's a concern on the AI side and deep learning. There's a concern
+
+35:39.840 --> 35:46.160
+ that we might, with over promising, hit another slowdown in funding, which does affect the number
+
+35:46.160 --> 35:51.280
+ of students, you know, that kind of thing. Yeah, it does. So the kind of ideas you had
+
+35:51.280 --> 35:56.400
+ in machines who think, did you continue that curiosity through the decades that followed?
+
+35:56.400 --> 36:04.800
+ Yes, I did. And what was your view, historical view of how AI community evolved, the conversations
+
+36:04.800 --> 36:11.520
+ about it, the work? Has it persisted the same way from its birth? No, of course not. It's just
+
+36:11.520 --> 36:21.360
+ we were just talking. The symbolic AI really kind of dried up and it all became algorithmic.
+
+36:22.400 --> 36:29.520
+ I remember a young AI student telling me what he was doing, and I had been away from the field
+
+36:29.520 --> 36:34.080
+ long enough. I'd gotten involved with complexity at the Santa Fe Institute.
+
+36:34.080 --> 36:40.960
+ I thought, algorithms, yeah, they're in the service of, but they're not the main event.
+
+36:41.680 --> 36:49.200
+ No, they became the main event. That surprised me. And we all know the downside of this. We
+
+36:49.200 --> 36:58.240
+ all know that if you're using an algorithm to make decisions based on a gazillion human decisions
+
+36:58.240 --> 37:05.040
+ baked into it, are all the mistakes that humans make, the bigotries, the short sightedness,
+
+37:06.000 --> 37:14.000
+ so on and so on. So you mentioned Santa Fe Institute. So you've written the novel Edge
+
+37:14.000 --> 37:21.200
+ of Chaos, but it's inspired by the ideas of complexity, a lot of which have been extensively
+
+37:21.200 --> 37:30.160
+ explored at the Santa Fe Institute. It's another fascinating topic of just sort of
+
+37:31.040 --> 37:37.600
+ emergent complexity from chaos. Nobody knows how it happens really, but it seems to wear
+
+37:37.600 --> 37:44.160
+ all the interesting stuff that does happen. So how do first, not your novel, but just
+
+37:44.160 --> 37:49.520
+ complexity in general in the work at Santa Fe fit into the bigger puzzle of the history of AI?
+
+37:49.520 --> 37:53.520
+ Or it may be even your personal journey through that.
+
+37:54.480 --> 38:03.040
+ One of the last projects I did concerning AI in particular was looking at the work of
+
+38:03.040 --> 38:12.960
+ Harold Cohen, the painter. And Harold was deeply involved with AI. He was a painter first.
+
+38:12.960 --> 38:27.600
+ And what his project, Aaron, which was a lifelong project, did, was reflect his own cognitive
+
+38:27.600 --> 38:34.800
+ processes. Okay. Harold and I, even though I wrote a book about it, we had a lot of friction between
+
+38:34.800 --> 38:44.560
+ us. And I went, I thought, this is it, you know, the book died. It was published and fell into a
+
+38:44.560 --> 38:53.040
+ ditch. This is it. I'm finished. It's time for me to do something different. By chance,
+
+38:53.040 --> 38:59.280
+ this was a sabbatical year for my husband. And we spent two months at the Santa Fe Institute
+
+38:59.280 --> 39:09.200
+ and two months at Caltech. And then the spring semester in Munich, Germany. Okay. Those two
+
+39:09.200 --> 39:19.520
+ months at the Santa Fe Institute were so restorative for me. And I began to, the institute was very
+
+39:19.520 --> 39:26.240
+ small then. It was in some kind of office complex on old Santa Fe trail. Everybody kept their door
+
+39:26.240 --> 39:33.440
+ open. So you could crack your head on a problem. And if you finally didn't get it, you could walk
+
+39:33.440 --> 39:42.480
+ in to see Stuart Kaufman or any number of people and say, I don't get this. Can you explain?
+
+39:43.680 --> 39:51.120
+ And one of the people that I was talking to about complex adaptive systems was Murray Gellemann.
+
+39:51.120 --> 39:58.960
+ And I told Murray what Harold Cohen had done. And I said, you know, this sounds to me
+
+39:58.960 --> 40:06.080
+ like a complex adaptive system. And he said, yeah, it is. Well, what do you know? Harold's
+
+40:06.080 --> 40:12.560
+ Aaron had all these kissing cousins all over the world in science and in economics and so on and
+
+40:12.560 --> 40:21.200
+ so forth. I was so relieved. I thought, okay, your instincts are okay. You're doing the right thing. I
+
+40:21.200 --> 40:25.920
+ didn't have the vocabulary. And that was one of the things that the Santa Fe Institute gave me.
+
+40:25.920 --> 40:31.040
+ If I could have rewritten that book, no, it had just come out. I couldn't rewrite it. I would have
+
+40:31.040 --> 40:37.680
+ had a vocabulary to explain what Aaron was doing. Okay. So I got really interested in
+
+40:37.680 --> 40:47.440
+ what was going on at the Institute. The people were again, bright and funny and willing to explain
+
+40:47.440 --> 40:54.800
+ anything to this amateur. George Cowan, who was then the head of the Institute, said he thought it
+
+40:54.800 --> 41:02.160
+ might be a nice idea if I wrote a book about the Institute. And I thought about it. And I had my
+
+41:02.160 --> 41:08.960
+ eye on some other project. God knows what. And I said, oh, I'm sorry, George. Yeah, I'd really love
+
+41:08.960 --> 41:13.840
+ to do it. But, you know, just not going to work for me at this moment. And he said, oh, too bad.
+
+41:13.840 --> 41:18.560
+ I think it would make an interesting book. Well, he was right and I was wrong. I wish I'd done it.
+
+41:18.560 --> 41:24.080
+ But that's interesting. I hadn't thought about that, that that was a road not taken that I wish
+
+41:24.080 --> 41:32.400
+ I'd taken. Well, you know what? That's just on that point. It's quite brave for you as a writer,
+
+41:32.400 --> 41:39.680
+ as sort of coming from a world of literature, the literary thinking and historical thinking. I mean,
+
+41:39.680 --> 41:52.640
+ just from that world and bravely talking to quite, I assume, large egos in AI or in complexity and so
+
+41:52.640 --> 42:00.560
+ on. How'd you do it? Like, where did you? I mean, I suppose they could be intimidated of you as well,
+
+42:00.560 --> 42:06.160
+ because it's two different worlds. I never picked up that anybody was intimidated by me.
+
+42:06.160 --> 42:11.040
+ But how were you brave enough? Where did you find the guts to just dumb, dumb luck? I mean,
+
+42:11.680 --> 42:16.160
+ this is an interesting rock to turn over. I'm going to write a book about and you know,
+
+42:16.160 --> 42:21.840
+ people have enough patience with writers, if they think they're going to end up at a book
+
+42:21.840 --> 42:27.840
+ that they let you flail around and so on. It's well, but they also look if the writer has.
+
+42:27.840 --> 42:33.440
+ There's like, if there's a sparkle in their eye, if they get it. Yeah, sure. Right. When were you
+
+42:33.440 --> 42:44.480
+ at the Santa Fe Institute? The time I'm talking about is 1990. Yeah, 1990, 1991, 1992. But we then,
+
+42:44.480 --> 42:49.920
+ because Joe was an external faculty member, we're in Santa Fe every summer, we bought a house there.
+
+42:49.920 --> 42:55.600
+ And I didn't have that much to do with the Institute anymore. I was writing my novels,
+
+42:55.600 --> 43:04.320
+ I was doing whatever I was doing. But I loved the Institute and I loved
+
+43:06.720 --> 43:12.960
+ the, again, the audacity of the ideas. That really appeals to me.
+
+43:12.960 --> 43:22.160
+ I think that there's this feeling, much like in great institutes of neuroscience, for example,
+
+43:23.040 --> 43:29.840
+ that they're in it for the long game of understanding something fundamental about
+
+43:29.840 --> 43:36.800
+ reality and nature. And that's really exciting. So if we start not to look a little bit more recently,
+
+43:36.800 --> 43:49.280
+ how AI is really popular today. How is this world, you mentioned algorithmic, but in general,
+
+43:50.080 --> 43:54.320
+ is the spirit of the people, the kind of conversations you hear through the grapevine
+
+43:54.320 --> 44:00.160
+ and so on, is that different than the roots that you remember? No, the same kind of excitement,
+
+44:00.160 --> 44:07.120
+ the same kind of, this is really going to make a difference in the world. And it will, it has.
+
+44:07.120 --> 44:13.280
+ A lot of folks, especially young, 20 years old or something, they think we've just found something
+
+44:14.080 --> 44:21.040
+ special here. We're going to change the world tomorrow. On a time scale, do you have
+
+44:22.000 --> 44:27.120
+ a sense of what, of the time scale at which breakthroughs in AI happen?
+
+44:27.120 --> 44:39.040
+ I really don't, because look at deep learning. That was, Jeffrey Hinton came up with the algorithm
+
+44:39.920 --> 44:48.960
+ in 86. But it took all these years for the technology to be good enough to actually
+
+44:48.960 --> 44:57.680
+ be applicable. So no, I can't predict that at all. I can't, I wouldn't even try.
+
+44:58.320 --> 45:05.440
+ Well, let me ask you to, not to try to predict, but to speak to the, I'm sure in the 60s,
+
+45:06.000 --> 45:10.320
+ as it continues now, there's people that think, let's call it, we can call it this
+
+45:11.040 --> 45:17.120
+ fun word, the singularity. When there's a phase shift, there's some profound feeling where
+
+45:17.120 --> 45:23.040
+ we're all really surprised by what's able to be achieved. I'm sure those dreams are there.
+
+45:23.040 --> 45:29.200
+ I remember reading quotes in the 60s and those continued. How have your own views,
+
+45:29.200 --> 45:34.960
+ maybe if you look back, about the timeline of a singularity changed?
+
+45:37.040 --> 45:45.760
+ Well, I'm not a big fan of the singularity as Ray Kurzweil has presented it.
+
+45:45.760 --> 45:52.480
+ But how would you define the Ray Kurzweil? How would you, how do you think of singularity in
+
+45:52.480 --> 45:58.880
+ those? If I understand Kurzweil's view, it's sort of, there's going to be this moment when
+
+45:58.880 --> 46:06.320
+ machines are smarter than humans and, you know, game over. However, the game over is,
+
+46:06.320 --> 46:10.800
+ I mean, do they put us on a reservation? Do they, et cetera, et cetera. And
+
+46:10.800 --> 46:19.840
+ first of all, machines are smarter than humans in some ways, all over the place. And they have been
+
+46:19.840 --> 46:27.280
+ since adding machines were invented. So it's not, it's not going to come like some great
+
+46:27.280 --> 46:34.320
+ edible crossroads, you know, where they meet each other and our offspring, Oedipus says,
+
+46:34.320 --> 46:41.040
+ you're dead. It's just not going to happen. Yeah. So it's already game over with calculators,
+
+46:41.040 --> 46:47.920
+ right? They're already out to do much better at basic arithmetic than us. But, you know,
+
+46:47.920 --> 46:55.840
+ there's a human like intelligence. And that's not the ones that destroy us. But, you know,
+
+46:55.840 --> 47:01.520
+ somebody that you can have as a, as a friend, you can have deep connections with that kind of
+
+47:01.520 --> 47:07.680
+ passing the Turing test and beyond, those kinds of ideas. Have you dreamt of those?
+
+47:07.680 --> 47:08.880
+ Oh, yes, yes, yes.
+
+47:08.880 --> 47:10.160
+ Those possibilities.
+
+47:10.160 --> 47:17.520
+ In a book I wrote with Ed Feigenbaum, there's a little story called the geriatric robot. And
+
+47:18.880 --> 47:24.240
+ how I came up with the geriatric robot is a story in itself. But here's, here's what the
+
+47:24.240 --> 47:29.520
+ geriatric robot does. It doesn't just clean you up and feed you and will you out into the sun.
+
+47:29.520 --> 47:42.720
+ It's great advantages. It listens. It says, tell me again about the great coup of 73.
+
+47:43.520 --> 47:52.080
+ Tell me again about how awful or how wonderful your grandchildren are and so on and so forth.
+
+47:53.040 --> 47:59.440
+ And it isn't hanging around to inherit your money. It isn't hanging around because it can't get
+
+47:59.440 --> 48:08.320
+ any other job. This is its job and so on and so forth. Well, I would love something like that.
+
+48:09.120 --> 48:15.600
+ Yeah. I mean, for me, that deeply excites me. So I think there's a lot of us.
+
+48:15.600 --> 48:20.880
+ Lex, you got to know it was a joke. I dreamed it up because I needed to talk to college students
+
+48:20.880 --> 48:26.880
+ and I needed to give them some idea of what AI might be. And they were rolling in the aisles
+
+48:26.880 --> 48:32.160
+ as I elaborated and elaborated and elaborated. When it went into the book,
+
+48:34.320 --> 48:40.240
+ they took my hide off in the New York Review of Books. This is just what we've thought about
+
+48:40.240 --> 48:44.400
+ these people in AI. They're inhuman. Oh, come on. Get over it.
+
+48:45.120 --> 48:49.120
+ Don't you think that's a good thing for the world that AI could potentially...
+
+48:49.120 --> 48:58.800
+ Why? I do. Absolutely. And furthermore, I want... I'm pushing 80 now. By the time I need help
+
+48:59.360 --> 49:04.160
+ like that, I also want it to roll itself in a corner and shut the fuck up.
+
+49:06.960 --> 49:09.920
+ Let me linger on that point. Do you really, though?
+
+49:09.920 --> 49:11.040
+ Yeah, I do. Here's what.
+
+49:11.040 --> 49:15.120
+ But you wanted to push back a little bit a little.
+
+49:15.120 --> 49:21.520
+ But I have watched my friends go through the whole issue around having help in the house.
+
+49:22.480 --> 49:29.760
+ And some of them have been very lucky and had fabulous help. And some of them have had people
+
+49:29.760 --> 49:35.120
+ in the house who want to keep the television going on all day, who want to talk on their phones all day.
+
+49:35.760 --> 49:38.960
+ No. So basically... Just roll yourself in the corner.
+
+49:38.960 --> 49:45.760
+ Unfortunately, us humans, when we're assistants, we care... We're still...
+
+49:45.760 --> 49:48.400
+ Even when we're assisting others, we care about ourselves more.
+
+49:48.400 --> 49:49.280
+ Of course.
+
+49:49.280 --> 49:56.800
+ And so you create more frustration. And a robot AI assistant can really optimize
+
+49:57.360 --> 50:03.040
+ the experience for you. I was just speaking to the point... You actually bring up a very,
+
+50:03.040 --> 50:07.120
+ very good point. But I was speaking to the fact that us humans are a little complicated,
+
+50:07.120 --> 50:14.560
+ that we don't necessarily want a perfect servant. I don't... Maybe you disagree with that.
+
+50:14.560 --> 50:21.360
+ But there's... I think there's a push and pull with humans.
+
+50:22.240 --> 50:28.080
+ A little tension, a little mystery that, of course, that's really difficult for you to get right.
+
+50:28.080 --> 50:35.120
+ But I do sense, especially in today with social media, that people are getting more and more
+
+50:35.120 --> 50:42.000
+ lonely, even young folks. And sometimes, especially young folks, that loneliness,
+
+50:42.000 --> 50:47.760
+ there's a longing for connection. And AI can help alleviate some of that loneliness.
+
+50:48.560 --> 50:52.880
+ Some. Just somebody who listens. Like in person.
+
+50:54.640 --> 50:54.960
+ That...
+
+50:54.960 --> 50:55.680
+ So to speak.
+
+50:56.240 --> 51:00.080
+ So to speak, yeah. So to speak.
+
+51:00.080 --> 51:07.120
+ Yeah, that to me is really exciting. But so if we look at that level of intelligence,
+
+51:07.120 --> 51:12.560
+ which is exceptionally difficult to achieve, actually, as the singularity, or whatever,
+
+51:12.560 --> 51:18.320
+ that's the human level bar, that people have dreamt of that too. Touring dreamt of it.
+
+51:19.520 --> 51:26.320
+ He had a date timeline. Do you have how of your own timeline evolved on past time?
+
+51:26.320 --> 51:27.520
+ I don't even think about it.
+
+51:27.520 --> 51:28.240
+ You don't even think?
+
+51:28.240 --> 51:35.520
+ No. Just this field has been so full of surprises for me.
+
+51:35.520 --> 51:39.120
+ That you're just taking in and see a fun bunch of basic science?
+
+51:39.120 --> 51:45.840
+ Yeah, I just can't. Maybe that's because I've been around the field long enough to think,
+
+51:45.840 --> 51:52.240
+ you know, don't go that way. Herb Simon was terrible about making these predictions of
+
+51:52.240 --> 51:53.840
+ when this and that would happen.
+
+51:53.840 --> 51:54.320
+ Right.
+
+51:54.320 --> 51:58.400
+ And he was a sensible guy.
+
+51:58.400 --> 52:01.600
+ Yeah. And his quotes are often used, right, as a...
+
+52:01.600 --> 52:02.880
+ That's a legend, yeah.
+
+52:02.880 --> 52:13.840
+ Yeah. Do you have concerns about AI, the existential threats that many people like Elon Musk and
+
+52:13.840 --> 52:16.160
+ Sam Harris and others that are thinking about?
+
+52:16.160 --> 52:21.200
+ Oh, yeah, yeah. That takes up a half a chapter in my book.
+
+52:21.200 --> 52:27.120
+ I call it the male gaze.
+
+52:27.120 --> 52:35.200
+ Well, you hear me out. The male gaze is actually a term from film criticism.
+
+52:36.240 --> 52:40.640
+ And I'm blocking on the woman who dreamed this up.
+
+52:41.280 --> 52:48.880
+ But she pointed out how most movies were made from the male point of view, that women were
+
+52:48.880 --> 52:55.520
+ objects, not subjects. They didn't have any agency and so on and so forth.
+
+52:56.080 --> 53:00.560
+ So when Elon and his pals, Hawking and so on,
+
+53:00.560 --> 53:05.760
+ okay, AI is going to eat our lunch and our dinner and our midnight snack too,
+
+53:06.640 --> 53:11.600
+ I thought, what? And I said to Ed Feigenbaum, oh, this is the first guy.
+
+53:11.600 --> 53:14.880
+ First, these guys have always been the smartest guy on the block.
+
+53:14.880 --> 53:20.880
+ And here comes something that might be smarter. Ooh, let's stamp it out before it takes over.
+
+53:20.880 --> 53:23.360
+ And Ed laughed. He said, I didn't think about it that way.
+
+53:24.080 --> 53:30.320
+ But I did. I did. And it is the male gaze.
+
+53:32.000 --> 53:37.120
+ Okay, suppose these things do have agency. Well, let's wait and see what happens.
+
+53:37.120 --> 53:47.040
+ Can we imbue them with ethics? Can we imbue them with a sense of empathy?
+
+53:48.560 --> 53:54.960
+ Or are they just going to be, I don't know, we've had centuries of guys like that?
+
+53:55.760 --> 54:03.600
+ That's interesting that the ego, the male gaze is immediately threatened.
+
+54:03.600 --> 54:12.320
+ And so you can't think in a patient, calm way of how the tech could evolve.
+
+54:13.280 --> 54:21.520
+ Speaking of which, here in 96 book, The Future of Women, I think at the time and now, certainly now,
+
+54:21.520 --> 54:27.760
+ I mean, I'm sorry, maybe at the time, but I'm more cognizant of now, is extremely relevant.
+
+54:27.760 --> 54:34.160
+ And you and Nancy Ramsey talk about four possible futures of women in science and tech.
+
+54:35.120 --> 54:42.400
+ So if we look at the decades before and after the book was released, can you tell a history,
+
+54:43.120 --> 54:50.560
+ sorry, of women in science and tech and how it has evolved? How have things changed? Where do we
+
+54:50.560 --> 55:01.520
+ stand? Not enough. They have not changed enough. The way that women are ground down in computing is
+
+55:02.320 --> 55:09.680
+ simply unbelievable. But what are the four possible futures for women in tech from the book?
+
+55:10.800 --> 55:16.720
+ What you're really looking at are various aspects of the present. So for each of those,
+
+55:16.720 --> 55:22.800
+ you could say, oh, yeah, we do have backlash. Look at what's happening with abortion and so on and so
+
+55:22.800 --> 55:31.280
+ forth. We have one step forward, one step back. The golden age of equality was the hardest chapter
+
+55:31.280 --> 55:37.040
+ to write. And I used something from the Santa Fe Institute, which is the sand pile effect,
+
+55:37.760 --> 55:44.480
+ that you drop sand very slowly onto a pile, and it grows and it grows and it grows until
+
+55:44.480 --> 55:56.640
+ suddenly it just breaks apart. And in a way, MeToo has done that. That was the last drop of sand
+
+55:56.640 --> 56:02.800
+ that broke everything apart. That was a perfect example of the sand pile effect. And that made
+
+56:02.800 --> 56:07.440
+ me feel good. It didn't change all of society, but it really woke a lot of people up.
+
+56:07.440 --> 56:15.760
+ But are you in general optimistic about maybe after MeToo? MeToo is about a very specific kind
+
+56:15.760 --> 56:21.920
+ of thing. Boy, solve that and you'll solve everything. But are you in general optimistic
+
+56:21.920 --> 56:27.600
+ about the future? Yes, I'm a congenital optimistic. I can't help it.
+
+56:28.400 --> 56:35.600
+ What about AI? What are your thoughts about the future of AI? Of course, I get asked,
+
+56:35.600 --> 56:41.280
+ what do you worry about? And the one thing I worry about is the things we can't anticipate.
+
+56:44.320 --> 56:47.440
+ There's going to be something out of that field that we will just say,
+
+56:47.440 --> 56:59.040
+ we weren't prepared for that. I am generally optimistic. When I first took up being interested
+
+56:59.040 --> 57:06.560
+ in AI, like most people in the field, more intelligence was like more virtue. What could be
+
+57:06.560 --> 57:15.760
+ bad? And in a way, I still believe that, but I realize that my notion of intelligence has
+
+57:15.760 --> 57:22.240
+ broadened. There are many kinds of intelligence, and we need to imbue our machines with those many
+
+57:22.240 --> 57:32.720
+ kinds. So you've now just finished or in the process of finishing the book, even working on
+
+57:32.720 --> 57:40.160
+ the memoir. How have you changed? I know it's just writing, but how have you changed the process?
+
+57:40.800 --> 57:48.880
+ If you look back, what kind of stuff did it bring up to you that surprised you looking at the entirety
+
+57:48.880 --> 58:01.200
+ of it all? The biggest thing, and it really wasn't a surprise, is how lucky I was, oh my, to be,
+
+58:03.680 --> 58:08.880
+ to have access to the beginning of a scientific field that is going to change the world.
+
+58:08.880 --> 58:20.880
+ How did I luck out? And yes, of course, my view of things has widened a lot.
+
+58:23.040 --> 58:32.080
+ If I can get back to one feminist part of our conversation, without knowing it, it really
+
+58:32.080 --> 58:40.400
+ was subconscious. I wanted AI to succeed because I was so tired of hearing that intelligence was
+
+58:40.400 --> 58:47.920
+ inside the male cranium. And I thought if there was something out there that wasn't a male
+
+58:49.360 --> 58:57.120
+ thinking and doing well, then that would put a lie to this whole notion of intelligence resides
+
+58:57.120 --> 59:04.560
+ in the male cranium. I did not know that until one night, Harold Cohen and I were
+
+59:05.760 --> 59:12.560
+ having a glass of wine, maybe two, and he said, what drew you to AI? And I said, oh, you know,
+
+59:12.560 --> 59:18.080
+ smartest people I knew, great project, blah, blah, blah. And I said, and I wanted something
+
+59:18.080 --> 59:30.640
+ besides male smarts. And it just bubbled up out of me, Lex. It's brilliant, actually. So AI really
+
+59:32.160 --> 59:36.080
+ humbles all of us and humbles the people that need to be humbled the most.
+
+59:37.040 --> 59:43.440
+ Let's hope. Oh, wow, that is so beautiful. Pamela, thank you so much for talking to
+
+59:43.440 --> 59:49.440
+ us. Oh, it's been a great pleasure. Thank you.
+

vtt/episode_037_large.vtt

@@ -0,0 +1,3590 @@
+WEBVTT
+
+00:00.000 --> 00:03.080
+ The following is a conversation with Vijay Kumar.
+
+00:03.080 --> 00:05.760
+ He's one of the top roboticists in the world,
+
+00:05.760 --> 00:08.760
+ a professor at the University of Pennsylvania,
+
+00:08.760 --> 00:12.880
+ a dean of pen engineering, former director of Grasp Lab,
+
+00:12.880 --> 00:15.300
+ or the General Robotics Automation Sensing
+
+00:15.300 --> 00:17.560
+ and Perception Laboratory at Penn,
+
+00:17.560 --> 00:22.560
+ that was established back in 1979, that's 40 years ago.
+
+00:22.600 --> 00:25.280
+ Vijay is perhaps best known for his work
+
+00:25.280 --> 00:28.520
+ in multi robot systems, robot swarms,
+
+00:28.520 --> 00:30.880
+ and micro aerial vehicles,
+
+00:30.880 --> 00:34.020
+ robots that elegantly cooperate in flight
+
+00:34.020 --> 00:36.200
+ under all the uncertainty and challenges
+
+00:36.200 --> 00:38.760
+ that the real world conditions present.
+
+00:38.760 --> 00:41.920
+ This is the Artificial Intelligence Podcast.
+
+00:41.920 --> 00:44.320
+ If you enjoy it, subscribe on YouTube,
+
+00:44.320 --> 00:47.560
+ give it five stars on iTunes, support on Patreon,
+
+00:47.560 --> 00:49.500
+ or simply connect with me on Twitter
+
+00:49.500 --> 00:53.280
+ at Lex Friedman, spelled F R I D M A N.
+
+00:53.280 --> 00:58.280
+ And now, here's my conversation with Vijay Kumar.
+
+00:58.700 --> 01:01.080
+ What is the first robot you've ever built
+
+01:01.080 --> 01:02.840
+ or were a part of building?
+
+01:02.840 --> 01:04.760
+ Way back when I was in graduate school,
+
+01:04.760 --> 01:06.760
+ I was part of a fairly big project
+
+01:06.760 --> 01:11.760
+ that involved building a very large hexapod.
+
+01:12.040 --> 01:16.700
+ It's weighed close to 7,000 pounds,
+
+01:17.520 --> 01:21.620
+ and it was powered by hydraulic actuation,
+
+01:21.620 --> 01:26.620
+ or it was actuated by hydraulics with 18 motors,
+
+01:27.720 --> 01:32.720
+ hydraulic motors, each controlled by an Intel 8085 processor
+
+01:34.160 --> 01:36.680
+ and an 8086 co processor.
+
+01:38.120 --> 01:43.120
+ And so imagine this huge monster that had 18 joints,
+
+01:44.800 --> 01:46.960
+ each controlled by an independent computer,
+
+01:46.960 --> 01:49.320
+ and there was a 19th computer that actually did
+
+01:49.320 --> 01:52.320
+ the coordination between these 18 joints.
+
+01:52.320 --> 01:53.720
+ So I was part of this project,
+
+01:53.720 --> 01:58.720
+ and my thesis work was how do you coordinate the 18 legs?
+
+02:02.080 --> 02:06.320
+ And in particular, the pressures in the hydraulic cylinders
+
+02:06.320 --> 02:09.200
+ to get efficient locomotion.
+
+02:09.200 --> 02:11.640
+ It sounds like a giant mess.
+
+02:11.640 --> 02:14.440
+ So how difficult is it to make all the motors communicate?
+
+02:14.440 --> 02:17.600
+ Presumably, you have to send signals hundreds of times
+
+02:17.600 --> 02:18.440
+ a second, or at least.
+
+02:18.440 --> 02:19.880
+ So this was not my work,
+
+02:19.880 --> 02:23.960
+ but the folks who worked on this wrote what I believe
+
+02:23.960 --> 02:26.640
+ to be the first multiprocessor operating system.
+
+02:26.640 --> 02:30.320
+ This was in the 80s, and you had to make sure
+
+02:30.320 --> 02:32.800
+ that obviously messages got across
+
+02:32.800 --> 02:34.640
+ from one joint to another.
+
+02:34.640 --> 02:37.960
+ You have to remember the clock speeds on those computers
+
+02:37.960 --> 02:39.660
+ were about half a megahertz.
+
+02:39.660 --> 02:42.180
+ Right, the 80s.
+
+02:42.180 --> 02:45.320
+ So not to romanticize the notion,
+
+02:45.320 --> 02:49.700
+ but how did it make you feel to see that robot move?
+
+02:51.080 --> 02:52.280
+ It was amazing.
+
+02:52.280 --> 02:55.280
+ In hindsight, it looks like, well, we built this thing
+
+02:55.280 --> 02:57.320
+ which really should have been much smaller.
+
+02:57.320 --> 02:59.160
+ And of course, today's robots are much smaller.
+
+02:59.160 --> 03:03.120
+ You look at Boston Dynamics or Ghost Robotics,
+
+03:03.120 --> 03:04.780
+ a spinoff from Penn.
+
+03:06.080 --> 03:10.080
+ But back then, you were stuck with the substrate you had,
+
+03:10.080 --> 03:13.720
+ the compute you had, so things were unnecessarily big.
+
+03:13.720 --> 03:18.040
+ But at the same time, and this is just human psychology,
+
+03:18.040 --> 03:20.400
+ somehow bigger means grander.
+
+03:21.600 --> 03:23.640
+ People never had the same appreciation
+
+03:23.640 --> 03:26.360
+ for nanotechnology or nanodevices
+
+03:26.360 --> 03:30.160
+ as they do for the Space Shuttle or the Boeing 747.
+
+03:30.160 --> 03:32.760
+ Yeah, you've actually done quite a good job
+
+03:32.760 --> 03:36.000
+ at illustrating that small is beautiful
+
+03:36.000 --> 03:37.760
+ in terms of robotics.
+
+03:37.760 --> 03:42.600
+ So what is on that topic is the most beautiful
+
+03:42.600 --> 03:46.200
+ or elegant robot in motion that you've ever seen?
+
+03:46.200 --> 03:47.880
+ Not to pick favorites or whatever,
+
+03:47.880 --> 03:51.000
+ but something that just inspires you that you remember.
+
+03:51.000 --> 03:54.000
+ Well, I think the thing that I'm most proud of
+
+03:54.000 --> 03:57.200
+ that my students have done is really think about
+
+03:57.200 --> 04:00.360
+ small UAVs that can maneuver in constrained spaces
+
+04:00.360 --> 04:03.640
+ and in particular, their ability to coordinate
+
+04:03.640 --> 04:06.760
+ with each other and form three dimensional patterns.
+
+04:06.760 --> 04:08.920
+ So once you can do that,
+
+04:08.920 --> 04:13.920
+ you can essentially create 3D objects in the sky
+
+04:14.960 --> 04:17.680
+ and you can deform these objects on the fly.
+
+04:17.680 --> 04:21.560
+ So in some sense, your toolbox of what you can create
+
+04:21.560 --> 04:23.400
+ has suddenly got enhanced.
+
+04:25.240 --> 04:27.800
+ And before that, we did the two dimensional version of this.
+
+04:27.800 --> 04:31.680
+ So we had ground robots forming patterns and so on.
+
+04:31.680 --> 04:34.960
+ So that was not as impressive, that was not as beautiful.
+
+04:34.960 --> 04:36.560
+ But if you do it in 3D,
+
+04:36.560 --> 04:40.240
+ suspended in midair, and you've got to go back to 2011
+
+04:40.240 --> 04:43.040
+ when we did this, now it's actually pretty standard
+
+04:43.040 --> 04:45.600
+ to do these things eight years later.
+
+04:45.600 --> 04:47.680
+ But back then it was a big accomplishment.
+
+04:47.680 --> 04:50.280
+ So the distributed cooperation
+
+04:50.280 --> 04:53.480
+ is where beauty emerges in your eyes?
+
+04:53.480 --> 04:55.800
+ Well, I think beauty to an engineer is very different
+
+04:55.800 --> 04:59.400
+ from beauty to someone who's looking at robots
+
+04:59.400 --> 05:01.240
+ from the outside, if you will.
+
+05:01.240 --> 05:04.800
+ But what I meant there, so before we said that grand,
+
+05:04.800 --> 05:09.800
+ so before we said that grand is associated with size.
+
+05:10.520 --> 05:13.720
+ And another way of thinking about this
+
+05:13.720 --> 05:15.600
+ is just the physical shape
+
+05:15.600 --> 05:18.400
+ and the idea that you can get physical shapes in midair
+
+05:18.400 --> 05:21.560
+ and have them deform, that's beautiful.
+
+05:21.560 --> 05:23.040
+ But the individual components,
+
+05:23.040 --> 05:24.880
+ the agility is beautiful too, right?
+
+05:24.880 --> 05:25.720
+ That is true too.
+
+05:25.720 --> 05:28.480
+ So then how quickly can you actually manipulate
+
+05:28.480 --> 05:29.560
+ these three dimensional shapes
+
+05:29.560 --> 05:31.280
+ and the individual components?
+
+05:31.280 --> 05:32.240
+ Yes, you're right.
+
+05:32.240 --> 05:36.760
+ But by the way, you said UAV, unmanned aerial vehicle.
+
+05:36.760 --> 05:41.760
+ What's a good term for drones, UAVs, quad copters?
+
+05:41.840 --> 05:44.560
+ Is there a term that's being standardized?
+
+05:44.560 --> 05:45.440
+ I don't know if there is.
+
+05:45.440 --> 05:47.920
+ Everybody wants to use the word drones.
+
+05:47.920 --> 05:51.080
+ And I've often said this, drones to me is a pejorative word.
+
+05:51.080 --> 05:53.960
+ It signifies something that's dumb,
+
+05:53.960 --> 05:56.360
+ that's pre programmed, that does one little thing
+
+05:56.360 --> 05:58.600
+ and robots are anything but drones.
+
+05:58.600 --> 06:00.680
+ So I actually don't like that word,
+
+06:00.680 --> 06:02.960
+ but that's what everybody uses.
+
+06:02.960 --> 06:04.880
+ You could call it unpiloted.
+
+06:04.880 --> 06:05.800
+ Unpiloted.
+
+06:05.800 --> 06:08.120
+ But even unpiloted could be radio controlled,
+
+06:08.120 --> 06:11.560
+ could be remotely controlled in many different ways.
+
+06:11.560 --> 06:12.960
+ And I think the right word is,
+
+06:12.960 --> 06:15.040
+ thinking about it as an aerial robot.
+
+06:15.040 --> 06:19.080
+ You also say agile, autonomous, aerial robot, right?
+
+06:19.080 --> 06:22.160
+ Yeah, so agility is an attribute, but they don't have to be.
+
+06:23.080 --> 06:24.800
+ So what biological system,
+
+06:24.800 --> 06:27.200
+ because you've also drawn a lot of inspiration with those.
+
+06:27.200 --> 06:30.360
+ I've seen bees and ants that you've talked about.
+
+06:30.360 --> 06:35.240
+ What living creatures have you found to be most inspiring
+
+06:35.240 --> 06:38.520
+ as an engineer, instructive in your work in robotics?
+
+06:38.520 --> 06:43.440
+ To me, so ants are really quite incredible creatures, right?
+
+06:43.440 --> 06:47.880
+ So you, I mean, the individuals arguably are very simple
+
+06:47.880 --> 06:52.360
+ in how they're built and yet they're incredibly resilient
+
+06:52.360 --> 06:53.960
+ as a population.
+
+06:53.960 --> 06:56.760
+ And as individuals, they're incredibly robust.
+
+06:56.760 --> 07:00.600
+ So, if you take an ant, it's six legs,
+
+07:00.600 --> 07:04.120
+ you remove one leg, it still works just fine.
+
+07:04.120 --> 07:05.760
+ And it moves along.
+
+07:05.760 --> 07:08.720
+ And I don't know that he even realizes it's lost a leg.
+
+07:09.760 --> 07:12.520
+ So that's the robustness at the individual ant level.
+
+07:13.400 --> 07:15.360
+ But then you look about this instinct
+
+07:15.360 --> 07:17.680
+ for self preservation of the colonies
+
+07:17.680 --> 07:20.400
+ and they adapt in so many amazing ways.
+
+07:20.400 --> 07:25.400
+ You know, transcending gaps by just chaining themselves
+
+07:26.800 --> 07:29.600
+ together when you have a flood,
+
+07:29.600 --> 07:32.360
+ being able to recruit other teammates
+
+07:32.360 --> 07:34.320
+ to carry big morsels of food,
+
+07:35.760 --> 07:38.760
+ and then going out in different directions looking for food,
+
+07:38.760 --> 07:43.160
+ and then being able to demonstrate consensus,
+
+07:43.160 --> 07:47.040
+ even though they don't communicate directly with each other
+
+07:47.040 --> 07:49.080
+ the way we communicate with each other.
+
+07:49.080 --> 07:51.880
+ In some sense, they also know how to do democracy,
+
+07:51.880 --> 07:53.640
+ probably better than what we do.
+
+07:53.640 --> 07:57.000
+ Yeah, somehow it's even democracy is emergent.
+
+07:57.000 --> 07:59.120
+ It seems like all of the phenomena that we see
+
+07:59.120 --> 08:00.480
+ is all emergent.
+
+08:00.480 --> 08:03.560
+ It seems like there's no centralized communicator.
+
+08:03.560 --> 08:06.520
+ There is, so I think a lot is made about that word,
+
+08:06.520 --> 08:09.640
+ emergent, and it means lots of things to different people.
+
+08:09.640 --> 08:10.680
+ But you're absolutely right.
+
+08:10.680 --> 08:13.040
+ I think as an engineer, you think about
+
+08:13.040 --> 08:17.720
+ what element, elemental behaviors
+
+08:17.720 --> 08:21.320
+ were primitives you could synthesize
+
+08:21.320 --> 08:25.240
+ so that the whole looks incredibly powerful,
+
+08:25.240 --> 08:26.520
+ incredibly synergistic,
+
+08:26.520 --> 08:29.520
+ the whole definitely being greater than some of the parts,
+
+08:29.520 --> 08:31.480
+ and ants are living proof of that.
+
+08:32.480 --> 08:34.960
+ So when you see these beautiful swarms
+
+08:34.960 --> 08:37.520
+ where there's biological systems of robots,
+
+08:38.520 --> 08:40.200
+ do you sometimes think of them
+
+08:40.200 --> 08:44.640
+ as a single individual living intelligent organism?
+
+08:44.640 --> 08:47.400
+ So it's the same as thinking of our human beings
+
+08:47.400 --> 08:51.160
+ are human civilization as one organism,
+
+08:51.160 --> 08:52.960
+ or do you still, as an engineer,
+
+08:52.960 --> 08:54.600
+ think about the individual components
+
+08:54.600 --> 08:55.440
+ and all the engineering
+
+08:55.440 --> 08:57.320
+ that went into the individual components?
+
+08:57.320 --> 08:58.640
+ Well, that's very interesting.
+
+08:58.640 --> 09:01.480
+ So again, philosophically as engineers,
+
+09:01.480 --> 09:05.400
+ what we wanna do is to go beyond
+
+09:05.400 --> 09:08.280
+ the individual components, the individual units,
+
+09:08.280 --> 09:11.520
+ and think about it as a unit, as a cohesive unit,
+
+09:11.520 --> 09:15.120
+ without worrying about the individual components.
+
+09:15.120 --> 09:17.760
+ If you start obsessing about
+
+09:17.760 --> 09:22.120
+ the individual building blocks and what they do,
+
+09:23.320 --> 09:27.960
+ you inevitably will find it hard to scale up.
+
+09:27.960 --> 09:29.000
+ Just mathematically,
+
+09:29.000 --> 09:31.600
+ just think about individual things you wanna model,
+
+09:31.600 --> 09:34.040
+ and if you want to have 10 of those,
+
+09:34.040 --> 09:36.440
+ then you essentially are taking Cartesian products
+
+09:36.440 --> 09:39.320
+ of 10 things, and that makes it really complicated.
+
+09:39.320 --> 09:41.840
+ Then to do any kind of synthesis or design
+
+09:41.840 --> 09:44.200
+ in that high dimension space is really hard.
+
+09:44.200 --> 09:45.800
+ So the right way to do this
+
+09:45.800 --> 09:49.040
+ is to think about the individuals in a clever way
+
+09:49.040 --> 09:51.120
+ so that at the higher level,
+
+09:51.120 --> 09:53.400
+ when you look at lots and lots of them,
+
+09:53.400 --> 09:55.320
+ abstractly, you can think of them
+
+09:55.320 --> 09:57.120
+ in some low dimensional space.
+
+09:57.120 --> 09:58.680
+ So what does that involve?
+
+09:58.680 --> 10:02.160
+ For the individual, do you have to try to make
+
+10:02.160 --> 10:05.160
+ the way they see the world as local as possible?
+
+10:05.160 --> 10:06.440
+ And the other thing,
+
+10:06.440 --> 10:09.560
+ do you just have to make them robust to collisions?
+
+10:09.560 --> 10:10.880
+ Like you said with the ants,
+
+10:10.880 --> 10:15.320
+ if something fails, the whole swarm doesn't fail.
+
+10:15.320 --> 10:17.760
+ Right, I think as engineers, we do this.
+
+10:17.760 --> 10:19.760
+ I mean, you think about, we build planes,
+
+10:19.760 --> 10:21.240
+ or we build iPhones,
+
+10:22.240 --> 10:26.280
+ and we know that by taking individual components,
+
+10:26.280 --> 10:30.080
+ well engineered components with well specified interfaces
+
+10:30.080 --> 10:31.680
+ that behave in a predictable way,
+
+10:31.680 --> 10:33.560
+ you can build complex systems.
+
+10:34.440 --> 10:36.880
+ So that's ingrained, I would claim,
+
+10:36.880 --> 10:39.400
+ in most engineers thinking,
+
+10:39.400 --> 10:41.600
+ and it's true for computer scientists as well.
+
+10:41.600 --> 10:44.760
+ I think what's different here is that you want
+
+10:44.760 --> 10:49.480
+ the individuals to be robust in some sense,
+
+10:49.480 --> 10:52.000
+ as we do in these other settings,
+
+10:52.000 --> 10:54.480
+ but you also want some degree of resiliency
+
+10:54.480 --> 10:56.320
+ for the population.
+
+10:56.320 --> 11:00.560
+ And so you really want them to be able to reestablish
+
+11:02.040 --> 11:03.840
+ communication with their neighbors.
+
+11:03.840 --> 11:08.840
+ You want them to rethink their strategy for group behavior.
+
+11:08.840 --> 11:10.760
+ You want them to reorganize.
+
+11:12.200 --> 11:15.920
+ And that's where I think a lot of the challenges lie.
+
+11:15.920 --> 11:18.160
+ So just at a high level,
+
+11:18.160 --> 11:20.880
+ what does it take for a bunch of,
+
+11:22.200 --> 11:24.440
+ what should we call them, flying robots,
+
+11:24.440 --> 11:26.680
+ to create a formation?
+
+11:26.680 --> 11:28.680
+ Just for people who are not familiar
+
+11:28.680 --> 11:32.760
+ with robotics in general, how much information is needed?
+
+11:32.760 --> 11:35.840
+ How do you even make it happen
+
+11:35.840 --> 11:39.520
+ without a centralized controller?
+
+11:39.520 --> 11:41.080
+ So, I mean, there are a couple of different ways
+
+11:41.080 --> 11:43.160
+ of looking at this.
+
+11:43.160 --> 11:45.680
+ If you are a purist,
+
+11:45.680 --> 11:50.680
+ you think of it as a way of recreating what nature does.
+
+11:53.560 --> 11:58.440
+ So nature forms groups for several reasons,
+
+11:58.440 --> 12:02.000
+ but mostly it's because of this instinct
+
+12:02.000 --> 12:05.680
+ that organisms have of preserving their colonies,
+
+12:05.680 --> 12:09.520
+ their population, which means what?
+
+12:09.520 --> 12:12.920
+ You need shelter, you need food, you need to procreate,
+
+12:12.920 --> 12:14.760
+ and that's basically it.
+
+12:14.760 --> 12:18.440
+ So the kinds of interactions you see are all organic.
+
+12:18.440 --> 12:19.760
+ They're all local.
+
+12:20.760 --> 12:24.080
+ And the only information that they share,
+
+12:24.080 --> 12:27.520
+ and mostly it's indirectly, is to, again,
+
+12:27.520 --> 12:30.000
+ preserve the herd or the flock,
+
+12:30.000 --> 12:35.000
+ or the swarm, and either by looking for new sources of food
+
+12:37.480 --> 12:39.440
+ or looking for new shelters, right?
+
+12:39.440 --> 12:40.280
+ Right.
+
+12:41.240 --> 12:45.360
+ As engineers, when we build swarms, we have a mission.
+
+12:46.560 --> 12:51.560
+ And when you think of a mission, and it involves mobility,
+
+12:52.480 --> 12:55.000
+ most often it's described in some kind
+
+12:55.000 --> 12:56.880
+ of a global coordinate system.
+
+12:56.880 --> 12:59.440
+ As a human, as an operator, as a commander,
+
+12:59.440 --> 13:03.560
+ or as a collaborator, I have my coordinate system,
+
+13:03.560 --> 13:06.640
+ and I want the robots to be consistent with that.
+
+13:07.600 --> 13:11.240
+ So I might think of it slightly differently.
+
+13:11.240 --> 13:15.440
+ I might want the robots to recognize that coordinate system,
+
+13:15.440 --> 13:17.720
+ which means not only do they have to think locally
+
+13:17.720 --> 13:19.600
+ in terms of who their immediate neighbors are,
+
+13:19.600 --> 13:20.920
+ but they have to be cognizant
+
+13:20.920 --> 13:24.040
+ of what the global environment is.
+
+13:24.040 --> 13:27.040
+ They have to be cognizant of what the global environment
+
+13:27.040 --> 13:28.280
+ looks like.
+
+13:28.280 --> 13:31.040
+ So if I say, surround this building
+
+13:31.040 --> 13:33.240
+ and protect this from intruders,
+
+13:33.240 --> 13:35.600
+ well, they're immediately in a building centered
+
+13:35.600 --> 13:37.040
+ coordinate system, and I have to tell them
+
+13:37.040 --> 13:38.680
+ where the building is.
+
+13:38.680 --> 13:40.040
+ And they're globally collaborating
+
+13:40.040 --> 13:41.280
+ on the map of that building.
+
+13:41.280 --> 13:44.160
+ They're maintaining some kind of global,
+
+13:44.160 --> 13:45.480
+ not just in the frame of the building,
+
+13:45.480 --> 13:49.000
+ but there's information that's ultimately being built up
+
+13:49.000 --> 13:53.280
+ explicitly as opposed to kind of implicitly,
+
+13:53.280 --> 13:54.360
+ like nature might.
+
+13:54.360 --> 13:55.200
+ Correct, correct.
+
+13:55.200 --> 13:57.680
+ So in some sense, nature is very, very sophisticated,
+
+13:57.680 --> 14:01.880
+ but the tasks that nature solves or needs to solve
+
+14:01.880 --> 14:05.160
+ are very different from the kind of engineered tasks,
+
+14:05.160 --> 14:09.760
+ artificial tasks that we are forced to address.
+
+14:09.760 --> 14:12.520
+ And again, there's nothing preventing us
+
+14:12.520 --> 14:15.160
+ from solving these other problems,
+
+14:15.160 --> 14:16.600
+ but ultimately it's about impact.
+
+14:16.600 --> 14:19.360
+ You want these swarms to do something useful.
+
+14:19.360 --> 14:24.360
+ And so you're kind of driven into this very unnatural,
+
+14:24.640 --> 14:25.480
+ if you will.
+
+14:25.480 --> 14:29.160
+ Unnatural, meaning not like how nature does, setting.
+
+14:29.160 --> 14:31.920
+ And it's probably a little bit more expensive
+
+14:31.920 --> 14:33.760
+ to do it the way nature does,
+
+14:33.760 --> 14:37.560
+ because nature is less sensitive
+
+14:37.560 --> 14:39.480
+ to the loss of the individual.
+
+14:39.480 --> 14:42.280
+ And cost wise in robotics,
+
+14:42.280 --> 14:45.480
+ I think you're more sensitive to losing individuals.
+
+14:45.480 --> 14:49.000
+ I think that's true, although if you look at the price
+
+14:49.000 --> 14:51.520
+ to performance ratio of robotic components,
+
+14:51.520 --> 14:54.720
+ it's coming down dramatically, right?
+
+14:54.720 --> 14:56.040
+ It continues to come down.
+
+14:56.040 --> 14:58.920
+ So I think we're asymptotically approaching the point
+
+14:58.920 --> 14:59.960
+ where we would get, yeah,
+
+14:59.960 --> 15:05.040
+ the cost of individuals would really become insignificant.
+
+15:05.040 --> 15:07.640
+ So let's step back at a high level view,
+
+15:07.640 --> 15:12.480
+ the impossible question of what kind of, as an overview,
+
+15:12.480 --> 15:14.400
+ what kind of autonomous flying vehicles
+
+15:14.400 --> 15:16.200
+ are there in general?
+
+15:16.200 --> 15:19.720
+ I think the ones that receive a lot of notoriety
+
+15:19.720 --> 15:22.560
+ are obviously the military vehicles.
+
+15:22.560 --> 15:26.280
+ Military vehicles are controlled by a base station,
+
+15:26.280 --> 15:29.640
+ but have a lot of human supervision.
+
+15:29.640 --> 15:31.800
+ But they have limited autonomy,
+
+15:31.800 --> 15:34.760
+ which is the ability to go from point A to point B.
+
+15:34.760 --> 15:37.080
+ And even the more sophisticated now,
+
+15:37.080 --> 15:40.400
+ sophisticated vehicles can do autonomous takeoff
+
+15:40.400 --> 15:41.760
+ and landing.
+
+15:41.760 --> 15:44.360
+ And those usually have wings and they're heavy.
+
+15:44.360 --> 15:45.360
+ Usually they're wings,
+
+15:45.360 --> 15:47.440
+ but then there's nothing preventing us from doing this
+
+15:47.440 --> 15:49.000
+ for helicopters as well.
+
+15:49.000 --> 15:52.480
+ There are many military organizations
+
+15:52.480 --> 15:56.560
+ that have autonomous helicopters in the same vein.
+
+15:56.560 --> 16:00.080
+ And by the way, you look at autopilots and airplanes
+
+16:00.080 --> 16:02.840
+ and it's actually very similar.
+
+16:02.840 --> 16:07.160
+ In fact, one interesting question we can ask is,
+
+16:07.160 --> 16:12.120
+ if you look at all the air safety violations,
+
+16:12.120 --> 16:14.080
+ all the crashes that occurred,
+
+16:14.080 --> 16:18.640
+ would they have happened if the plane were truly autonomous?
+
+16:18.640 --> 16:21.960
+ And I think you'll find that in many of the cases,
+
+16:21.960 --> 16:24.600
+ because of pilot error, we made silly decisions.
+
+16:24.600 --> 16:26.960
+ And so in some sense, even in air traffic,
+
+16:26.960 --> 16:29.800
+ commercial air traffic, there's a lot of applications,
+
+16:29.800 --> 16:33.960
+ although we only see autonomy being enabled
+
+16:33.960 --> 16:38.960
+ at very high altitudes when the plane is an autopilot.
+
+16:38.960 --> 16:41.960
+ The plane is an autopilot.
+
+16:41.960 --> 16:42.800
+ There's still a role for the human
+
+16:42.800 --> 16:47.640
+ and that kind of autonomy is, you're kind of implying,
+
+16:47.640 --> 16:48.680
+ I don't know what the right word is,
+
+16:48.680 --> 16:53.480
+ but it's a little dumber than it could be.
+
+16:53.480 --> 16:55.720
+ Right, so in the lab, of course,
+
+16:55.720 --> 16:59.200
+ we can afford to be a lot more aggressive.
+
+16:59.200 --> 17:04.200
+ And the question we try to ask is,
+
+17:04.200 --> 17:09.200
+ can we make robots that will be able to make decisions
+
+17:10.360 --> 17:13.680
+ without any kind of external infrastructure?
+
+17:13.680 --> 17:14.880
+ So what does that mean?
+
+17:14.880 --> 17:16.960
+ So the most common piece of infrastructure
+
+17:16.960 --> 17:19.640
+ that airplanes use today is GPS.
+
+17:20.560 --> 17:25.160
+ GPS is also the most brittle form of information.
+
+17:26.680 --> 17:30.480
+ If you have driven in a city, try to use GPS navigation,
+
+17:30.480 --> 17:32.760
+ in tall buildings, you immediately lose GPS.
+
+17:32.760 --> 17:36.280
+ And so that's not a very sophisticated way
+
+17:36.280 --> 17:37.840
+ of building autonomy.
+
+17:37.840 --> 17:39.560
+ I think the second piece of infrastructure
+
+17:39.560 --> 17:41.920
+ they rely on is communications.
+
+17:41.920 --> 17:46.200
+ Again, it's very easy to jam communications.
+
+17:47.360 --> 17:51.320
+ In fact, if you use wifi, you know that wifi signals
+
+17:51.320 --> 17:53.520
+ drop out, cell signals drop out.
+
+17:53.520 --> 17:56.800
+ So to rely on something like that is not good.
+
+17:58.560 --> 18:01.200
+ The third form of infrastructure we use,
+
+18:01.200 --> 18:02.920
+ and I hate to call it infrastructure,
+
+18:02.920 --> 18:06.360
+ but it is that, in the sense of robots, is people.
+
+18:06.360 --> 18:08.640
+ So you could rely on somebody to pilot you.
+
+18:09.960 --> 18:11.600
+ And so the question you wanna ask is,
+
+18:11.600 --> 18:14.760
+ if there are no pilots, there's no communications
+
+18:14.760 --> 18:18.720
+ with any base station, if there's no knowledge of position,
+
+18:18.720 --> 18:21.640
+ and if there's no a priori map,
+
+18:21.640 --> 18:24.880
+ a priori knowledge of what the environment looks like,
+
+18:24.880 --> 18:28.240
+ a priori model of what might happen in the future,
+
+18:28.240 --> 18:29.560
+ can robots navigate?
+
+18:29.560 --> 18:31.480
+ So that is true autonomy.
+
+18:31.480 --> 18:34.160
+ So that's true autonomy, and we're talking about,
+
+18:34.160 --> 18:36.880
+ you mentioned like military application of drones.
+
+18:36.880 --> 18:38.320
+ Okay, so what else is there?
+
+18:38.320 --> 18:42.080
+ You talk about agile, autonomous flying robots,
+
+18:42.080 --> 18:45.680
+ aerial robots, so that's a different kind of,
+
+18:45.680 --> 18:48.160
+ it's not winged, it's not big, at least it's small.
+
+18:48.160 --> 18:50.840
+ So I use the word agility mostly,
+
+18:50.840 --> 18:53.520
+ or at least we're motivated to do agile robots,
+
+18:53.520 --> 18:58.000
+ mostly because robots can operate
+
+18:58.000 --> 19:01.120
+ and should be operating in constrained environments.
+
+19:02.120 --> 19:06.960
+ And if you want to operate the way a global hawk operates,
+
+19:06.960 --> 19:09.120
+ I mean, the kinds of conditions in which you operate
+
+19:09.120 --> 19:10.760
+ are very, very restrictive.
+
+19:11.760 --> 19:13.720
+ If you wanna go inside a building,
+
+19:13.720 --> 19:15.600
+ for example, for search and rescue,
+
+19:15.600 --> 19:18.120
+ or to locate an active shooter,
+
+19:18.120 --> 19:22.120
+ or you wanna navigate under the canopy in an orchard
+
+19:22.120 --> 19:23.880
+ to look at health of plants,
+
+19:23.880 --> 19:28.240
+ or to look for, to count fruits,
+
+19:28.240 --> 19:31.240
+ to measure the tree trunks.
+
+19:31.240 --> 19:33.240
+ These are things we do, by the way.
+
+19:33.240 --> 19:35.400
+ There's some cool agriculture stuff you've shown
+
+19:35.400 --> 19:37.080
+ in the past, it's really awesome.
+
+19:37.080 --> 19:40.360
+ So in those kinds of settings, you do need that agility.
+
+19:40.360 --> 19:42.560
+ Agility does not necessarily mean
+
+19:42.560 --> 19:45.440
+ you break records for the 100 meters dash.
+
+19:45.440 --> 19:48.000
+ What it really means is you see the unexpected
+
+19:48.000 --> 19:51.480
+ and you're able to maneuver in a safe way,
+
+19:51.480 --> 19:55.400
+ and in a way that gets you the most information
+
+19:55.400 --> 19:57.640
+ about the thing you're trying to do.
+
+19:57.640 --> 20:00.440
+ By the way, you may be the only person
+
+20:00.440 --> 20:04.200
+ who, in a TED Talk, has used a math equation,
+
+20:04.200 --> 20:07.600
+ which is amazing, people should go see one of your TED Talks.
+
+20:07.600 --> 20:08.800
+ Actually, it's very interesting,
+
+20:08.800 --> 20:12.400
+ because the TED curator, Chris Anderson,
+
+20:12.400 --> 20:15.360
+ told me, you can't show math.
+
+20:15.360 --> 20:18.200
+ And I thought about it, but that's who I am.
+
+20:18.200 --> 20:20.760
+ I mean, that's our work.
+
+20:20.760 --> 20:25.760
+ And so I felt compelled to give the audience a taste
+
+20:25.760 --> 20:27.640
+ for at least some math.
+
+20:27.640 --> 20:32.640
+ So on that point, simply, what does it take
+
+20:32.880 --> 20:37.360
+ to make a thing with four motors fly, a quadcopter,
+
+20:37.360 --> 20:40.640
+ one of these little flying robots?
+
+20:41.760 --> 20:43.960
+ How hard is it to make it fly?
+
+20:43.960 --> 20:46.560
+ How do you coordinate the four motors?
+
+20:46.560 --> 20:51.560
+ How do you convert those motors into actual movement?
+
+20:52.600 --> 20:54.800
+ So this is an interesting question.
+
+20:54.800 --> 20:58.080
+ We've been trying to do this since 2000.
+
+20:58.080 --> 21:00.560
+ It is a commentary on the sensors
+
+21:00.560 --> 21:02.080
+ that were available back then,
+
+21:02.080 --> 21:04.280
+ the computers that were available back then.
+
+21:05.560 --> 21:10.280
+ And a number of things happened between 2000 and 2007.
+
+21:11.520 --> 21:14.120
+ One is the advances in computing,
+
+21:14.120 --> 21:16.760
+ which is, so we all know about Moore's Law,
+
+21:16.760 --> 21:19.680
+ but I think 2007 was a tipping point,
+
+21:19.680 --> 21:22.720
+ the year of the iPhone, the year of the cloud.
+
+21:22.720 --> 21:24.640
+ Lots of things happened in 2007.
+
+21:25.600 --> 21:27.600
+ But going back even further,
+
+21:27.600 --> 21:31.360
+ inertial measurement units as a sensor really matured.
+
+21:31.360 --> 21:33.040
+ Again, lots of reasons for that.
+
+21:33.920 --> 21:35.400
+ Certainly, there's a lot of federal funding,
+
+21:35.400 --> 21:37.360
+ particularly DARPA in the US,
+
+21:38.320 --> 21:42.760
+ but they didn't anticipate this boom in IMUs.
+
+21:42.760 --> 21:46.560
+ But if you look, subsequently what happened
+
+21:46.560 --> 21:50.040
+ is that every car manufacturer had to put an airbag in,
+
+21:50.040 --> 21:52.600
+ which meant you had to have an accelerometer on board.
+
+21:52.600 --> 21:55.000
+ And so that drove down the price to performance ratio.
+
+21:55.000 --> 21:56.880
+ Wow, I should know this.
+
+21:56.880 --> 21:57.960
+ That's very interesting.
+
+21:57.960 --> 21:59.360
+ That's very interesting, the connection there.
+
+21:59.360 --> 22:01.320
+ And that's why research is very,
+
+22:01.320 --> 22:03.280
+ it's very hard to predict the outcomes.
+
+22:04.840 --> 22:07.640
+ And again, the federal government spent a ton of money
+
+22:07.640 --> 22:12.280
+ on things that they thought were useful for resonators,
+
+22:12.280 --> 22:16.840
+ but it ended up enabling these small UAVs, which is great,
+
+22:16.840 --> 22:18.520
+ because I could have never raised that much money
+
+22:18.520 --> 22:20.760
+ and sold this project,
+
+22:20.760 --> 22:22.200
+ hey, we want to build these small UAVs.
+
+22:22.200 --> 22:25.440
+ Can you actually fund the development of low cost IMUs?
+
+22:25.440 --> 22:27.600
+ So why do you need an IMU on an IMU?
+
+22:27.600 --> 22:31.000
+ So I'll come back to that.
+
+22:31.000 --> 22:33.320
+ So in 2007, 2008, we were able to build these.
+
+22:33.320 --> 22:35.200
+ And then the question you're asking was a good one.
+
+22:35.200 --> 22:40.240
+ How do you coordinate the motors to develop this?
+
+22:40.240 --> 22:43.880
+ But over the last 10 years, everything is commoditized.
+
+22:43.880 --> 22:46.240
+ A high school kid today can pick up
+
+22:46.240 --> 22:50.560
+ a Raspberry Pi kit and build this.
+
+22:50.560 --> 22:53.200
+ All the low levels functionality is all automated.
+
+22:54.160 --> 22:56.360
+ But basically at some level,
+
+22:56.360 --> 23:01.360
+ you have to drive the motors at the right RPMs,
+
+23:01.360 --> 23:03.680
+ the right velocity,
+
+23:04.560 --> 23:07.480
+ in order to generate the right amount of thrust,
+
+23:07.480 --> 23:10.360
+ in order to position it and orient it in a way
+
+23:10.360 --> 23:12.840
+ that you need to in order to fly.
+
+23:13.800 --> 23:16.680
+ The feedback that you get is from onboard sensors,
+
+23:16.680 --> 23:18.400
+ and the IMU is an important part of it.
+
+23:18.400 --> 23:23.400
+ The IMU tells you what the acceleration is,
+
+23:23.840 --> 23:26.400
+ as well as what the angular velocity is.
+
+23:26.400 --> 23:29.200
+ And those are important pieces of information.
+
+23:30.440 --> 23:34.200
+ In addition to that, you need some kind of local position
+
+23:34.200 --> 23:37.480
+ or velocity information.
+
+23:37.480 --> 23:39.360
+ For example, when we walk,
+
+23:39.360 --> 23:41.560
+ we implicitly have this information
+
+23:41.560 --> 23:45.840
+ because we kind of know what our stride length is.
+
+23:46.720 --> 23:51.480
+ We also are looking at images fly past our retina,
+
+23:51.480 --> 23:54.280
+ if you will, and so we can estimate velocity.
+
+23:54.280 --> 23:56.360
+ We also have accelerometers in our head,
+
+23:56.360 --> 23:59.160
+ and we're able to integrate all these pieces of information
+
+23:59.160 --> 24:02.360
+ to determine where we are as we walk.
+
+24:02.360 --> 24:04.320
+ And so robots have to do something very similar.
+
+24:04.320 --> 24:08.160
+ You need an IMU, you need some kind of a camera
+
+24:08.160 --> 24:11.640
+ or other sensor that's measuring velocity,
+
+24:12.560 --> 24:15.800
+ and then you need some kind of a global reference frame
+
+24:15.800 --> 24:19.520
+ if you really want to think about doing something
+
+24:19.520 --> 24:21.280
+ in a world coordinate system.
+
+24:21.280 --> 24:23.680
+ And so how do you estimate your position
+
+24:23.680 --> 24:25.160
+ with respect to that global reference frame?
+
+24:25.160 --> 24:26.560
+ That's important as well.
+
+24:26.560 --> 24:29.520
+ So coordinating the RPMs of the four motors
+
+24:29.520 --> 24:32.640
+ is what allows you to, first of all, fly and hover,
+
+24:32.640 --> 24:35.600
+ and then you can change the orientation
+
+24:35.600 --> 24:37.600
+ and the velocity and so on.
+
+24:37.600 --> 24:38.440
+ Exactly, exactly.
+
+24:38.440 --> 24:40.320
+ So it's a bunch of degrees of freedom
+
+24:40.320 --> 24:41.160
+ that you're complaining about.
+
+24:41.160 --> 24:42.200
+ There's six degrees of freedom,
+
+24:42.200 --> 24:44.920
+ but you only have four inputs, the four motors.
+
+24:44.920 --> 24:49.920
+ And it turns out to be a remarkably versatile configuration.
+
+24:50.920 --> 24:53.080
+ You think at first, well, I only have four motors,
+
+24:53.080 --> 24:55.000
+ how do I go sideways?
+
+24:55.000 --> 24:57.280
+ But it's not too hard to say, well, if I tilt myself,
+
+24:57.280 --> 25:00.440
+ I can go sideways, and then you have four motors
+
+25:00.440 --> 25:03.320
+ pointing up, how do I rotate in place
+
+25:03.320 --> 25:05.360
+ about a vertical axis?
+
+25:05.360 --> 25:07.800
+ Well, you rotate them at different speeds
+
+25:07.800 --> 25:09.720
+ and that generates reaction moments
+
+25:09.720 --> 25:11.520
+ and that allows you to turn.
+
+25:11.520 --> 25:14.960
+ So it's actually a pretty, it's an optimal configuration
+
+25:14.960 --> 25:17.040
+ from an engineer standpoint.
+
+25:18.360 --> 25:23.360
+ It's very simple, very cleverly done, and very versatile.
+
+25:23.360 --> 25:27.240
+ So if you could step back to a time,
+
+25:27.240 --> 25:30.000
+ so I've always known flying robots as,
+
+25:31.040 --> 25:35.760
+ to me, it was natural that a quadcopter should fly.
+
+25:35.760 --> 25:37.880
+ But when you first started working with it,
+
+25:38.800 --> 25:42.000
+ how surprised are you that you can make,
+
+25:42.000 --> 25:45.520
+ do so much with the four motors?
+
+25:45.520 --> 25:47.600
+ How surprising is it that you can make this thing fly,
+
+25:47.600 --> 25:49.760
+ first of all, that you can make it hover,
+
+25:49.760 --> 25:52.000
+ that you can add control to it?
+
+25:52.000 --> 25:55.080
+ Firstly, this is not, the four motor configuration
+
+25:55.080 --> 25:56.400
+ is not ours.
+
+25:56.400 --> 25:59.320
+ You can, it has at least a hundred year history.
+
+26:00.320 --> 26:04.160
+ And various people, various people try to get quadrotors
+
+26:04.160 --> 26:06.840
+ to fly without much success.
+
+26:08.480 --> 26:10.760
+ As I said, we've been working on this since 2000.
+
+26:10.760 --> 26:14.400
+ Our first designs were, well, this is way too complicated.
+
+26:14.400 --> 26:18.480
+ Why not we try to get an omnidirectional flying robot?
+
+26:18.480 --> 26:21.760
+ So our early designs, we had eight rotors.
+
+26:21.760 --> 26:25.200
+ And so these eight rotors were arranged uniformly
+
+26:26.600 --> 26:28.000
+ on a sphere, if you will.
+
+26:28.000 --> 26:30.440
+ So you can imagine a symmetric configuration.
+
+26:30.440 --> 26:33.280
+ And so you should be able to fly anywhere.
+
+26:33.280 --> 26:36.240
+ But the real challenge we had is the strength to weight ratio
+
+26:36.240 --> 26:37.080
+ is not enough.
+
+26:37.080 --> 26:39.680
+ And of course, we didn't have the sensors and so on.
+
+26:40.520 --> 26:43.040
+ So everybody knew, or at least the people
+
+26:43.040 --> 26:44.800
+ who worked with rotorcrafts knew,
+
+26:44.800 --> 26:46.520
+ four rotors will get it done.
+
+26:47.520 --> 26:49.400
+ So that was not our idea.
+
+26:49.400 --> 26:52.800
+ But it took a while before we could actually do
+
+26:52.800 --> 26:56.920
+ the onboard sensing and the computation that was needed
+
+26:56.920 --> 27:01.000
+ for the kinds of agile maneuvering that we wanted to do
+
+27:01.000 --> 27:03.000
+ in our little aerial robots.
+
+27:03.000 --> 27:07.560
+ And that only happened between 2007 and 2009 in our lab.
+
+27:07.560 --> 27:09.960
+ Yeah, and you have to send the signal
+
+27:09.960 --> 27:12.480
+ maybe a hundred times a second.
+
+27:12.480 --> 27:15.960
+ So the compute there, everything has to come down in price.
+
+27:15.960 --> 27:20.960
+ And what are the steps of getting from point A to point B?
+
+27:21.720 --> 27:25.200
+ So we just talked about like local control.
+
+27:25.200 --> 27:30.200
+ But if all the kind of cool dancing in the air
+
+27:30.840 --> 27:34.520
+ that I've seen you show, how do you make it happen?
+
+27:34.520 --> 27:37.360
+ How do you make a trajectory?
+
+27:37.360 --> 27:40.520
+ First of all, okay, figure out a trajectory.
+
+27:40.520 --> 27:41.680
+ So plan a trajectory.
+
+27:41.680 --> 27:44.400
+ And then how do you make that trajectory happen?
+
+27:44.400 --> 27:47.280
+ Yeah, I think planning is a very fundamental problem
+
+27:47.280 --> 27:48.120
+ in robotics.
+
+27:48.120 --> 27:50.800
+ I think 10 years ago it was an esoteric thing,
+
+27:50.800 --> 27:53.040
+ but today with self driving cars,
+
+27:53.040 --> 27:55.840
+ everybody can understand this basic idea
+
+27:55.840 --> 27:57.920
+ that a car sees a whole bunch of things
+
+27:57.920 --> 28:00.320
+ and it has to keep a lane or maybe make a right turn
+
+28:00.320 --> 28:01.280
+ or switch lanes.
+
+28:01.280 --> 28:02.680
+ It has to plan a trajectory.
+
+28:02.680 --> 28:03.560
+ It has to be safe.
+
+28:03.560 --> 28:04.840
+ It has to be efficient.
+
+28:04.840 --> 28:06.640
+ So everybody's familiar with that.
+
+28:06.640 --> 28:10.240
+ That's kind of the first step that you have to think about
+
+28:10.240 --> 28:14.800
+ when you say autonomy.
+
+28:14.800 --> 28:19.120
+ And so for us, it's about finding smooth motions,
+
+28:19.120 --> 28:21.320
+ motions that are safe.
+
+28:21.320 --> 28:22.880
+ So we think about these two things.
+
+28:22.880 --> 28:24.680
+ One is optimality, one is safety.
+
+28:24.680 --> 28:27.200
+ Clearly you cannot compromise safety.
+
+28:28.440 --> 28:31.360
+ So you're looking for safe, optimal motions.
+
+28:31.360 --> 28:34.480
+ The other thing you have to think about is
+
+28:34.480 --> 28:38.160
+ can you actually compute a reasonable trajectory
+
+28:38.160 --> 28:40.760
+ in a small amount of time?
+
+28:40.760 --> 28:42.280
+ Cause you have a time budget.
+
+28:42.280 --> 28:45.160
+ So the optimal becomes suboptimal,
+
+28:45.160 --> 28:50.160
+ but in our lab we focus on synthesizing smooth trajectory
+
+28:51.160 --> 28:53.000
+ that satisfy all the constraints.
+
+28:53.000 --> 28:57.120
+ In other words, don't violate any safety constraints
+
+28:58.440 --> 29:02.880
+ and is as efficient as possible.
+
+29:02.880 --> 29:04.360
+ And when I say efficient,
+
+29:04.360 --> 29:06.600
+ it could mean I want to get from point A to point B
+
+29:06.600 --> 29:08.360
+ as quickly as possible,
+
+29:08.360 --> 29:11.840
+ or I want to get to it as gracefully as possible,
+
+29:12.840 --> 29:15.960
+ or I want to consume as little energy as possible.
+
+29:15.960 --> 29:18.240
+ But always staying within the safety constraints.
+
+29:18.240 --> 29:22.800
+ But yes, always finding a safe trajectory.
+
+29:22.800 --> 29:25.040
+ So there's a lot of excitement and progress
+
+29:25.040 --> 29:27.360
+ in the field of machine learning
+
+29:27.360 --> 29:29.360
+ and reinforcement learning
+
+29:29.360 --> 29:32.200
+ and the neural network variant of that
+
+29:32.200 --> 29:33.920
+ with deep reinforcement learning.
+
+29:33.920 --> 29:36.360
+ Do you see a role of machine learning
+
+29:36.360 --> 29:40.560
+ in, so a lot of the success of flying robots
+
+29:40.560 --> 29:42.320
+ did not rely on machine learning,
+
+29:42.320 --> 29:45.040
+ except for maybe a little bit of the perception
+
+29:45.040 --> 29:46.600
+ on the computer vision side.
+
+29:46.600 --> 29:48.440
+ On the control side and the planning,
+
+29:48.440 --> 29:50.400
+ do you see there's a role in the future
+
+29:50.400 --> 29:51.680
+ for machine learning?
+
+29:51.680 --> 29:53.800
+ So let me disagree a little bit with you.
+
+29:53.800 --> 29:56.800
+ I think we never perhaps called out in my work,
+
+29:56.800 --> 29:57.720
+ called out learning,
+
+29:57.720 --> 30:00.600
+ but even this very simple idea of being able to fly
+
+30:00.600 --> 30:02.200
+ through a constrained space.
+
+30:02.200 --> 30:05.680
+ The first time you try it, you'll invariably,
+
+30:05.680 --> 30:08.440
+ you might get it wrong if the task is challenging.
+
+30:08.440 --> 30:12.200
+ And the reason is to get it perfectly right,
+
+30:12.200 --> 30:14.600
+ you have to model everything in the environment.
+
+30:15.600 --> 30:19.960
+ And flying is notoriously hard to model.
+
+30:19.960 --> 30:24.960
+ There are aerodynamic effects that we constantly discover.
+
+30:26.520 --> 30:29.440
+ Even just before I was talking to you,
+
+30:29.440 --> 30:33.440
+ I was talking to a student about how blades flap
+
+30:33.440 --> 30:35.320
+ when they fly.
+
+30:35.320 --> 30:40.320
+ And that ends up changing how a rotorcraft
+
+30:40.880 --> 30:43.960
+ is accelerated in the angular direction.
+
+30:43.960 --> 30:46.360
+ Does he use like micro flaps or something?
+
+30:46.360 --> 30:47.280
+ It's not micro flaps.
+
+30:47.280 --> 30:49.640
+ So we assume that each blade is rigid,
+
+30:49.640 --> 30:51.720
+ but actually it flaps a little bit.
+
+30:51.720 --> 30:52.880
+ It bends.
+
+30:52.880 --> 30:53.720
+ Interesting, yeah.
+
+30:53.720 --> 30:56.040
+ And so the models rely on the fact,
+
+30:56.040 --> 30:58.640
+ on the assumption that they're not rigid.
+
+30:58.640 --> 31:00.640
+ On the assumption that they're actually rigid,
+
+31:00.640 --> 31:02.240
+ but that's not true.
+
+31:02.240 --> 31:03.720
+ If you're flying really quickly,
+
+31:03.720 --> 31:06.920
+ these effects become significant.
+
+31:06.920 --> 31:09.240
+ If you're flying close to the ground,
+
+31:09.240 --> 31:12.160
+ you get pushed off by the ground, right?
+
+31:12.160 --> 31:14.920
+ Something which every pilot knows when he tries to land
+
+31:14.920 --> 31:18.000
+ or she tries to land, this is called a ground effect.
+
+31:18.920 --> 31:21.000
+ Something very few pilots think about
+
+31:21.000 --> 31:23.040
+ is what happens when you go close to a ceiling
+
+31:23.040 --> 31:25.320
+ or you get sucked into a ceiling.
+
+31:25.320 --> 31:26.880
+ There are very few aircrafts
+
+31:26.880 --> 31:29.520
+ that fly close to any kind of ceiling.
+
+31:29.520 --> 31:33.520
+ Likewise, when you go close to a wall,
+
+31:33.520 --> 31:35.720
+ there are these wall effects.
+
+31:35.720 --> 31:37.680
+ And if you've gone on a train
+
+31:37.680 --> 31:39.600
+ and you pass another train that's traveling
+
+31:39.600 --> 31:42.400
+ in the opposite direction, you feel the buffeting.
+
+31:42.400 --> 31:45.400
+ And so these kinds of microclimates
+
+31:45.400 --> 31:47.880
+ affect our UAV significantly.
+
+31:47.880 --> 31:48.720
+ So if you want...
+
+31:48.720 --> 31:50.640
+ And they're impossible to model, essentially.
+
+31:50.640 --> 31:52.480
+ I wouldn't say they're impossible to model,
+
+31:52.480 --> 31:54.880
+ but the level of sophistication you would need
+
+31:54.880 --> 31:58.600
+ in the model and the software would be tremendous.
+
+32:00.000 --> 32:02.920
+ Plus, to get everything right would be awfully tedious.
+
+32:02.920 --> 32:05.080
+ So the way we do this is over time,
+
+32:05.080 --> 32:09.000
+ we figure out how to adapt to these conditions.
+
+32:10.360 --> 32:13.160
+ So early on, we use the form of learning
+
+32:13.160 --> 32:15.760
+ that we call iterative learning.
+
+32:15.760 --> 32:18.600
+ So this idea, if you want to perform a task,
+
+32:18.600 --> 32:22.120
+ there are a few things that you need to change
+
+32:22.120 --> 32:24.960
+ and iterate over a few parameters
+
+32:24.960 --> 32:29.280
+ that over time you can figure out.
+
+32:29.280 --> 32:33.400
+ So I could call it policy gradient reinforcement learning,
+
+32:33.400 --> 32:34.920
+ but actually it was just iterative learning.
+
+32:34.920 --> 32:36.000
+ Iterative learning.
+
+32:36.000 --> 32:37.800
+ And so this was there way back.
+
+32:37.800 --> 32:39.440
+ I think what's interesting is,
+
+32:39.440 --> 32:41.640
+ if you look at autonomous vehicles today,
+
+32:43.120 --> 32:45.680
+ learning occurs, could occur in two pieces.
+
+32:45.680 --> 32:47.960
+ One is perception, understanding the world.
+
+32:47.960 --> 32:50.080
+ Second is action, taking actions.
+
+32:50.080 --> 32:52.240
+ Everything that I've seen that is successful
+
+32:52.240 --> 32:54.360
+ is on the perception side of things.
+
+32:54.360 --> 32:55.400
+ So in computer vision,
+
+32:55.400 --> 32:57.840
+ we've made amazing strides in the last 10 years.
+
+32:57.840 --> 33:01.640
+ So recognizing objects, actually detecting objects,
+
+33:01.640 --> 33:06.400
+ classifying them and tagging them in some sense,
+
+33:06.400 --> 33:07.440
+ annotating them.
+
+33:07.440 --> 33:09.640
+ This is all done through machine learning.
+
+33:09.640 --> 33:12.160
+ On the action side, on the other hand,
+
+33:12.160 --> 33:13.720
+ I don't know of any examples
+
+33:13.720 --> 33:15.560
+ where there are fielded systems
+
+33:15.560 --> 33:17.560
+ where we actually learn
+
+33:17.560 --> 33:20.560
+ the right behavior.
+
+33:20.560 --> 33:22.760
+ Outside of single demonstration is successful.
+
+33:22.760 --> 33:24.640
+ In the laboratory, this is the holy grail.
+
+33:24.640 --> 33:26.040
+ Can you do end to end learning?
+
+33:26.040 --> 33:28.800
+ Can you go from pixels to motor currents?
+
+33:30.200 --> 33:31.600
+ This is really, really hard.
+
+33:32.800 --> 33:35.080
+ And I think if you go forward,
+
+33:35.080 --> 33:37.600
+ the right way to think about these things
+
+33:37.600 --> 33:40.720
+ is data driven approaches,
+
+33:40.720 --> 33:42.400
+ learning based approaches,
+
+33:42.400 --> 33:45.280
+ in concert with model based approaches,
+
+33:45.280 --> 33:47.320
+ which is the traditional way of doing things.
+
+33:47.320 --> 33:48.720
+ So I think there's a piece,
+
+33:48.720 --> 33:51.400
+ there's a role for each of these methodologies.
+
+33:51.400 --> 33:52.440
+ So what do you think,
+
+33:52.440 --> 33:53.880
+ just jumping out on topic
+
+33:53.880 --> 33:56.200
+ since you mentioned autonomous vehicles,
+
+33:56.200 --> 33:58.480
+ what do you think are the limits on the perception side?
+
+33:58.480 --> 34:01.080
+ So I've talked to Elon Musk
+
+34:01.080 --> 34:03.320
+ and there on the perception side,
+
+34:03.320 --> 34:05.960
+ they're using primarily computer vision
+
+34:05.960 --> 34:08.080
+ to perceive the environment.
+
+34:08.080 --> 34:09.760
+ In your work with,
+
+34:09.760 --> 34:12.560
+ because you work with the real world a lot
+
+34:12.560 --> 34:13.720
+ and the physical world,
+
+34:13.720 --> 34:15.800
+ what are the limits of computer vision?
+
+34:15.800 --> 34:18.000
+ Do you think we can solve autonomous vehicles
+
+34:19.160 --> 34:20.880
+ on the perception side,
+
+34:20.880 --> 34:24.240
+ focusing on vision alone and machine learning?
+
+34:24.240 --> 34:27.480
+ So, we also have a spinoff company,
+
+34:27.480 --> 34:31.840
+ Exxon Technologies that works underground in mines.
+
+34:31.840 --> 34:35.600
+ So you go into mines, they're dark, they're dirty.
+
+34:36.480 --> 34:38.600
+ You fly in a dirty area,
+
+34:38.600 --> 34:41.120
+ there's stuff you kick up from by the propellers,
+
+34:41.120 --> 34:42.720
+ the downwash kicks up dust.
+
+34:42.720 --> 34:45.520
+ I challenge you to get a computer vision algorithm
+
+34:45.520 --> 34:46.680
+ to work there.
+
+34:46.680 --> 34:49.600
+ So we use LIDARs in that setting.
+
+34:51.200 --> 34:55.360
+ Indoors and even outdoors when we fly through fields,
+
+34:55.360 --> 34:57.120
+ I think there's a lot of potential
+
+34:57.120 --> 34:59.960
+ for just solving the problem using computer vision alone.
+
+35:01.240 --> 35:02.760
+ But I think the bigger question is,
+
+35:02.760 --> 35:06.160
+ can you actually solve
+
+35:06.160 --> 35:09.440
+ or can you actually identify all the corner cases
+
+35:09.440 --> 35:13.920
+ using a single sensing modality and using learning alone?
+
+35:13.920 --> 35:15.400
+ So what's your intuition there?
+
+35:15.400 --> 35:17.920
+ So look, if you have a corner case
+
+35:17.920 --> 35:20.000
+ and your algorithm doesn't work,
+
+35:20.000 --> 35:23.200
+ your instinct is to go get data about the corner case
+
+35:23.200 --> 35:26.640
+ and patch it up, learn how to deal with that corner case.
+
+35:27.640 --> 35:32.040
+ But at some point, this is gonna saturate,
+
+35:32.040 --> 35:34.200
+ this approach is not viable.
+
+35:34.200 --> 35:38.000
+ So today, computer vision algorithms can detect
+
+35:38.000 --> 35:41.360
+ 90% of the objects or can detect objects 90% of the time,
+
+35:41.360 --> 35:43.920
+ classify them 90% of the time.
+
+35:43.920 --> 35:47.960
+ Cats on the internet probably can do 95%, I don't know.
+
+35:47.960 --> 35:52.520
+ But to get from 90% to 99%, you need a lot more data.
+
+35:52.520 --> 35:54.480
+ And then I tell you, well, that's not enough
+
+35:54.480 --> 35:56.680
+ because I have a safety critical application,
+
+35:56.680 --> 36:00.160
+ I wanna go from 99% to 99.9%.
+
+36:00.160 --> 36:01.600
+ That's even more data.
+
+36:01.600 --> 36:08.600
+ So I think if you look at wanting accuracy on the X axis
+
+36:09.600 --> 36:14.080
+ and look at the amount of data on the Y axis,
+
+36:14.080 --> 36:16.440
+ I believe that curve is an exponential curve.
+
+36:16.440 --> 36:19.480
+ Wow, okay, it's even hard if it's linear.
+
+36:19.480 --> 36:20.800
+ It's hard if it's linear, totally,
+
+36:20.800 --> 36:22.560
+ but I think it's exponential.
+
+36:22.560 --> 36:24.120
+ And the other thing you have to think about
+
+36:24.120 --> 36:29.600
+ is that this process is a very, very power hungry process
+
+36:29.600 --> 36:32.880
+ to run data farms or servers.
+
+36:32.880 --> 36:34.600
+ Power, do you mean literally power?
+
+36:34.600 --> 36:36.600
+ Literally power, literally power.
+
+36:36.600 --> 36:41.760
+ So in 2014, five years ago, and I don't have more recent data,
+
+36:41.760 --> 36:48.360
+ 2% of US electricity consumption was from data farms.
+
+36:48.360 --> 36:52.080
+ So we think about this as an information science
+
+36:52.080 --> 36:54.240
+ and information processing problem.
+
+36:54.240 --> 36:57.840
+ Actually, it is an energy processing problem.
+
+36:57.840 --> 37:00.440
+ And so unless we figured out better ways of doing this,
+
+37:00.440 --> 37:02.440
+ I don't think this is viable.
+
+37:02.440 --> 37:06.600
+ So talking about driving, which is a safety critical application
+
+37:06.600 --> 37:10.440
+ and some aspect of flight is safety critical,
+
+37:10.440 --> 37:12.960
+ maybe philosophical question, maybe an engineering one,
+
+37:12.960 --> 37:15.000
+ what problem do you think is harder to solve,
+
+37:15.000 --> 37:18.120
+ autonomous driving or autonomous flight?
+
+37:18.120 --> 37:19.920
+ That's a really interesting question.
+
+37:19.920 --> 37:25.440
+ I think autonomous flight has several advantages
+
+37:25.440 --> 37:29.360
+ that autonomous driving doesn't have.
+
+37:29.360 --> 37:32.400
+ So look, if I want to go from point A to point B,
+
+37:32.400 --> 37:34.320
+ I have a very, very safe trajectory.
+
+37:34.320 --> 37:36.800
+ Go vertically up to a maximum altitude,
+
+37:36.800 --> 37:39.480
+ fly horizontally to just about the destination,
+
+37:39.480 --> 37:42.400
+ and then come down vertically.
+
+37:42.400 --> 37:45.400
+ This is preprogrammed.
+
+37:45.400 --> 37:48.040
+ The equivalent of that is very hard to find
+
+37:48.040 --> 37:51.560
+ in the self driving car world because you're on the ground,
+
+37:51.560 --> 37:53.560
+ you're in a two dimensional surface,
+
+37:53.560 --> 37:56.680
+ and the trajectories on the two dimensional surface
+
+37:56.680 --> 38:00.200
+ are more likely to encounter obstacles.
+
+38:00.200 --> 38:03.280
+ I mean this in an intuitive sense, but mathematically true.
+
+38:03.280 --> 38:06.360
+ That's mathematically as well, that's true.
+
+38:06.360 --> 38:10.040
+ There's other option on the 2G space of platooning,
+
+38:10.040 --> 38:11.640
+ or because there's so many obstacles,
+
+38:11.640 --> 38:13.280
+ you can connect with those obstacles
+
+38:13.280 --> 38:14.560
+ and all these kind of options.
+
+38:14.560 --> 38:16.560
+ Sure, but those exist in the three dimensional space as well.
+
+38:16.560 --> 38:17.560
+ So they do.
+
+38:17.560 --> 38:21.800
+ So the question also implies how difficult are obstacles
+
+38:21.800 --> 38:23.800
+ in the three dimensional space in flight?
+
+38:23.800 --> 38:25.600
+ So that's the downside.
+
+38:25.600 --> 38:26.920
+ I think in three dimensional space,
+
+38:26.920 --> 38:29.080
+ you're modeling three dimensional world,
+
+38:29.080 --> 38:31.280
+ not just because you want to avoid it,
+
+38:31.280 --> 38:33.040
+ but you want to reason about it,
+
+38:33.040 --> 38:35.360
+ and you want to work in the three dimensional environment,
+
+38:35.360 --> 38:37.480
+ and that's significantly harder.
+
+38:37.480 --> 38:38.920
+ So that's one disadvantage.
+
+38:38.920 --> 38:41.040
+ I think the second disadvantage is of course,
+
+38:41.040 --> 38:43.200
+ anytime you fly, you have to put up
+
+38:43.200 --> 38:46.560
+ with the peculiarities of aerodynamics
+
+38:46.560 --> 38:48.720
+ and their complicated environments.
+
+38:48.720 --> 38:49.800
+ How do you negotiate that?
+
+38:49.800 --> 38:51.880
+ So that's always a problem.
+
+38:51.880 --> 38:55.240
+ Do you see a time in the future where there is,
+
+38:55.240 --> 38:58.720
+ you mentioned there's agriculture applications.
+
+38:58.720 --> 39:01.680
+ So there's a lot of applications of flying robots,
+
+39:01.680 --> 39:03.040
+ but do you see a time in the future
+
+39:03.040 --> 39:05.360
+ where there's tens of thousands,
+
+39:05.360 --> 39:08.160
+ or maybe hundreds of thousands of delivery drones
+
+39:08.160 --> 39:12.160
+ that fill the sky, delivery flying robots?
+
+39:12.160 --> 39:14.200
+ I think there's a lot of potential
+
+39:14.200 --> 39:15.920
+ for the last mile delivery.
+
+39:15.920 --> 39:19.240
+ And so in crowded cities, I don't know,
+
+39:19.240 --> 39:21.400
+ if you go to a place like Hong Kong,
+
+39:21.400 --> 39:24.400
+ just crossing the river can take half an hour,
+
+39:24.400 --> 39:29.400
+ and while a drone can just do it in five minutes at most.
+
+39:29.400 --> 39:34.400
+ I think you look at delivery of supplies to remote villages.
+
+39:35.800 --> 39:38.680
+ I work with a nonprofit called Weave Robotics.
+
+39:38.680 --> 39:40.920
+ So they work in the Peruvian Amazon,
+
+39:40.920 --> 39:44.680
+ where the only highways that are available
+
+39:44.680 --> 39:47.440
+ are the only highways or rivers.
+
+39:47.440 --> 39:52.440
+ And to get from point A to point B may take five hours,
+
+39:52.960 --> 39:55.600
+ while with a drone, you can get there in 30 minutes.
+
+39:56.680 --> 39:59.880
+ So just delivering drugs,
+
+39:59.880 --> 40:04.880
+ retrieving samples for testing vaccines,
+
+40:05.160 --> 40:07.120
+ I think there's huge potential here.
+
+40:07.120 --> 40:09.960
+ So I think the challenges are not technological,
+
+40:09.960 --> 40:12.040
+ but the challenge is economical.
+
+40:12.040 --> 40:15.560
+ The one thing I'll tell you that nobody thinks about
+
+40:15.560 --> 40:18.920
+ is the fact that we've not made huge strides
+
+40:18.920 --> 40:20.840
+ in battery technology.
+
+40:20.840 --> 40:23.520
+ Yes, it's true, batteries are becoming less expensive
+
+40:23.520 --> 40:26.240
+ because we have these mega factories that are coming up,
+
+40:26.240 --> 40:28.800
+ but they're all based on lithium based technologies.
+
+40:28.800 --> 40:31.480
+ And if you look at the energy density
+
+40:31.480 --> 40:33.240
+ and the power density,
+
+40:33.240 --> 40:38.000
+ those are two fundamentally limiting numbers.
+
+40:38.000 --> 40:39.680
+ So power density is important
+
+40:39.680 --> 40:42.480
+ because for a UAV to take off vertically into the air,
+
+40:42.480 --> 40:46.360
+ which most drones do, they don't have a runway,
+
+40:46.360 --> 40:50.240
+ you consume roughly 200 watts per kilo at the small size.
+
+40:51.560 --> 40:53.920
+ That's a lot, right?
+
+40:53.920 --> 40:57.520
+ In contrast, the human brain consumes less than 80 watts,
+
+40:57.520 --> 40:58.920
+ the whole of the human brain.
+
+40:59.920 --> 41:03.600
+ So just imagine just lifting yourself into the air
+
+41:03.600 --> 41:06.000
+ is like two or three light bulbs,
+
+41:06.000 --> 41:07.840
+ which makes no sense to me.
+
+41:07.840 --> 41:10.440
+ Yeah, so you're going to have to at scale
+
+41:10.440 --> 41:12.880
+ solve the energy problem then,
+
+41:12.880 --> 41:17.880
+ charging the batteries, storing the energy and so on.
+
+41:18.920 --> 41:20.680
+ And then the storage is the second problem,
+
+41:20.680 --> 41:22.960
+ but storage limits the range.
+
+41:22.960 --> 41:27.960
+ But you have to remember that you have to burn
+
+41:28.680 --> 41:31.600
+ a lot of it per given time.
+
+41:31.600 --> 41:32.920
+ So the burning is another problem.
+
+41:32.920 --> 41:34.640
+ Which is a power question.
+
+41:34.640 --> 41:38.640
+ Yes, and do you think just your intuition,
+
+41:38.640 --> 41:43.640
+ there are breakthroughs in batteries on the horizon?
+
+41:44.960 --> 41:46.440
+ How hard is that problem?
+
+41:46.440 --> 41:47.600
+ Look, there are a lot of companies
+
+41:47.600 --> 41:52.600
+ that are promising flying cars that are autonomous
+
+41:53.880 --> 41:55.120
+ and that are clean.
+
+41:59.400 --> 42:01.680
+ I think they're over promising.
+
+42:01.680 --> 42:04.800
+ The autonomy piece is doable.
+
+42:04.800 --> 42:07.040
+ The clean piece, I don't think so.
+
+42:08.000 --> 42:11.840
+ There's another company that I work with called JetOptra.
+
+42:11.840 --> 42:14.360
+ They make small jet engines.
+
+42:15.760 --> 42:18.080
+ And they can get up to 50 miles an hour very easily
+
+42:18.080 --> 42:19.960
+ and lift 50 kilos.
+
+42:19.960 --> 42:22.840
+ But they're jet engines, they're efficient,
+
+42:23.920 --> 42:26.320
+ they're a little louder than electric vehicles,
+
+42:26.320 --> 42:28.960
+ but they can build flying cars.
+
+42:28.960 --> 42:32.440
+ So your sense is that there's a lot of pieces
+
+42:32.440 --> 42:33.520
+ that have come together.
+
+42:33.520 --> 42:37.360
+ So on this crazy question,
+
+42:37.360 --> 42:39.720
+ if you look at companies like Kitty Hawk,
+
+42:39.720 --> 42:42.080
+ working on electric, so the clean,
+
+42:43.880 --> 42:45.840
+ talking to Sebastian Thrun, right?
+
+42:45.840 --> 42:48.840
+ It's a crazy dream, you know?
+
+42:48.840 --> 42:52.080
+ But you work with flight a lot.
+
+42:52.080 --> 42:55.760
+ You've mentioned before that manned flights
+
+42:55.760 --> 43:00.760
+ or carrying a human body is very difficult to do.
+
+43:01.640 --> 43:04.240
+ So how crazy is flying cars?
+
+43:04.240 --> 43:05.400
+ Do you think there'll be a day
+
+43:05.400 --> 43:10.400
+ when we have vertical takeoff and landing vehicles
+
+43:11.080 --> 43:14.040
+ that are sufficiently affordable
+
+43:14.960 --> 43:17.440
+ that we're going to see a huge amount of them?
+
+43:17.440 --> 43:19.680
+ And they would look like something like we dream of
+
+43:19.680 --> 43:21.080
+ when we think about flying cars.
+
+43:21.080 --> 43:22.200
+ Yeah, like the Jetsons.
+
+43:22.200 --> 43:23.160
+ The Jetsons, yeah.
+
+43:23.160 --> 43:25.560
+ So look, there are a lot of smart people working on this
+
+43:25.560 --> 43:29.640
+ and you never say something is not possible
+
+43:29.640 --> 43:32.200
+ when you have people like Sebastian Thrun working on it.
+
+43:32.200 --> 43:35.160
+ So I totally think it's viable.
+
+43:35.160 --> 43:38.240
+ I question, again, the electric piece.
+
+43:38.240 --> 43:39.520
+ The electric piece, yeah.
+
+43:39.520 --> 43:41.680
+ And again, for short distances, you can do it.
+
+43:41.680 --> 43:43.640
+ And there's no reason to suggest
+
+43:43.640 --> 43:45.840
+ that these all just have to be rotorcrafts.
+
+43:45.840 --> 43:46.920
+ You take off vertically,
+
+43:46.920 --> 43:49.680
+ but then you morph into a forward flight.
+
+43:49.680 --> 43:51.600
+ I think there are a lot of interesting designs.
+
+43:51.600 --> 43:56.040
+ The question to me is, are these economically viable?
+
+43:56.040 --> 43:59.160
+ And if you agree to do this with fossil fuels,
+
+43:59.160 --> 44:01.960
+ it instantly immediately becomes viable.
+
+44:01.960 --> 44:03.480
+ That's a real challenge.
+
+44:03.480 --> 44:06.560
+ Do you think it's possible for robots and humans
+
+44:06.560 --> 44:08.840
+ to collaborate successfully on tasks?
+
+44:08.840 --> 44:13.640
+ So a lot of robotics folks that I talk to and work with,
+
+44:13.640 --> 44:18.000
+ I mean, humans just add a giant mess to the picture.
+
+44:18.000 --> 44:20.320
+ So it's best to remove them from consideration
+
+44:20.320 --> 44:22.400
+ when solving specific tasks.
+
+44:22.400 --> 44:23.600
+ It's very difficult to model.
+
+44:23.600 --> 44:26.000
+ There's just a source of uncertainty.
+
+44:26.000 --> 44:31.000
+ In your work with these agile flying robots,
+
+44:32.560 --> 44:35.680
+ do you think there's a role for collaboration with humans?
+
+44:35.680 --> 44:38.600
+ Or is it best to model tasks in a way
+
+44:38.600 --> 44:43.400
+ that doesn't have a human in the picture?
+
+44:43.400 --> 44:46.760
+ Well, I don't think we should ever think about robots
+
+44:46.760 --> 44:48.120
+ without human in the picture.
+
+44:48.120 --> 44:50.960
+ Ultimately, robots are there because we want them
+
+44:50.960 --> 44:54.360
+ to solve problems for humans.
+
+44:54.360 --> 44:58.280
+ But there's no general solution to this problem.
+
+44:58.280 --> 45:00.000
+ I think if you look at human interaction
+
+45:00.000 --> 45:02.400
+ and how humans interact with robots,
+
+45:02.400 --> 45:05.280
+ you know, we think of these in sort of three different ways.
+
+45:05.280 --> 45:07.600
+ One is the human commanding the robot.
+
+45:08.880 --> 45:12.880
+ The second is the human collaborating with the robot.
+
+45:12.880 --> 45:15.520
+ So for example, we work on how a robot
+
+45:15.520 --> 45:18.720
+ can actually pick up things with a human and carry things.
+
+45:18.720 --> 45:20.880
+ That's like true collaboration.
+
+45:20.880 --> 45:25.000
+ And third, we think about humans as bystanders,
+
+45:25.000 --> 45:27.240
+ self driving cars, what's the human's role
+
+45:27.240 --> 45:30.320
+ and how do self driving cars
+
+45:30.320 --> 45:32.920
+ acknowledge the presence of humans?
+
+45:32.920 --> 45:35.840
+ So I think all of these things are different scenarios.
+
+45:35.840 --> 45:38.480
+ It depends on what kind of humans, what kind of task.
+
+45:39.640 --> 45:41.840
+ And I think it's very difficult to say
+
+45:41.840 --> 45:45.520
+ that there's a general theory that we all have for this.
+
+45:45.520 --> 45:48.440
+ But at the same time, it's also silly to say
+
+45:48.440 --> 45:52.000
+ that we should think about robots independent of humans.
+
+45:52.000 --> 45:55.760
+ So to me, human robot interaction
+
+45:55.760 --> 45:59.760
+ is almost a mandatory aspect of everything we do.
+
+45:59.760 --> 46:02.440
+ Yes, but to which degree, so your thoughts,
+
+46:02.440 --> 46:05.240
+ if we jump to autonomous vehicles, for example,
+
+46:05.240 --> 46:08.680
+ there's a big debate between what's called
+
+46:08.680 --> 46:10.640
+ level two and level four.
+
+46:10.640 --> 46:13.680
+ So semi autonomous and autonomous vehicles.
+
+46:13.680 --> 46:16.440
+ And so the Tesla approach currently at least
+
+46:16.440 --> 46:18.960
+ has a lot of collaboration between human and machine.
+
+46:18.960 --> 46:22.040
+ So the human is supposed to actively supervise
+
+46:22.040 --> 46:23.880
+ the operation of the robot.
+
+46:23.880 --> 46:28.880
+ Part of the safety definition of how safe a robot is
+
+46:29.160 --> 46:32.880
+ in that case is how effective is the human in monitoring it.
+
+46:32.880 --> 46:37.880
+ Do you think that's ultimately not a good approach
+
+46:37.880 --> 46:42.360
+ in sort of having a human in the picture,
+
+46:42.360 --> 46:47.360
+ not as a bystander or part of the infrastructure,
+
+46:47.400 --> 46:50.000
+ but really as part of what's required
+
+46:50.000 --> 46:51.560
+ to make the system safe?
+
+46:51.560 --> 46:53.720
+ This is harder than it sounds.
+
+46:53.720 --> 46:58.200
+ I think, you know, if you, I mean,
+
+46:58.200 --> 47:01.360
+ I'm sure you've driven before in highways and so on.
+
+47:01.360 --> 47:06.120
+ It's really very hard to have to relinquish control
+
+47:06.120 --> 47:10.440
+ to a machine and then take over when needed.
+
+47:10.440 --> 47:12.280
+ So I think Tesla's approach is interesting
+
+47:12.280 --> 47:14.800
+ because it allows you to periodically establish
+
+47:14.800 --> 47:18.520
+ some kind of contact with the car.
+
+47:18.520 --> 47:20.640
+ Toyota, on the other hand, is thinking about
+
+47:20.640 --> 47:24.800
+ shared autonomy or collaborative autonomy as a paradigm.
+
+47:24.800 --> 47:27.480
+ If I may argue, these are very, very simple ways
+
+47:27.480 --> 47:29.680
+ of human robot collaboration,
+
+47:29.680 --> 47:31.880
+ because the task is pretty boring.
+
+47:31.880 --> 47:35.000
+ You sit in a vehicle, you go from point A to point B.
+
+47:35.000 --> 47:37.360
+ I think the more interesting thing to me is,
+
+47:37.360 --> 47:38.760
+ for example, search and rescue.
+
+47:38.760 --> 47:41.980
+ I've got a human first responder, robot first responders.
+
+47:43.160 --> 47:45.120
+ I gotta do something.
+
+47:45.120 --> 47:46.000
+ It's important.
+
+47:46.000 --> 47:47.800
+ I have to do it in two minutes.
+
+47:47.800 --> 47:49.240
+ The building is burning.
+
+47:49.240 --> 47:50.440
+ There's been an explosion.
+
+47:50.440 --> 47:51.360
+ It's collapsed.
+
+47:51.360 --> 47:52.800
+ How do I do it?
+
+47:52.800 --> 47:54.740
+ I think to me, those are the interesting things
+
+47:54.740 --> 47:57.160
+ where it's very, very unstructured.
+
+47:57.160 --> 47:58.480
+ And what's the role of the human?
+
+47:58.480 --> 48:00.200
+ What's the role of the robot?
+
+48:00.200 --> 48:02.440
+ Clearly, there's lots of interesting challenges
+
+48:02.440 --> 48:03.440
+ and there's a field.
+
+48:03.440 --> 48:05.760
+ I think we're gonna make a lot of progress in this area.
+
+48:05.760 --> 48:07.600
+ Yeah, it's an exciting form of collaboration.
+
+48:07.600 --> 48:08.440
+ You're right.
+
+48:08.440 --> 48:11.120
+ In autonomous driving, the main enemy
+
+48:11.120 --> 48:13.120
+ is just boredom of the human.
+
+48:13.120 --> 48:13.960
+ Yes.
+
+48:13.960 --> 48:15.680
+ As opposed to in rescue operations,
+
+48:15.680 --> 48:18.360
+ it's literally life and death.
+
+48:18.360 --> 48:22.080
+ And the collaboration enables
+
+48:22.080 --> 48:23.820
+ the effective completion of the mission.
+
+48:23.820 --> 48:24.760
+ So it's exciting.
+
+48:24.760 --> 48:27.400
+ In some sense, we're also doing this.
+
+48:27.400 --> 48:30.520
+ You think about the human driving a car
+
+48:30.520 --> 48:33.800
+ and almost invariably, the human's trying
+
+48:33.800 --> 48:35.000
+ to estimate the state of the car,
+
+48:35.000 --> 48:37.280
+ they estimate the state of the environment and so on.
+
+48:37.280 --> 48:40.120
+ But what if the car were to estimate the state of the human?
+
+48:40.120 --> 48:41.960
+ So for example, I'm sure you have a smartphone
+
+48:41.960 --> 48:44.580
+ and the smartphone tries to figure out what you're doing
+
+48:44.580 --> 48:48.320
+ and send you reminders and oftentimes telling you
+
+48:48.320 --> 48:49.540
+ to drive to a certain place,
+
+48:49.540 --> 48:51.400
+ although you have no intention of going there
+
+48:51.400 --> 48:53.880
+ because it thinks that that's where you should be
+
+48:53.880 --> 48:56.240
+ because of some Gmail calendar entry
+
+48:57.520 --> 48:58.960
+ or something like that.
+
+48:58.960 --> 49:01.600
+ And it's trying to constantly figure out who you are,
+
+49:01.600 --> 49:02.740
+ what you're doing.
+
+49:02.740 --> 49:04.200
+ If a car were to do that,
+
+49:04.200 --> 49:06.840
+ maybe that would make the driver safer
+
+49:06.840 --> 49:08.160
+ because the car is trying to figure out
+
+49:08.160 --> 49:09.760
+ is the driver paying attention,
+
+49:09.760 --> 49:11.600
+ looking at his or her eyes,
+
+49:12.480 --> 49:14.400
+ looking at circadian movements.
+
+49:14.400 --> 49:16.480
+ So I think the potential is there,
+
+49:16.480 --> 49:18.600
+ but from the reverse side,
+
+49:18.600 --> 49:21.640
+ it's not robot modeling, but it's human modeling.
+
+49:21.640 --> 49:22.880
+ It's more on the human, right.
+
+49:22.880 --> 49:25.320
+ And I think the robots can do a very good job
+
+49:25.320 --> 49:29.120
+ of modeling humans if you really think about the framework
+
+49:29.120 --> 49:32.640
+ that you have a human sitting in a cockpit,
+
+49:32.640 --> 49:35.820
+ surrounded by sensors, all staring at him,
+
+49:35.820 --> 49:37.860
+ in addition to be staring outside,
+
+49:37.860 --> 49:39.160
+ but also staring at him.
+
+49:39.160 --> 49:40.960
+ I think there's a real synergy there.
+
+49:40.960 --> 49:42.360
+ Yeah, I love that problem
+
+49:42.360 --> 49:45.560
+ because it's the new 21st century form of psychology,
+
+49:45.560 --> 49:48.520
+ actually AI enabled psychology.
+
+49:48.520 --> 49:51.280
+ A lot of people have sci fi inspired fears
+
+49:51.280 --> 49:54.080
+ of walking robots like those from Boston Dynamics.
+
+49:54.080 --> 49:56.480
+ If you just look at shows on Netflix and so on,
+
+49:56.480 --> 49:59.040
+ or flying robots like those you work with,
+
+49:59.920 --> 50:03.160
+ how would you, how do you think about those fears?
+
+50:03.160 --> 50:05.040
+ How would you alleviate those fears?
+
+50:05.040 --> 50:09.040
+ Do you have inklings, echoes of those same concerns?
+
+50:09.040 --> 50:11.760
+ You know, anytime we develop a technology
+
+50:11.760 --> 50:14.160
+ meaning to have positive impact in the world,
+
+50:14.160 --> 50:15.780
+ there's always the worry that,
+
+50:17.440 --> 50:21.000
+ you know, somebody could subvert those technologies
+
+50:21.000 --> 50:23.280
+ and use it in an adversarial setting.
+
+50:23.280 --> 50:25.280
+ And robotics is no exception, right?
+
+50:25.280 --> 50:29.280
+ So I think it's very easy to weaponize robots.
+
+50:29.280 --> 50:30.880
+ I think we talk about swarms.
+
+50:31.720 --> 50:33.960
+ One thing I worry a lot about is,
+
+50:33.960 --> 50:35.880
+ so, you know, for us to get swarms to work
+
+50:35.880 --> 50:38.280
+ and do something reliably, it's really hard.
+
+50:38.280 --> 50:42.040
+ But suppose I have this challenge
+
+50:42.040 --> 50:44.360
+ of trying to destroy something,
+
+50:44.360 --> 50:45.720
+ and I have a swarm of robots,
+
+50:45.720 --> 50:47.280
+ where only one out of the swarm
+
+50:47.280 --> 50:48.920
+ needs to get to its destination.
+
+50:48.920 --> 50:52.640
+ So that suddenly becomes a lot more doable.
+
+50:52.640 --> 50:54.720
+ And so I worry about, you know,
+
+50:54.720 --> 50:56.920
+ this general idea of using autonomy
+
+50:56.920 --> 50:58.600
+ with lots and lots of agents.
+
+51:00.040 --> 51:01.320
+ I mean, having said that, look,
+
+51:01.320 --> 51:03.760
+ a lot of this technology is not very mature.
+
+51:03.760 --> 51:05.520
+ My favorite saying is that
+
+51:06.560 --> 51:10.520
+ if somebody had to develop this technology,
+
+51:10.520 --> 51:12.320
+ wouldn't you rather the good guys do it?
+
+51:12.320 --> 51:13.880
+ So the good guys have a good understanding
+
+51:13.880 --> 51:15.560
+ of the technology, so they can figure out
+
+51:15.560 --> 51:18.320
+ how this technology is being used in a bad way,
+
+51:18.320 --> 51:21.360
+ or could be used in a bad way and try to defend against it.
+
+51:21.360 --> 51:22.760
+ So we think a lot about that.
+
+51:22.760 --> 51:25.400
+ So we have, we're doing research
+
+51:25.400 --> 51:28.240
+ on how to defend against swarms, for example.
+
+51:28.240 --> 51:29.600
+ That's interesting.
+
+51:29.600 --> 51:32.960
+ There's in fact a report by the National Academies
+
+51:32.960 --> 51:35.520
+ on counter UAS technologies.
+
+51:36.680 --> 51:38.200
+ This is a real threat,
+
+51:38.200 --> 51:40.320
+ but we're also thinking about how to defend against this
+
+51:40.320 --> 51:42.920
+ and knowing how swarms work.
+
+51:42.920 --> 51:47.160
+ Knowing how autonomy works is, I think, very important.
+
+51:47.160 --> 51:49.280
+ So it's not just politicians?
+
+51:49.280 --> 51:51.640
+ Do you think engineers have a role in this discussion?
+
+51:51.640 --> 51:52.480
+ Absolutely.
+
+51:52.480 --> 51:55.280
+ I think the days where politicians
+
+51:55.280 --> 51:57.680
+ can be agnostic to technology are gone.
+
+51:59.200 --> 52:02.640
+ I think every politician needs to be
+
+52:03.840 --> 52:05.680
+ literate in technology.
+
+52:05.680 --> 52:08.640
+ And I often say technology is the new liberal art.
+
+52:09.800 --> 52:12.920
+ Understanding how technology will change your life,
+
+52:12.920 --> 52:14.480
+ I think is important.
+
+52:14.480 --> 52:18.080
+ And every human being needs to understand that.
+
+52:18.080 --> 52:20.160
+ And maybe we can elect some engineers
+
+52:20.160 --> 52:22.720
+ to office as well on the other side.
+
+52:22.720 --> 52:24.840
+ What are the biggest open problems in robotics?
+
+52:24.840 --> 52:27.760
+ And you said we're in the early days in some sense.
+
+52:27.760 --> 52:31.040
+ What are the problems we would like to solve in robotics?
+
+52:31.040 --> 52:32.520
+ I think there are lots of problems, right?
+
+52:32.520 --> 52:36.440
+ But I would phrase it in the following way.
+
+52:36.440 --> 52:39.520
+ If you look at the robots we're building,
+
+52:39.520 --> 52:43.160
+ they're still very much tailored towards
+
+52:43.160 --> 52:46.520
+ doing specific tasks and specific settings.
+
+52:46.520 --> 52:49.480
+ I think the question of how do you get them to operate
+
+52:49.480 --> 52:51.080
+ in much broader settings
+
+52:53.560 --> 52:58.040
+ where things can change in unstructured environments
+
+52:58.040 --> 52:59.160
+ is up in the air.
+
+52:59.160 --> 53:01.200
+ So think of self driving cars.
+
+53:02.920 --> 53:05.680
+ Today, we can build a self driving car in a parking lot.
+
+53:05.680 --> 53:09.000
+ We can do level five autonomy in a parking lot.
+
+53:10.040 --> 53:13.240
+ But can you do a level five autonomy
+
+53:13.240 --> 53:16.840
+ in the streets of Napoli in Italy or Mumbai in India?
+
+53:16.840 --> 53:17.760
+ No.
+
+53:17.760 --> 53:22.400
+ So in some sense, when we think about robotics,
+
+53:22.400 --> 53:25.120
+ we have to think about where they're functioning,
+
+53:25.120 --> 53:27.760
+ what kind of environment, what kind of a task.
+
+53:27.760 --> 53:29.800
+ We have no understanding
+
+53:29.800 --> 53:32.800
+ of how to put both those things together.
+
+53:32.800 --> 53:34.000
+ So we're in the very early days
+
+53:34.000 --> 53:35.920
+ of applying it to the physical world.
+
+53:35.920 --> 53:38.800
+ And I was just in Naples actually.
+
+53:38.800 --> 53:42.200
+ And there's levels of difficulty and complexity
+
+53:42.200 --> 53:45.880
+ depending on which area you're applying it to.
+
+53:45.880 --> 53:46.720
+ I think so.
+
+53:46.720 --> 53:49.320
+ And we don't have a systematic way of understanding that.
+
+53:51.040 --> 53:53.800
+ Everybody says, just because a computer
+
+53:53.800 --> 53:56.520
+ can now beat a human at any board game,
+
+53:56.520 --> 53:59.920
+ we certainly know something about intelligence.
+
+53:59.920 --> 54:01.360
+ That's not true.
+
+54:01.360 --> 54:04.400
+ A computer board game is very, very structured.
+
+54:04.400 --> 54:08.480
+ It is the equivalent of working in a Henry Ford factory
+
+54:08.480 --> 54:11.680
+ where things, parts come, you assemble, move on.
+
+54:11.680 --> 54:14.120
+ It's a very, very, very structured setting.
+
+54:14.120 --> 54:15.680
+ That's the easiest thing.
+
+54:15.680 --> 54:17.040
+ And we know how to do that.
+
+54:18.400 --> 54:20.400
+ So you've done a lot of incredible work
+
+54:20.400 --> 54:23.720
+ at the UPenn, University of Pennsylvania, GraspLab.
+
+54:23.720 --> 54:26.560
+ You're now Dean of Engineering at UPenn.
+
+54:26.560 --> 54:31.320
+ What advice do you have for a new bright eyed undergrad
+
+54:31.320 --> 54:34.640
+ interested in robotics or AI or engineering?
+
+54:34.640 --> 54:36.560
+ Well, I think there's really three things.
+
+54:36.560 --> 54:40.600
+ One is you have to get used to the idea
+
+54:40.600 --> 54:42.840
+ that the world will not be the same in five years
+
+54:42.840 --> 54:45.160
+ or four years whenever you graduate, right?
+
+54:45.160 --> 54:46.120
+ Which is really hard to do.
+
+54:46.120 --> 54:48.960
+ So this thing about predicting the future,
+
+54:48.960 --> 54:50.520
+ every one of us needs to be trying
+
+54:50.520 --> 54:52.360
+ to predict the future always.
+
+54:53.280 --> 54:54.960
+ Not because you'll be any good at it,
+
+54:54.960 --> 54:56.440
+ but by thinking about it,
+
+54:56.440 --> 55:00.880
+ I think you sharpen your senses and you become smarter.
+
+55:00.880 --> 55:02.080
+ So that's number one.
+
+55:02.080 --> 55:05.760
+ Number two, it's a corollary of the first piece,
+
+55:05.760 --> 55:09.360
+ which is you really don't know what's gonna be important.
+
+55:09.360 --> 55:12.080
+ So this idea that I'm gonna specialize in something
+
+55:12.080 --> 55:15.320
+ which will allow me to go in a particular direction,
+
+55:15.320 --> 55:16.480
+ it may be interesting,
+
+55:16.480 --> 55:18.480
+ but it's important also to have this breadth
+
+55:18.480 --> 55:20.360
+ so you have this jumping off point.
+
+55:22.000 --> 55:23.000
+ I think the third thing,
+
+55:23.000 --> 55:25.360
+ and this is where I think Penn excels.
+
+55:25.360 --> 55:27.240
+ I mean, we teach engineering,
+
+55:27.240 --> 55:29.960
+ but it's always in the context of the liberal arts.
+
+55:29.960 --> 55:32.360
+ It's always in the context of society.
+
+55:32.360 --> 55:35.840
+ As engineers, we cannot afford to lose sight of that.
+
+55:35.840 --> 55:37.640
+ So I think that's important.
+
+55:37.640 --> 55:39.960
+ But I think one thing that people underestimate
+
+55:39.960 --> 55:40.920
+ when they do robotics
+
+55:40.920 --> 55:43.440
+ is the importance of mathematical foundations,
+
+55:43.440 --> 55:46.880
+ the importance of representations.
+
+55:47.720 --> 55:50.040
+ Not everything can just be solved
+
+55:50.040 --> 55:52.440
+ by looking for Ross packages on the internet
+
+55:52.440 --> 55:56.280
+ or to find a deep neural network that works.
+
+55:56.280 --> 55:59.080
+ I think the representation question is key,
+
+55:59.080 --> 56:00.400
+ even to machine learning,
+
+56:00.400 --> 56:05.400
+ where if you ever hope to achieve or get to explainable AI,
+
+56:05.400 --> 56:07.760
+ somehow there need to be representations
+
+56:07.760 --> 56:09.080
+ that you can understand.
+
+56:09.080 --> 56:11.120
+ So if you wanna do robotics,
+
+56:11.120 --> 56:12.680
+ you should also do mathematics.
+
+56:12.680 --> 56:15.080
+ And you said liberal arts, a little literature.
+
+56:16.160 --> 56:17.200
+ If you wanna build a robot,
+
+56:17.200 --> 56:19.320
+ it should be reading Dostoyevsky.
+
+56:19.320 --> 56:20.360
+ I agree with that.
+
+56:20.360 --> 56:21.200
+ Very good.
+
+56:21.200 --> 56:23.560
+ So Vijay, thank you so much for talking today.
+
+56:23.560 --> 56:24.400
+ It was an honor.
+
+56:24.400 --> 56:25.240
+ Thank you.
+
+56:25.240 --> 56:26.200
+ It was just a very exciting conversation.
+
+56:26.200 --> 56:46.200
+ Thank you.
+

vtt/episode_037_small.vtt

@@ -0,0 +1,3143 @@
+WEBVTT
+
+00:00.000 --> 00:03.160
+ The following is a conversation with Vijay Kumar.
+
+00:03.160 --> 00:05.800
+ He's one of the top roboticists in the world,
+
+00:05.800 --> 00:08.760
+ a professor at the University of Pennsylvania,
+
+00:08.760 --> 00:10.680
+ a Dean of Penn Engineering,
+
+00:10.680 --> 00:12.880
+ former director of Grasp Lab,
+
+00:12.880 --> 00:15.320
+ or the General Robotics Automation Sensing
+
+00:15.320 --> 00:17.560
+ and Perception Laboratory at Penn,
+
+00:17.560 --> 00:22.560
+ that was established back in 1979, that's 40 years ago.
+
+00:22.600 --> 00:24.720
+ Vijay is perhaps best known
+
+00:24.720 --> 00:28.520
+ for his work in multi robot systems, robot swarms,
+
+00:28.520 --> 00:30.880
+ and micro aerial vehicles.
+
+00:30.880 --> 00:34.040
+ Robots that elegantly cooperate in flight
+
+00:34.040 --> 00:36.200
+ under all the uncertainty and challenges
+
+00:36.200 --> 00:38.760
+ that the real world conditions present.
+
+00:38.760 --> 00:41.960
+ This is the Artificial Intelligence Podcast.
+
+00:41.960 --> 00:44.320
+ If you enjoy it, subscribe on YouTube,
+
+00:44.320 --> 00:46.080
+ give it five stars on iTunes,
+
+00:46.080 --> 00:47.560
+ support it on Patreon,
+
+00:47.560 --> 00:49.480
+ or simply connect with me on Twitter
+
+00:49.480 --> 00:53.280
+ at Lex Freedman spelled FRID MAN.
+
+00:53.280 --> 00:57.560
+ And now here's my conversation with Vijay Kumar.
+
+00:58.680 --> 01:01.080
+ What is the first robot you've ever built
+
+01:01.080 --> 01:02.840
+ or were a part of building?
+
+01:02.840 --> 01:04.760
+ Way back when I was in graduate school,
+
+01:04.760 --> 01:06.760
+ I was part of a fairly big project
+
+01:06.760 --> 01:11.760
+ that involved building a very large hexapod.
+
+01:12.000 --> 01:17.000
+ This weighed close to 7,000 pounds.
+
+01:17.520 --> 01:21.600
+ And it was powered by hydraulic actuation,
+
+01:21.600 --> 01:26.600
+ or was actuated by hydraulics with 18 motors,
+
+01:27.760 --> 01:32.760
+ hydraulic motors, each controlled by an Intel 8085 processor
+
+01:34.200 --> 01:36.680
+ and an 8086 co processor.
+
+01:38.160 --> 01:43.160
+ And so imagine this huge monster that had 18 joints,
+
+01:44.840 --> 01:47.000
+ each controlled by an independent computer.
+
+01:47.000 --> 01:48.560
+ And there was a 19th computer
+
+01:48.560 --> 01:50.160
+ that actually did the coordination
+
+01:50.160 --> 01:52.360
+ between these 18 joints.
+
+01:52.360 --> 01:53.760
+ So as part of this project,
+
+01:53.760 --> 01:57.960
+ and my thesis work was,
+
+01:57.960 --> 02:01.080
+ how do you coordinate the 18 legs?
+
+02:02.120 --> 02:06.360
+ And in particular, the pressures in the hydraulic cylinders
+
+02:06.360 --> 02:09.240
+ to get efficient locomotion.
+
+02:09.240 --> 02:11.680
+ It sounds like a giant mess.
+
+02:11.680 --> 02:14.480
+ So how difficult is it to make all the motors communicate?
+
+02:14.480 --> 02:16.880
+ Presumably you have to send signals
+
+02:16.880 --> 02:18.720
+ hundreds of times a second, or at least...
+
+02:18.720 --> 02:22.800
+ This was not my work, but the folks who worked on this
+
+02:22.800 --> 02:24.200
+ wrote what I believe to be
+
+02:24.200 --> 02:26.640
+ the first multiprocessor operating system.
+
+02:26.640 --> 02:27.960
+ This was in the 80s.
+
+02:29.080 --> 02:32.240
+ And you had to make sure that obviously messages
+
+02:32.240 --> 02:34.640
+ got across from one joint to another.
+
+02:34.640 --> 02:37.960
+ You have to remember the clock speeds on those computers
+
+02:37.960 --> 02:39.640
+ were about half a megahertz.
+
+02:39.640 --> 02:40.480
+ Right.
+
+02:40.480 --> 02:42.200
+ So the 80s.
+
+02:42.200 --> 02:45.320
+ So not to romanticize the notion,
+
+02:45.320 --> 02:47.960
+ but how did it make you feel to make,
+
+02:47.960 --> 02:49.680
+ to see that robot move?
+
+02:51.040 --> 02:52.240
+ It was amazing.
+
+02:52.240 --> 02:54.160
+ In hindsight, it looks like, well,
+
+02:54.160 --> 02:57.240
+ we built the thing which really should have been much smaller.
+
+02:57.240 --> 02:59.080
+ And of course, today's robots are much smaller.
+
+02:59.080 --> 03:02.200
+ You look at, you know, Boston Dynamics,
+
+03:02.200 --> 03:04.720
+ our ghost robotics has been off from pen.
+
+03:06.000 --> 03:08.640
+ But back then, you were stuck
+
+03:08.640 --> 03:11.120
+ with the substrate you had, the compute you had,
+
+03:11.120 --> 03:13.640
+ so things were unnecessarily big.
+
+03:13.640 --> 03:18.000
+ But at the same time, and this is just human psychology,
+
+03:18.000 --> 03:20.380
+ somehow bigger means grander.
+
+03:21.280 --> 03:23.600
+ You know, people never have the same appreciation
+
+03:23.600 --> 03:26.320
+ for nanotechnology or nano devices
+
+03:26.320 --> 03:30.120
+ as they do for the space shuttle or the Boeing 747.
+
+03:30.120 --> 03:32.720
+ Yeah, you've actually done quite a good job
+
+03:32.720 --> 03:35.960
+ at illustrating that small is beautiful
+
+03:35.960 --> 03:37.680
+ in terms of robotics.
+
+03:37.680 --> 03:42.520
+ So what is on that topic is the most beautiful
+
+03:42.520 --> 03:46.200
+ or elegant robot emotion that you've ever seen.
+
+03:46.200 --> 03:47.880
+ Not to pick favorites or whatever,
+
+03:47.880 --> 03:51.000
+ but something that just inspires you that you remember.
+
+03:51.000 --> 03:54.000
+ Well, I think the thing that I'm most proud of
+
+03:54.000 --> 03:57.200
+ that my students have done is really think about
+
+03:57.200 --> 04:00.360
+ small UAVs that can maneuver and constrain spaces
+
+04:00.360 --> 04:03.640
+ and in particular, their ability to coordinate
+
+04:03.640 --> 04:06.760
+ with each other and form three dimensional patterns.
+
+04:06.760 --> 04:08.920
+ So once you can do that,
+
+04:08.920 --> 04:13.920
+ you can essentially create 3D objects in the sky
+
+04:17.000 --> 04:19.800
+ and you can deform these objects on the fly.
+
+04:19.800 --> 04:23.520
+ So in some sense, your toolbox of what you can create
+
+04:23.520 --> 04:25.300
+ has suddenly got enhanced.
+
+04:27.400 --> 04:29.920
+ And before that, we did the two dimensional version of this.
+
+04:29.920 --> 04:33.760
+ So we had ground robots forming patterns and so on.
+
+04:33.760 --> 04:37.080
+ So that was not as impressive, that was not as beautiful.
+
+04:37.080 --> 04:40.480
+ But if you do it in 3D, suspend it in midair
+
+04:40.480 --> 04:43.640
+ and you've got to go back to 2011 when we did this.
+
+04:43.640 --> 04:45.960
+ Now it's actually pretty standard to do these things
+
+04:45.960 --> 04:49.800
+ eight years later, but back then it was a big accomplishment.
+
+04:49.800 --> 04:52.440
+ So the distributed cooperation
+
+04:52.440 --> 04:55.640
+ is where beauty emerges in your eyes?
+
+04:55.640 --> 04:57.960
+ Well, I think beauty to an engineer is very different
+
+04:57.960 --> 05:01.520
+ from beauty to someone who's looking at robots
+
+05:01.520 --> 05:03.400
+ from the outside, if you will.
+
+05:03.400 --> 05:07.920
+ But what I meant there, so before we said that grand
+
+05:07.920 --> 05:10.480
+ is associated with size.
+
+05:10.480 --> 05:13.640
+ And another way of thinking about this
+
+05:13.640 --> 05:16.480
+ is just the physical shape and the idea
+
+05:16.480 --> 05:18.320
+ that you can create physical shapes in midair
+
+05:18.320 --> 05:21.520
+ and have them deform, that's beautiful.
+
+05:21.520 --> 05:23.000
+ But the individual components,
+
+05:23.000 --> 05:24.840
+ the agility is beautiful too, right?
+
+05:24.840 --> 05:25.680
+ That is true too.
+
+05:25.680 --> 05:28.400
+ So then how quickly can you actually manipulate
+
+05:28.400 --> 05:29.560
+ these three dimensional shapes
+
+05:29.560 --> 05:31.200
+ and the individual components?
+
+05:31.200 --> 05:32.200
+ Yes, you're right.
+
+05:32.200 --> 05:36.760
+ Oh, by the way, said UAV, unmanned aerial vehicle.
+
+05:36.760 --> 05:41.760
+ What's a good term for drones, UAVs, quadcopters?
+
+05:41.840 --> 05:44.520
+ Is there a term that's being standardized?
+
+05:44.520 --> 05:45.440
+ I don't know if there is.
+
+05:45.440 --> 05:47.880
+ Everybody wants to use the word drones.
+
+05:47.880 --> 05:49.760
+ And I've often said there's drones to me
+
+05:49.760 --> 05:51.000
+ is a pejorative word.
+
+05:51.000 --> 05:53.960
+ It signifies something that's dumb,
+
+05:53.960 --> 05:56.320
+ a pre program that does one little thing
+
+05:56.320 --> 05:58.600
+ and robots are anything but drones.
+
+05:58.600 --> 06:00.680
+ So I actually don't like that word,
+
+06:00.680 --> 06:02.960
+ but that's what everybody uses.
+
+06:02.960 --> 06:04.880
+ You could call it unpiloted.
+
+06:04.880 --> 06:05.800
+ Unpiloted.
+
+06:05.800 --> 06:08.080
+ But even unpiloted could be radio controlled,
+
+06:08.080 --> 06:10.480
+ could be remotely controlled in many different ways.
+
+06:11.560 --> 06:12.680
+ And I think the right word
+
+06:12.680 --> 06:15.040
+ is thinking about it as an aerial robot.
+
+06:15.040 --> 06:19.080
+ You also say agile, autonomous aerial robot, right?
+
+06:19.080 --> 06:20.600
+ Yeah, so agility is an attribute,
+
+06:20.600 --> 06:22.200
+ but they don't have to be.
+
+06:23.080 --> 06:24.800
+ So what biological system,
+
+06:24.800 --> 06:26.880
+ because you've also drawn a lot of inspiration
+
+06:26.880 --> 06:28.640
+ with those I've seen bees and ants
+
+06:28.640 --> 06:32.400
+ that you've talked about, what living creatures
+
+06:32.400 --> 06:35.280
+ have you found to be most inspiring
+
+06:35.280 --> 06:38.560
+ as an engineer, instructive in your work in robotics?
+
+06:38.560 --> 06:43.480
+ To me, so ants are really quite incredible creatures, right?
+
+06:43.480 --> 06:47.920
+ So you, I mean, the individuals arguably are very simple
+
+06:47.920 --> 06:52.400
+ in how they're built, and yet they're incredibly resilient
+
+06:52.400 --> 06:54.000
+ as a population.
+
+06:54.000 --> 06:56.800
+ And as individuals, they're incredibly robust.
+
+06:56.800 --> 07:01.800
+ So, if you take an ant with six legs, you remove one leg,
+
+07:02.080 --> 07:04.160
+ it still works just fine.
+
+07:04.160 --> 07:05.800
+ And it moves along,
+
+07:05.800 --> 07:08.800
+ and I don't know that he even realizes it's lost a leg.
+
+07:09.800 --> 07:13.480
+ So that's the robustness at the individual ant level.
+
+07:13.480 --> 07:15.400
+ But then you look about this instinct
+
+07:15.400 --> 07:17.760
+ for self preservation of the colonies,
+
+07:17.760 --> 07:20.440
+ and they adapt in so many amazing ways,
+
+07:20.440 --> 07:25.440
+ you know, transcending gaps by just chaining themselves
+
+07:27.680 --> 07:30.400
+ together when you have a flood,
+
+07:30.400 --> 07:33.160
+ being able to recruit other teammates
+
+07:33.160 --> 07:35.200
+ to carry big morsels of food,
+
+07:36.560 --> 07:38.240
+ and then going out in different directions,
+
+07:38.240 --> 07:39.560
+ looking for food,
+
+07:39.560 --> 07:43.040
+ and then being able to demonstrate consensus,
+
+07:43.880 --> 07:47.240
+ even though they don't communicate directly
+
+07:47.240 --> 07:50.400
+ with each other the way we communicate with each other,
+
+07:50.400 --> 07:53.840
+ in some sense, they also know how to do democracy,
+
+07:53.840 --> 07:55.480
+ probably better than what we do.
+
+07:55.480 --> 07:59.000
+ Yeah, somehow, even democracy is emergent.
+
+07:59.000 --> 08:00.640
+ It seems like all of the phenomena
+
+08:00.640 --> 08:02.480
+ that we see is all emergent.
+
+08:02.480 --> 08:05.600
+ It seems like there's no centralized communicator.
+
+08:05.600 --> 08:09.840
+ There is, so I think a lot is made about that word, emergent,
+
+08:09.840 --> 08:11.560
+ and it means lots of things to different people,
+
+08:11.560 --> 08:12.600
+ but you're absolutely right.
+
+08:12.600 --> 08:17.600
+ I think as an engineer, you think about what element,
+
+08:17.600 --> 08:22.600
+ elemental behaviors, what primitives you could synthesize
+
+08:22.720 --> 08:26.640
+ so that the whole looks incredibly powerful,
+
+08:26.640 --> 08:27.920
+ incredibly synergistic,
+
+08:27.920 --> 08:30.960
+ the whole definitely being greater than some of the parts,
+
+08:30.960 --> 08:33.760
+ and ants are living proof of that.
+
+08:33.760 --> 08:36.280
+ So when you see these beautiful swarms
+
+08:36.280 --> 08:38.800
+ where there's biological systems of robots,
+
+08:39.920 --> 08:41.560
+ do you sometimes think of them
+
+08:41.560 --> 08:45.920
+ as a single individual living intelligent organism?
+
+08:45.920 --> 08:49.400
+ So it's the same as thinking of our human civilization
+
+08:49.400 --> 08:52.920
+ as one organism, or do you still, as an engineer,
+
+08:52.920 --> 08:54.560
+ think about the individual components
+
+08:54.560 --> 08:55.840
+ and all the engineering that went into
+
+08:55.840 --> 08:57.280
+ the individual components?
+
+08:57.280 --> 08:58.600
+ Well, that's very interesting.
+
+08:58.600 --> 09:01.440
+ So again, philosophically, as engineers,
+
+09:01.440 --> 09:06.440
+ what we want to do is to go beyond the individual components,
+
+09:06.800 --> 09:10.240
+ the individual units, and think about it as a unit,
+
+09:10.240 --> 09:11.480
+ as a cohesive unit,
+
+09:11.480 --> 09:14.240
+ without worrying about the individual components.
+
+09:14.240 --> 09:19.240
+ If you start obsessing about the individual building blocks
+
+09:19.680 --> 09:24.680
+ and what they do, you inevitably will find it hard
+
+09:26.400 --> 09:27.920
+ to scale up.
+
+09:27.920 --> 09:28.960
+ Just mathematically,
+
+09:28.960 --> 09:31.560
+ just think about individual things you want to model,
+
+09:31.560 --> 09:34.000
+ and if you want to have 10 of those,
+
+09:34.000 --> 09:36.400
+ then you essentially are taking Cartesian products
+
+09:36.400 --> 09:39.280
+ of 10 things, and that makes it really complicated
+
+09:39.280 --> 09:41.800
+ than to do any kind of synthesis or design
+
+09:41.800 --> 09:44.160
+ in that high dimension space is really hard.
+
+09:44.160 --> 09:45.840
+ So the right way to do this
+
+09:45.840 --> 09:49.080
+ is to think about the individuals in a clever way
+
+09:49.080 --> 09:51.160
+ so that at the higher level,
+
+09:51.160 --> 09:53.400
+ when you look at lots and lots of them,
+
+09:53.400 --> 09:55.320
+ abstractly, you can think of them
+
+09:55.320 --> 09:57.120
+ in some low dimensional space.
+
+09:57.120 --> 09:58.680
+ So what does that involve?
+
+09:58.680 --> 10:02.160
+ For the individual, you have to try to make
+
+10:02.160 --> 10:05.160
+ the way they see the world as local as possible,
+
+10:05.160 --> 10:06.440
+ and the other thing,
+
+10:06.440 --> 10:09.560
+ do you just have to make them robust to collisions?
+
+10:09.560 --> 10:10.880
+ Like you said, with the ants,
+
+10:10.880 --> 10:15.320
+ if something fails, the whole swarm doesn't fail.
+
+10:15.320 --> 10:17.760
+ Right, I think as engineers, we do this.
+
+10:17.760 --> 10:18.840
+ I mean, you know, think about,
+
+10:18.840 --> 10:21.280
+ we build planes or we build iPhones,
+
+10:22.240 --> 10:26.280
+ and we know that by taking individual components,
+
+10:26.280 --> 10:27.600
+ well engineered components,
+
+10:27.600 --> 10:30.080
+ with well specified interfaces
+
+10:30.080 --> 10:31.680
+ that behave in a predictable way,
+
+10:31.680 --> 10:33.560
+ you can build complex systems.
+
+10:34.400 --> 10:36.840
+ So that's ingrained I would claim
+
+10:36.840 --> 10:39.400
+ in most engineers thinking,
+
+10:39.400 --> 10:41.600
+ and it's true for computer scientists as well.
+
+10:41.600 --> 10:44.760
+ I think what's different here is that you want
+
+10:44.760 --> 10:49.480
+ the individuals to be robust in some sense,
+
+10:49.480 --> 10:52.000
+ as we do in these other settings,
+
+10:52.000 --> 10:54.480
+ but you also want some degree of resiliency
+
+10:54.480 --> 10:56.320
+ for the population.
+
+10:56.320 --> 10:58.720
+ And so you really want them to be able
+
+10:58.720 --> 11:03.720
+ to reestablish communication with their neighbors.
+
+11:03.840 --> 11:07.320
+ You want them to rethink their strategy
+
+11:07.320 --> 11:09.040
+ for group behavior.
+
+11:09.040 --> 11:11.000
+ You want them to reorganize.
+
+11:12.440 --> 11:16.120
+ And that's where I think a lot of the challenges lie.
+
+11:16.120 --> 11:18.400
+ So just at a high level,
+
+11:18.400 --> 11:21.120
+ what does it take for a bunch of,
+
+11:22.400 --> 11:23.560
+ what should we call them,
+
+11:23.560 --> 11:26.920
+ flying robots to create a formation?
+
+11:26.920 --> 11:30.400
+ Just for people who are not familiar with robotics
+
+11:30.400 --> 11:33.000
+ in general, how much information is needed?
+
+11:33.000 --> 11:36.040
+ How do you even make it happen
+
+11:36.040 --> 11:39.720
+ without a centralized controller?
+
+11:39.720 --> 11:41.320
+ So I mean, there are a couple of different ways
+
+11:41.320 --> 11:43.400
+ of looking at this.
+
+11:43.400 --> 11:48.400
+ If you are a purist, you think of it as a way
+
+11:50.040 --> 11:52.160
+ of recreating what nature does.
+
+11:53.800 --> 11:58.680
+ So nature forms groups for several reasons,
+
+11:58.680 --> 12:02.200
+ but mostly it's because of this instinct
+
+12:02.200 --> 12:07.200
+ that organisms have of preserving their colonies,
+
+12:07.280 --> 12:11.200
+ their population, which means what?
+
+12:11.200 --> 12:14.640
+ You need shelter, you need food, you need to procreate,
+
+12:14.640 --> 12:16.480
+ and that's basically it.
+
+12:16.480 --> 12:20.120
+ So the kinds of interactions you see are all organic.
+
+12:20.120 --> 12:21.320
+ They're all local.
+
+12:22.320 --> 12:25.760
+ And the only information that they share,
+
+12:25.760 --> 12:27.800
+ and mostly it's indirectly,
+
+12:27.800 --> 12:31.040
+ is to again preserve the herd or the flock
+
+12:31.040 --> 12:36.040
+ or the swarm and either by looking for new sources of food
+
+12:39.400 --> 12:41.240
+ or looking for new shelters, right?
+
+12:42.960 --> 12:47.200
+ As engineers, when we build swarms, we have a mission.
+
+12:48.280 --> 12:50.760
+ And when you think of a mission,
+
+12:52.080 --> 12:54.360
+ and it involves mobility,
+
+12:54.360 --> 12:56.840
+ most often it's described in some kind
+
+12:56.840 --> 12:58.800
+ of a global coordinate system.
+
+12:58.800 --> 13:03.080
+ As a human, as an operator, as a commander,
+
+13:03.080 --> 13:07.120
+ or as a collaborator, I have my coordinate system
+
+13:07.120 --> 13:10.160
+ and I want the robots to be consistent with that.
+
+13:11.120 --> 13:14.720
+ So I might think of it slightly differently.
+
+13:14.720 --> 13:18.960
+ I might want the robots to recognize that coordinate system,
+
+13:18.960 --> 13:21.360
+ which means not only do they have to think locally
+
+13:21.360 --> 13:23.160
+ in terms of who their immediate neighbors are,
+
+13:23.160 --> 13:24.640
+ but they have to be cognizant
+
+13:24.640 --> 13:28.320
+ of what the global environment looks like.
+
+13:28.320 --> 13:31.080
+ So if I go, if I say surround this building
+
+13:31.080 --> 13:33.280
+ and protect this from intruders,
+
+13:33.280 --> 13:35.160
+ well, they're immediately in a building
+
+13:35.160 --> 13:36.520
+ centered coordinate system
+
+13:36.520 --> 13:38.720
+ and I have to tell them where the building is.
+
+13:38.720 --> 13:40.080
+ And they're globally collaborating
+
+13:40.080 --> 13:41.360
+ on the map of that building.
+
+13:41.360 --> 13:44.240
+ They're maintaining some kind of global,
+
+13:44.240 --> 13:45.560
+ not just in the frame of the building,
+
+13:45.560 --> 13:49.040
+ but there's information that's ultimately being built up
+
+13:49.040 --> 13:53.320
+ explicitly as opposed to kind of implicitly,
+
+13:53.320 --> 13:54.400
+ like nature might.
+
+13:54.400 --> 13:55.240
+ Correct, correct.
+
+13:55.240 --> 13:57.720
+ So in some sense, nature is very, very sophisticated,
+
+13:57.720 --> 14:00.480
+ but the tasks that nature solves
+
+14:00.480 --> 14:03.040
+ or needs to solve are very different
+
+14:03.040 --> 14:05.160
+ from the kind of engineered tasks,
+
+14:05.160 --> 14:09.800
+ artificial tasks that we are forced to address.
+
+14:09.800 --> 14:12.560
+ And again, there's nothing preventing us
+
+14:12.560 --> 14:15.200
+ from solving these other problems,
+
+14:15.200 --> 14:16.640
+ but ultimately it's about impact.
+
+14:16.640 --> 14:19.400
+ You want these swarms to do something useful.
+
+14:19.400 --> 14:24.400
+ And so you're kind of driven into this very unnatural,
+
+14:24.400 --> 14:27.840
+ if you will, unnatural meaning, not like how nature does,
+
+14:27.840 --> 14:29.000
+ setting.
+
+14:29.000 --> 14:31.720
+ And it's probably a little bit more expensive
+
+14:31.720 --> 14:33.560
+ to do it the way nature does,
+
+14:33.560 --> 14:38.560
+ because nature is less sensitive to the loss of the individual
+
+14:39.280 --> 14:42.080
+ and cost wise in robotics,
+
+14:42.080 --> 14:45.280
+ I think you're more sensitive to losing individuals.
+
+14:45.280 --> 14:48.800
+ I think that's true, although if you look at the price
+
+14:48.800 --> 14:51.320
+ to performance ratio of robotic components,
+
+14:51.320 --> 14:53.640
+ it's coming down dramatically.
+
+14:53.640 --> 14:54.480
+ I'm interested.
+
+14:54.480 --> 14:56.040
+ Right, it continues to come down.
+
+14:56.040 --> 14:58.920
+ So I think we're asymptotically approaching the point
+
+14:58.920 --> 14:59.960
+ where we would get, yeah,
+
+14:59.960 --> 15:05.080
+ the cost of individuals would really become insignificant.
+
+15:05.080 --> 15:07.600
+ So let's step back at a high level view,
+
+15:07.600 --> 15:11.680
+ the impossible question of what kind of,
+
+15:11.680 --> 15:14.400
+ as an overview, what kind of autonomous flying vehicles
+
+15:14.400 --> 15:16.200
+ are there in general?
+
+15:16.200 --> 15:19.720
+ I think the ones that receive a lot of notoriety
+
+15:19.720 --> 15:22.560
+ are obviously the military vehicles.
+
+15:22.560 --> 15:26.280
+ Military vehicles are controlled by a base station,
+
+15:26.280 --> 15:29.640
+ but have a lot of human supervision,
+
+15:29.640 --> 15:31.800
+ but have limited autonomy,
+
+15:31.800 --> 15:34.760
+ which is the ability to go from point A to point B,
+
+15:34.760 --> 15:38.320
+ and even the more sophisticated vehicles
+
+15:38.320 --> 15:41.760
+ can do autonomous takeoff and landing.
+
+15:41.760 --> 15:44.400
+ And those usually have wings and they're heavy?
+
+15:44.400 --> 15:45.360
+ Usually they're wings,
+
+15:45.360 --> 15:47.440
+ but there's nothing preventing us from doing this
+
+15:47.440 --> 15:49.000
+ for helicopters as well.
+
+15:49.000 --> 15:53.440
+ There are many military organizations that have
+
+15:53.440 --> 15:56.560
+ autonomous helicopters in the same vein.
+
+15:56.560 --> 16:00.080
+ And by the way, you look at autopilots and airplanes,
+
+16:00.080 --> 16:02.800
+ and it's actually very similar.
+
+16:02.800 --> 16:07.160
+ In fact, one interesting question we can ask is,
+
+16:07.160 --> 16:12.120
+ if you look at all the air safety violations,
+
+16:12.120 --> 16:14.080
+ all the crashes that occurred,
+
+16:14.080 --> 16:16.880
+ would they have happened if the plane
+
+16:16.880 --> 16:20.200
+ were truly autonomous, and I think you'll find
+
+16:20.200 --> 16:21.960
+ that in many of the cases,
+
+16:21.960 --> 16:24.600
+ because of pilot error, we made silly decisions.
+
+16:24.600 --> 16:26.920
+ And so in some sense, even in air traffic,
+
+16:26.920 --> 16:29.760
+ commercial air traffic, there's a lot of applications,
+
+16:29.760 --> 16:33.920
+ although we only see autonomy being enabled
+
+16:33.920 --> 16:38.920
+ at very high altitudes when the plane is an autopilot.
+
+16:41.160 --> 16:42.520
+ There's still a role for the human,
+
+16:42.520 --> 16:47.520
+ and that kind of autonomy is, you're kind of implying,
+
+16:47.640 --> 16:48.680
+ I don't know what the right word is,
+
+16:48.680 --> 16:52.600
+ but it's a little dumber than it could be.
+
+16:53.480 --> 16:55.720
+ Right, so in the lab, of course,
+
+16:55.720 --> 16:59.200
+ we can afford to be a lot more aggressive.
+
+16:59.200 --> 17:04.200
+ And the question we try to ask is,
+
+17:04.600 --> 17:09.600
+ can we make robots that will be able to make decisions
+
+17:09.600 --> 17:13.680
+ without any kind of external infrastructure?
+
+17:13.680 --> 17:14.880
+ So what does that mean?
+
+17:14.880 --> 17:16.960
+ So the most common piece of infrastructure
+
+17:16.960 --> 17:19.640
+ that airplanes use today is GPS.
+
+17:20.560 --> 17:25.160
+ GPS is also the most brittle form of information.
+
+17:26.680 --> 17:30.480
+ If you've driven in a city, try to use GPS navigation,
+
+17:30.480 --> 17:32.760
+ tall buildings, you immediately lose GPS.
+
+17:33.720 --> 17:36.320
+ And so that's not a very sophisticated way
+
+17:36.320 --> 17:37.880
+ of building autonomy.
+
+17:37.880 --> 17:39.560
+ I think the second piece of infrastructure
+
+17:39.560 --> 17:41.920
+ that I rely on is communications.
+
+17:41.920 --> 17:46.200
+ Again, it's very easy to jam communications.
+
+17:47.400 --> 17:49.680
+ In fact, if you use Wi Fi,
+
+17:49.680 --> 17:51.880
+ you know that Wi Fi signals drop out,
+
+17:51.880 --> 17:53.560
+ cell signals drop out.
+
+17:53.560 --> 17:56.840
+ So to rely on something like that is not good.
+
+17:58.600 --> 18:01.240
+ The third form of infrastructure we use,
+
+18:01.240 --> 18:02.960
+ and I hate to call it infrastructure,
+
+18:02.960 --> 18:06.400
+ but it is that in the sense of robots, it's people.
+
+18:06.400 --> 18:08.760
+ So you could rely on somebody to pilot you.
+
+18:08.760 --> 18:09.960
+ Right.
+
+18:09.960 --> 18:11.600
+ And so the question you wanna ask is
+
+18:11.600 --> 18:13.400
+ if there are no pilots,
+
+18:13.400 --> 18:16.200
+ if there's no communications with any base station,
+
+18:16.200 --> 18:18.720
+ if there's no knowledge of position,
+
+18:18.720 --> 18:21.640
+ and if there's no a priori map,
+
+18:21.640 --> 18:24.880
+ a priori knowledge of what the environment looks like,
+
+18:24.880 --> 18:28.280
+ a priori model of what might happen in the future.
+
+18:28.280 --> 18:29.560
+ Can robots navigate?
+
+18:29.560 --> 18:31.440
+ So that is true autonomy.
+
+18:31.440 --> 18:33.240
+ So that's true autonomy.
+
+18:33.240 --> 18:35.040
+ And we're talking about, you mentioned
+
+18:35.040 --> 18:36.880
+ like military applications and drones.
+
+18:36.880 --> 18:38.280
+ Okay, so what else is there?
+
+18:38.280 --> 18:43.280
+ You talk about agile autonomous flying robots, aerial robots.
+
+18:43.480 --> 18:46.320
+ So that's a different kind of, it's not winged,
+
+18:46.320 --> 18:48.120
+ it's not big, at least it's small.
+
+18:48.120 --> 18:50.800
+ So I use the word agility mostly,
+
+18:50.800 --> 18:53.480
+ or at least we're motivated to do agile robots,
+
+18:53.480 --> 18:57.960
+ mostly because robots can operate
+
+18:57.960 --> 19:01.120
+ and should be operating in constrained environments.
+
+19:02.120 --> 19:06.960
+ And if you want to operate the way a global hawk operates,
+
+19:06.960 --> 19:09.120
+ I mean, the kinds of conditions in which you operate
+
+19:09.120 --> 19:10.760
+ are very, very restrictive.
+
+19:11.760 --> 19:13.720
+ If you wanna go inside a building,
+
+19:13.720 --> 19:15.600
+ for example, for search and rescue,
+
+19:15.600 --> 19:18.120
+ or to locate an active shooter,
+
+19:18.120 --> 19:22.120
+ or you wanna navigate under the canopy in an orchard
+
+19:22.120 --> 19:23.880
+ to look at health of plants,
+
+19:23.880 --> 19:28.880
+ or to count fruits to measure the tree trunks.
+
+19:31.240 --> 19:33.240
+ These are things we do, by the way.
+
+19:33.240 --> 19:35.920
+ Yeah, some cool agriculture stuff you've shown in the past,
+
+19:35.920 --> 19:36.760
+ it's really awesome.
+
+19:36.760 --> 19:40.400
+ So in those kinds of settings, you do need that agility.
+
+19:40.400 --> 19:42.560
+ Agility does not necessarily mean
+
+19:42.560 --> 19:45.440
+ you break records for the 100 meters dash.
+
+19:45.440 --> 19:48.040
+ What it really means is you see the unexpected
+
+19:48.040 --> 19:51.520
+ and you're able to maneuver in a safe way,
+
+19:51.520 --> 19:55.440
+ and in a way that gets you the most information
+
+19:55.440 --> 19:57.720
+ about the thing you're trying to do.
+
+19:57.720 --> 20:00.520
+ By the way, you may be the only person
+
+20:00.520 --> 20:04.280
+ who in a TED talk has used a math equation,
+
+20:04.280 --> 20:07.720
+ which is amazing, people should go see one of your TED talks.
+
+20:07.720 --> 20:08.840
+ Actually, it's very interesting
+
+20:08.840 --> 20:13.560
+ because the TED curator, Chris Anderson, told me,
+
+20:13.560 --> 20:15.400
+ you can't show math.
+
+20:15.400 --> 20:18.240
+ And I thought about it, but that's who I am.
+
+20:18.240 --> 20:20.800
+ I mean, that's our work.
+
+20:20.800 --> 20:25.800
+ And so I felt compelled to give the audience a taste
+
+20:25.800 --> 20:27.680
+ for at least some math.
+
+20:27.680 --> 20:32.680
+ So on that point, simply, what does it take
+
+20:32.680 --> 20:37.120
+ to make a thing with four motors fly, a quadcopter,
+
+20:37.120 --> 20:40.400
+ one of these little flying robots?
+
+20:41.560 --> 20:43.800
+ How hard is it to make it fly?
+
+20:43.800 --> 20:46.360
+ How do you coordinate the four motors?
+
+20:47.360 --> 20:52.360
+ How do you convert those motors into actual movement?
+
+20:52.400 --> 20:54.600
+ So this is an interesting question.
+
+20:54.600 --> 20:57.840
+ We've been trying to do this since 2000.
+
+20:57.840 --> 21:00.360
+ It is a commentary on the sensors
+
+21:00.360 --> 21:01.880
+ that were available back then,
+
+21:01.880 --> 21:04.320
+ and the computers that were available back then.
+
+21:05.640 --> 21:08.080
+ And a number of things happened
+
+21:08.080 --> 21:10.320
+ between 2000 and 2007.
+
+21:11.640 --> 21:15.560
+ One is the advances in computing, which is,
+
+21:15.560 --> 21:16.840
+ so we all know about Moore's Law,
+
+21:16.840 --> 21:19.760
+ but I think 2007 was a tipping point,
+
+21:19.760 --> 21:22.800
+ the year of the iPhone, the year of the cloud.
+
+21:22.800 --> 21:24.720
+ Lots of things happened in 2007.
+
+21:25.680 --> 21:27.640
+ But going back even further,
+
+21:27.640 --> 21:31.440
+ inertial measurement units as a sensor really matured.
+
+21:31.440 --> 21:33.040
+ Again, lots of reasons for that.
+
+21:34.000 --> 21:35.480
+ Certainly there's a lot of federal funding,
+
+21:35.480 --> 21:37.440
+ particularly DARPA in the US,
+
+21:38.360 --> 21:42.840
+ but they didn't anticipate this boom in IMUs.
+
+21:43.800 --> 21:46.080
+ But if you look subsequently,
+
+21:46.080 --> 21:49.000
+ what happened is that every car manufacturer
+
+21:49.000 --> 21:50.120
+ had to put an airbag in,
+
+21:50.120 --> 21:52.720
+ which meant you had to have an accelerometer on board.
+
+21:52.720 --> 21:54.080
+ And so that drove down the price
+
+21:54.080 --> 21:56.280
+ to performance ratio of the sensors.
+
+21:56.280 --> 21:57.960
+ I should know this, that's very interesting.
+
+21:57.960 --> 21:59.480
+ It's very interesting, the connection there.
+
+21:59.480 --> 22:03.160
+ And that's why research is very hard to predict the outcomes.
+
+22:04.920 --> 22:07.760
+ And again, the federal government spent a ton of money
+
+22:07.760 --> 22:12.360
+ on things that they thought were useful for resonators,
+
+22:12.360 --> 22:16.920
+ but it ended up enabling these small UAVs, which is great,
+
+22:16.920 --> 22:18.600
+ because I could have never raised that much money
+
+22:18.600 --> 22:20.800
+ and told, sold this project,
+
+22:20.800 --> 22:22.280
+ hey, we want to build these small UAVs.
+
+22:22.280 --> 22:25.520
+ Can you actually fund the development of low cost IMUs?
+
+22:25.520 --> 22:27.720
+ So why do you need an IMU on an IMU?
+
+22:27.720 --> 22:30.440
+ So I'll come back to that,
+
+22:30.440 --> 22:33.400
+ but so in 2007, 2008, we were able to build these.
+
+22:33.400 --> 22:35.280
+ And then the question you're asking was a good one,
+
+22:35.280 --> 22:40.280
+ how do you coordinate the motors to develop this?
+
+22:40.320 --> 22:43.920
+ But over the last 10 years, everything is commoditized.
+
+22:43.920 --> 22:47.920
+ A high school kid today can pick up a Raspberry Pi kit
+
+22:49.520 --> 22:50.600
+ and build this,
+
+22:50.600 --> 22:53.240
+ all the low levels functionality is all automated.
+
+22:53.240 --> 22:58.240
+ But basically at some level, you have to drive the motors
+
+22:59.160 --> 23:03.660
+ at the right RPMs, the right velocity,
+
+23:04.560 --> 23:07.480
+ in order to generate the right amount of thrust
+
+23:07.480 --> 23:09.960
+ in order to position it and orient it
+
+23:09.960 --> 23:12.840
+ in a way that you need to in order to fly.
+
+23:13.800 --> 23:16.680
+ The feedback that you get is from onboard sensors
+
+23:16.680 --> 23:18.400
+ and the IMU is an important part of it.
+
+23:18.400 --> 23:23.400
+ The IMU tells you what the acceleration is
+
+23:23.840 --> 23:26.400
+ as well as what the angular velocity is.
+
+23:26.400 --> 23:29.200
+ And those are important pieces of information.
+
+23:30.440 --> 23:34.200
+ In addition to that, you need some kind of local position
+
+23:34.200 --> 23:36.480
+ or velocity information.
+
+23:37.440 --> 23:39.320
+ For example, when we walk,
+
+23:39.320 --> 23:41.520
+ we implicitly have this information
+
+23:41.520 --> 23:45.800
+ because we kind of know how, what our stride length is.
+
+23:45.800 --> 23:50.800
+ We also are looking at images fly past our retina,
+
+23:51.440 --> 23:54.240
+ if you will, and so we can estimate velocity.
+
+23:54.240 --> 23:56.280
+ We also have accelerometers in our head
+
+23:56.280 --> 23:59.120
+ and we're able to integrate all these pieces of information
+
+23:59.120 --> 24:02.320
+ to determine where we are as we walk.
+
+24:02.320 --> 24:04.320
+ And so robots have to do something very similar.
+
+24:04.320 --> 24:08.160
+ You need an IMU, you need some kind of a camera
+
+24:08.160 --> 24:11.600
+ or other sensor that's measuring velocity.
+
+24:11.600 --> 24:15.800
+ And then you need some kind of a global reference frame
+
+24:15.800 --> 24:19.480
+ if you really want to think about doing something
+
+24:19.480 --> 24:21.280
+ in a world coordinate system.
+
+24:21.280 --> 24:23.680
+ And so how do you estimate your position
+
+24:23.680 --> 24:25.160
+ with respect to that global reference frame?
+
+24:25.160 --> 24:26.560
+ That's important as well.
+
+24:26.560 --> 24:29.520
+ So coordinating the RPMs of the four motors
+
+24:29.520 --> 24:32.680
+ is what allows you to, first of all, fly and hover
+
+24:32.680 --> 24:35.640
+ and then you can change the orientation
+
+24:35.640 --> 24:37.640
+ and the velocity and so on.
+
+24:37.640 --> 24:38.480
+ Exactly, exactly.
+
+24:38.480 --> 24:40.360
+ So there's a bunch of degrees of freedom
+
+24:40.360 --> 24:42.240
+ or there's six degrees of freedom
+
+24:42.240 --> 24:44.960
+ but you only have four inputs, the four motors.
+
+24:44.960 --> 24:49.960
+ And it turns out to be a remarkably versatile configuration.
+
+24:50.960 --> 24:53.120
+ You think at first, well, I only have four motors,
+
+24:53.120 --> 24:55.040
+ how do I go sideways?
+
+24:55.040 --> 24:56.360
+ But it's not too hard to say, well,
+
+24:56.360 --> 24:59.200
+ if I tilt myself, I can go sideways.
+
+24:59.200 --> 25:01.200
+ And then you have four motors pointing up,
+
+25:01.200 --> 25:05.400
+ how do I rotate in place about a vertical axis?
+
+25:05.400 --> 25:07.840
+ Well, you rotate them at different speeds
+
+25:07.840 --> 25:09.760
+ and that generates reaction moments
+
+25:09.760 --> 25:11.560
+ and that allows you to turn.
+
+25:11.560 --> 25:13.400
+ So it's actually a pretty,
+
+25:13.400 --> 25:17.040
+ it's an optimal configuration from an engineer standpoint.
+
+25:17.960 --> 25:22.960
+ It's very simple, very cleverly done and very versatile.
+
+25:23.800 --> 25:26.520
+ So if you could step back to a time,
+
+25:27.320 --> 25:30.120
+ so I've always known flying robots as,
+
+25:31.120 --> 25:35.840
+ to me it was natural that the quadcopters should fly.
+
+25:35.840 --> 25:38.040
+ But when you first started working with it,
+
+25:38.040 --> 25:42.040
+ how surprised are you that you can make,
+
+25:42.040 --> 25:45.560
+ do so much with the four motors?
+
+25:45.560 --> 25:47.640
+ How surprising is that you can make this thing fly,
+
+25:47.640 --> 25:49.800
+ first of all, that you can make it hover,
+
+25:49.800 --> 25:52.040
+ then you can add control to it?
+
+25:52.920 --> 25:55.800
+ Firstly, this is not, the four motor configuration
+
+25:55.800 --> 26:00.120
+ is not ours, it has at least a hundred year history.
+
+26:01.080 --> 26:02.480
+ And various people,
+
+26:02.480 --> 26:06.280
+ various people try to get quadrotors to fly
+
+26:06.280 --> 26:08.120
+ without much success.
+
+26:09.240 --> 26:11.560
+ As I said, we've been working on this since 2000.
+
+26:11.560 --> 26:13.360
+ Our first designs were,
+
+26:13.360 --> 26:15.160
+ well, this is way too complicated.
+
+26:15.160 --> 26:19.160
+ Why not we try to get an omnidirectional flying robot?
+
+26:19.160 --> 26:22.760
+ So our early designs, we had eight rotors.
+
+26:22.760 --> 26:26.080
+ And so these eight rotors were arranged uniformly
+
+26:27.520 --> 26:28.880
+ on a sphere, if you will.
+
+26:28.880 --> 26:31.360
+ So you can imagine a symmetric configuration
+
+26:31.360 --> 26:34.160
+ and so you should be able to fly anywhere.
+
+26:34.160 --> 26:36.280
+ But the real challenge we had is the strength
+
+26:36.280 --> 26:37.880
+ to weight ratio is not enough,
+
+26:37.880 --> 26:41.240
+ and of course we didn't have the sensors and so on.
+
+26:41.240 --> 26:43.840
+ So everybody knew, or at least the people
+
+26:43.840 --> 26:45.680
+ who worked with rotor crafts knew,
+
+26:45.680 --> 26:47.320
+ four rotors would get it done.
+
+26:48.280 --> 26:50.200
+ So that was not our idea.
+
+26:50.200 --> 26:53.480
+ But it took a while before we could actually do
+
+26:53.480 --> 26:56.520
+ the onboard sensing and the computation
+
+26:56.520 --> 27:00.400
+ that was needed for the kinds of agile maneuvering
+
+27:00.400 --> 27:03.800
+ that we wanted to do in our little aerial robots.
+
+27:03.800 --> 27:08.320
+ And that only happened between 2007 and 2009 in our lab.
+
+27:08.320 --> 27:10.680
+ Yeah, and you have to send the signal
+
+27:10.680 --> 27:13.200
+ maybe a hundred times a second.
+
+27:13.200 --> 27:15.400
+ So the compute there is everything
+
+27:15.400 --> 27:16.720
+ has to come down in price.
+
+27:16.720 --> 27:21.720
+ And what are the steps of getting from point A to point B?
+
+27:22.320 --> 27:25.840
+ So we just talked about like local control,
+
+27:25.840 --> 27:30.840
+ but if all the kind of cool dancing in the air
+
+27:30.840 --> 27:34.480
+ that I've seen you show, how do you make it happen?
+
+27:34.480 --> 27:38.840
+ Make it trajectory, first of all, okay,
+
+27:38.840 --> 27:41.600
+ figure out a trajectory, so plan a trajectory,
+
+27:41.600 --> 27:44.320
+ and then how do you make that trajectory happen?
+
+27:44.320 --> 27:47.320
+ I think planning is a very fundamental problem in robotics.
+
+27:47.320 --> 27:50.120
+ I think 10 years ago it was an esoteric thing,
+
+27:50.120 --> 27:52.360
+ but today with self driving cars,
+
+27:52.360 --> 27:55.160
+ everybody can understand this basic idea
+
+27:55.160 --> 27:57.320
+ that a car sees a whole bunch of things
+
+27:57.320 --> 27:59.720
+ and it has to keep a lane or maybe make a right turn
+
+27:59.720 --> 28:02.160
+ or switch lanes, it has to plan a trajectory,
+
+28:02.160 --> 28:04.320
+ it has to be safe, it has to be efficient.
+
+28:04.320 --> 28:06.120
+ So everybody's familiar with that.
+
+28:06.120 --> 28:07.400
+ That's kind of the first step
+
+28:07.400 --> 28:12.400
+ that you have to think about when you say autonomy.
+
+28:14.320 --> 28:18.600
+ And so for us, it's about finding smooth motions,
+
+28:18.600 --> 28:20.760
+ motions that are safe.
+
+28:20.760 --> 28:22.360
+ So we think about these two things.
+
+28:22.360 --> 28:24.160
+ One is optimality, one is safety.
+
+28:24.160 --> 28:26.720
+ Clearly you cannot compromise safety.
+
+28:26.720 --> 28:30.160
+ So you're looking for safe, optimal motions.
+
+28:30.160 --> 28:33.160
+ The other thing you have to think about
+
+28:33.160 --> 28:37.360
+ is can you actually compute a reasonable trajectory
+
+28:37.360 --> 28:41.560
+ in a small amount of time, because you have a time budget.
+
+28:41.560 --> 28:44.360
+ So the optimal becomes suboptimal,
+
+28:44.360 --> 28:49.360
+ but in our lab we focus on synthesizing smooth trajectory
+
+28:50.560 --> 28:52.360
+ that satisfy all the constraints.
+
+28:52.360 --> 28:57.200
+ In other words, don't violate any safety constraints
+
+28:57.200 --> 29:02.200
+ and is as efficient as possible.
+
+29:02.200 --> 29:04.600
+ And when I say efficient, it could mean
+
+29:04.600 --> 29:07.760
+ I want to get from point A to point B as quickly as possible
+
+29:07.760 --> 29:11.200
+ or I want to get to it as gracefully as possible
+
+29:11.200 --> 29:15.360
+ or I want to consume as little energy as possible.
+
+29:15.360 --> 29:17.600
+ But always staying within the safety constraints.
+
+29:17.600 --> 29:22.440
+ But yes, always finding a safe trajectory.
+
+29:22.440 --> 29:24.440
+ So there's a lot of excitement and progress
+
+29:24.440 --> 29:26.440
+ in the field of machine learning.
+
+29:26.440 --> 29:31.440
+ And reinforcement learning and the neural network variant
+
+29:31.440 --> 29:33.440
+ of that with deeper reinforcement learning.
+
+29:33.440 --> 29:37.440
+ Do you see a role of machine learning in...
+
+29:37.440 --> 29:40.040
+ So a lot of the success with flying robots
+
+29:40.040 --> 29:41.840
+ did not rely on machine learning,
+
+29:41.840 --> 29:44.440
+ except for maybe a little bit of the perception
+
+29:44.440 --> 29:46.040
+ on the computer vision side.
+
+29:46.040 --> 29:48.040
+ On the control side and the planning,
+
+29:48.040 --> 29:50.040
+ do you see there's a role in the future
+
+29:50.040 --> 29:51.040
+ for machine learning?
+
+29:51.040 --> 29:53.040
+ So let me disagree a little bit with you.
+
+29:53.040 --> 29:56.040
+ I think we never perhaps called out in my work
+
+29:56.040 --> 29:59.040
+ called out learning, but even this very simple idea
+
+29:59.040 --> 30:04.040
+ of being able to fly through a constrained space.
+
+30:04.040 --> 30:07.040
+ The first time you try it, you'll invariably...
+
+30:07.040 --> 30:10.040
+ You might get it wrong if the task is challenging.
+
+30:10.040 --> 30:14.040
+ And the reason is to get it perfectly right,
+
+30:14.040 --> 30:17.040
+ you have to model everything in the environment.
+
+30:17.040 --> 30:22.040
+ And flying is notoriously hard to model.
+
+30:22.040 --> 30:28.040
+ There are aerodynamic effects that we constantly discover,
+
+30:28.040 --> 30:31.040
+ even just before I was talking to you,
+
+30:31.040 --> 30:37.040
+ I was talking to a student about how blades flap when they fly.
+
+30:37.040 --> 30:43.040
+ And that ends up changing how a rotorcraft
+
+30:43.040 --> 30:46.040
+ is accelerated in the angular direction.
+
+30:46.040 --> 30:48.040
+ Does it use like microflaps or something?
+
+30:48.040 --> 30:49.040
+ It's not microflaps.
+
+30:49.040 --> 30:52.040
+ We assume that each blade is rigid,
+
+30:52.040 --> 30:54.040
+ but actually it flaps a little bit.
+
+30:54.040 --> 30:55.040
+ It bends.
+
+30:55.040 --> 30:56.040
+ Interesting, yeah.
+
+30:56.040 --> 30:58.040
+ And so the models rely on the fact,
+
+30:58.040 --> 31:01.040
+ on the assumption that they're actually rigid.
+
+31:01.040 --> 31:02.040
+ But that's not true.
+
+31:02.040 --> 31:04.040
+ If you're flying really quickly,
+
+31:04.040 --> 31:07.040
+ these effects become significant.
+
+31:07.040 --> 31:09.040
+ If you're flying close to the ground,
+
+31:09.040 --> 31:12.040
+ you get pushed off by the ground.
+
+31:12.040 --> 31:15.040
+ Something which every pilot knows when he tries to land
+
+31:15.040 --> 31:19.040
+ or she tries to land, this is called a ground effect.
+
+31:19.040 --> 31:21.040
+ Something very few pilots think about
+
+31:21.040 --> 31:23.040
+ is what happens when you go close to a ceiling,
+
+31:23.040 --> 31:25.040
+ or you get sucked into a ceiling.
+
+31:25.040 --> 31:29.040
+ There are very few aircraft that fly close to any kind of ceiling.
+
+31:29.040 --> 31:33.040
+ Likewise, when you go close to a wall,
+
+31:33.040 --> 31:36.040
+ there are these wall effects.
+
+31:36.040 --> 31:39.040
+ And if you've gone on a train and you pass another train
+
+31:39.040 --> 31:41.040
+ that's traveling the opposite direction,
+
+31:41.040 --> 31:43.040
+ you can feel the buffeting.
+
+31:43.040 --> 31:46.040
+ And so these kinds of microclimates
+
+31:46.040 --> 31:48.040
+ affect our UAVs significantly.
+
+31:48.040 --> 31:51.040
+ And they're impossible to model, essentially.
+
+31:51.040 --> 31:53.040
+ I wouldn't say they're impossible to model,
+
+31:53.040 --> 31:55.040
+ but the level of sophistication you would need
+
+31:55.040 --> 32:00.040
+ in the model and the software would be tremendous.
+
+32:00.040 --> 32:03.040
+ Plus, to get everything right would be awfully tedious.
+
+32:03.040 --> 32:05.040
+ So the way we do this is over time,
+
+32:05.040 --> 32:10.040
+ we figure out how to adapt to these conditions.
+
+32:10.040 --> 32:13.040
+ So early on, we use the form of learning
+
+32:13.040 --> 32:15.040
+ that we call iterative learning.
+
+32:15.040 --> 32:18.040
+ So this idea, if you want to perform a task,
+
+32:18.040 --> 32:23.040
+ there are a few things that you need to change
+
+32:23.040 --> 32:25.040
+ and iterate over a few parameters
+
+32:25.040 --> 32:29.040
+ that over time you can figure out.
+
+32:29.040 --> 32:34.040
+ So I could call it policy gradient reinforcement learning,
+
+32:34.040 --> 32:36.040
+ but actually it was just iterative learning.
+
+32:36.040 --> 32:38.040
+ And so this was there way back.
+
+32:38.040 --> 32:40.040
+ I think what's interesting is,
+
+32:40.040 --> 32:43.040
+ if you look at autonomous vehicles today,
+
+32:43.040 --> 32:46.040
+ learning occurs, could occur in two pieces.
+
+32:46.040 --> 32:48.040
+ One is perception, understanding the world.
+
+32:48.040 --> 32:51.040
+ Second is action, taking actions.
+
+32:51.040 --> 32:54.040
+ Everything that I've seen that is successful
+
+32:54.040 --> 32:56.040
+ is on the perception side of things.
+
+32:56.040 --> 32:59.040
+ So in computer vision, we've made amazing strides
+
+32:59.040 --> 33:00.040
+ in the last 10 years.
+
+33:00.040 --> 33:03.040
+ So recognizing objects, actually detecting objects,
+
+33:03.040 --> 33:08.040
+ classifying them and tagging them in some sense,
+
+33:08.040 --> 33:11.040
+ annotating them, this is all done through machine learning.
+
+33:11.040 --> 33:13.040
+ On the action side, on the other hand,
+
+33:13.040 --> 33:17.040
+ I don't know if any examples where there are fielded systems
+
+33:17.040 --> 33:21.040
+ where we actually learn the right behavior.
+
+33:21.040 --> 33:23.040
+ Outside of single demonstration is successful.
+
+33:23.040 --> 33:25.040
+ On the laboratory, this is the Holy Grail.
+
+33:25.040 --> 33:27.040
+ Can you do end to end learning?
+
+33:27.040 --> 33:31.040
+ Can you go from pixels to motor currents?
+
+33:31.040 --> 33:33.040
+ This is really, really hard.
+
+33:33.040 --> 33:36.040
+ And I think if you go forward,
+
+33:36.040 --> 33:38.040
+ the right way to think about these things
+
+33:38.040 --> 33:43.040
+ is data driven approaches, learning based approaches,
+
+33:43.040 --> 33:46.040
+ in concert with model based approaches,
+
+33:46.040 --> 33:48.040
+ which is the traditional way of doing things.
+
+33:48.040 --> 33:50.040
+ So I think there's a piece,
+
+33:50.040 --> 33:52.040
+ there's a role for each of these methodologies.
+
+33:52.040 --> 33:55.040
+ So what do you think, just jumping out on topics,
+
+33:55.040 --> 33:57.040
+ since you mentioned autonomous vehicles,
+
+33:57.040 --> 33:59.040
+ what do you think are the limits on the perception side?
+
+33:59.040 --> 34:02.040
+ So I've talked to Elon Musk,
+
+34:02.040 --> 34:04.040
+ and there on the perception side,
+
+34:04.040 --> 34:07.040
+ they're using primarily computer vision
+
+34:07.040 --> 34:09.040
+ to perceive the environment.
+
+34:09.040 --> 34:13.040
+ In your work with, because you work with the real world a lot,
+
+34:13.040 --> 34:15.040
+ and the physical world,
+
+34:15.040 --> 34:17.040
+ what are the limits of computer vision?
+
+34:17.040 --> 34:20.040
+ Do you think you can solve autonomous vehicles,
+
+34:20.040 --> 34:22.040
+ focusing on the perception side,
+
+34:22.040 --> 34:25.040
+ focusing on vision alone and machine learning?
+
+34:25.040 --> 34:29.040
+ So we also have a spin off company, Exxon Technologies,
+
+34:29.040 --> 34:32.040
+ that works underground in mines.
+
+34:32.040 --> 34:35.040
+ So you go into mines, they're dark.
+
+34:35.040 --> 34:37.040
+ They're dirty.
+
+34:37.040 --> 34:39.040
+ You fly in a dirty area,
+
+34:39.040 --> 34:42.040
+ there's stuff you kick up by the propellers,
+
+34:42.040 --> 34:44.040
+ the downwash kicks up dust.
+
+34:44.040 --> 34:48.040
+ I challenge you to get a computer vision algorithm to work there.
+
+34:48.040 --> 34:53.040
+ So we use LIDARS in that setting.
+
+34:53.040 --> 34:57.040
+ Indoors, and even outdoors when we fly through fields,
+
+34:57.040 --> 34:59.040
+ I think there's a lot of potential
+
+34:59.040 --> 35:03.040
+ for just solving the problem using computer vision alone.
+
+35:03.040 --> 35:05.040
+ But I think the bigger question is,
+
+35:05.040 --> 35:08.040
+ can you actually solve,
+
+35:08.040 --> 35:11.040
+ or can you actually identify all the corner cases
+
+35:11.040 --> 35:14.040
+ using a single sensing modality
+
+35:14.040 --> 35:16.040
+ and using learning alone?
+
+35:16.040 --> 35:18.040
+ What's your intuition there?
+
+35:18.040 --> 35:20.040
+ So look, if you have a corner case
+
+35:20.040 --> 35:22.040
+ and your algorithm doesn't work,
+
+35:22.040 --> 35:25.040
+ your instinct is to go get data about the corner case
+
+35:25.040 --> 35:29.040
+ and patch it up, learn how to deal with that corner case.
+
+35:29.040 --> 35:32.040
+ But at some point,
+
+35:32.040 --> 35:36.040
+ this is going to saturate, this approach is not viable.
+
+35:36.040 --> 35:39.040
+ So today, computer vision algorithms
+
+35:39.040 --> 35:43.040
+ can detect objects 90% of the time,
+
+35:43.040 --> 35:45.040
+ classify them 90% of the time.
+
+35:45.040 --> 35:49.040
+ Cats on the internet probably can do 95%, I don't know.
+
+35:49.040 --> 35:54.040
+ But to get from 90% to 99%, you need a lot more data.
+
+35:54.040 --> 35:56.040
+ And then I tell you, well, that's not enough
+
+35:56.040 --> 35:58.040
+ because I have a safety critical application
+
+35:58.040 --> 36:01.040
+ that want to go from 99% to 99.9%,
+
+36:01.040 --> 36:03.040
+ well, that's even more data.
+
+36:03.040 --> 36:07.040
+ So I think if you look at
+
+36:07.040 --> 36:11.040
+ wanting accuracy on the x axis
+
+36:11.040 --> 36:15.040
+ and look at the amount of data on the y axis,
+
+36:15.040 --> 36:18.040
+ I believe that curve is an exponential curve.
+
+36:18.040 --> 36:21.040
+ Wow, okay, it's even hard if it's linear.
+
+36:21.040 --> 36:24.040
+ It's hard if it's linear, totally, but I think it's exponential.
+
+36:24.040 --> 36:26.040
+ And the other thing you have to think about
+
+36:26.040 --> 36:31.040
+ is that this process is a very, very power hungry process
+
+36:31.040 --> 36:34.040
+ to run data farms or servers.
+
+36:34.040 --> 36:36.040
+ Power, do you mean literally power?
+
+36:36.040 --> 36:38.040
+ Literally power, literally power.
+
+36:38.040 --> 36:43.040
+ So in 2014, five years ago, and I don't have more recent data,
+
+36:43.040 --> 36:50.040
+ 2% of US electricity consumption was from data farms.
+
+36:50.040 --> 36:54.040
+ So we think about this as an information science
+
+36:54.040 --> 36:56.040
+ and information processing problem.
+
+36:56.040 --> 36:59.040
+ Actually, it is an energy processing problem.
+
+36:59.040 --> 37:02.040
+ And so unless we've figured out better ways of doing this,
+
+37:02.040 --> 37:04.040
+ I don't think this is viable.
+
+37:04.040 --> 37:08.040
+ So talking about driving, which is a safety critical application
+
+37:08.040 --> 37:11.040
+ and some aspect of the flight is safety critical,
+
+37:11.040 --> 37:14.040
+ maybe philosophical question, maybe an engineering one.
+
+37:14.040 --> 37:16.040
+ What problem do you think is harder to solve?
+
+37:16.040 --> 37:19.040
+ Autonomous driving or autonomous flight?
+
+37:19.040 --> 37:21.040
+ That's a really interesting question.
+
+37:21.040 --> 37:26.040
+ I think autonomous flight has several advantages
+
+37:26.040 --> 37:30.040
+ that autonomous driving doesn't have.
+
+37:30.040 --> 37:33.040
+ So look, if I want to go from point A to point B,
+
+37:33.040 --> 37:35.040
+ I have a very, very safe trajectory.
+
+37:35.040 --> 37:38.040
+ Go vertically up to a maximum altitude,
+
+37:38.040 --> 37:41.040
+ fly horizontally to just about the destination
+
+37:41.040 --> 37:43.040
+ and then come down vertically.
+
+37:43.040 --> 37:46.040
+ This is preprogrammed.
+
+37:46.040 --> 37:49.040
+ The equivalent of that is very hard to find
+
+37:49.040 --> 37:53.040
+ in a self driving car world because you're on the ground,
+
+37:53.040 --> 37:55.040
+ you're in a two dimensional surface,
+
+37:55.040 --> 37:58.040
+ and the trajectories on the two dimensional surface
+
+37:58.040 --> 38:01.040
+ are more likely to encounter obstacles.
+
+38:01.040 --> 38:03.040
+ I mean this in an intuitive sense,
+
+38:03.040 --> 38:05.040
+ but mathematically true, that's...
+
+38:05.040 --> 38:08.040
+ Mathematically as well, that's true.
+
+38:08.040 --> 38:11.040
+ There's other option on the 2G space of platooning
+
+38:11.040 --> 38:13.040
+ or because there's so many obstacles,
+
+38:13.040 --> 38:15.040
+ you can connect with those obstacles
+
+38:15.040 --> 38:16.040
+ and all these kinds of problems.
+
+38:16.040 --> 38:18.040
+ But those exist in the three dimensional space as well.
+
+38:18.040 --> 38:19.040
+ So they do.
+
+38:19.040 --> 38:23.040
+ So the question also implies how difficult are obstacles
+
+38:23.040 --> 38:25.040
+ in the three dimensional space in flight?
+
+38:25.040 --> 38:27.040
+ So that's the downside.
+
+38:27.040 --> 38:29.040
+ I think in three dimensional space,
+
+38:29.040 --> 38:31.040
+ you're modeling three dimensional world,
+
+38:31.040 --> 38:33.040
+ not just because you want to avoid it,
+
+38:33.040 --> 38:35.040
+ but you want to reason about it
+
+38:35.040 --> 38:37.040
+ and you want to work in that three dimensional environment.
+
+38:37.040 --> 38:39.040
+ And that's significantly harder.
+
+38:39.040 --> 38:41.040
+ So that's one disadvantage.
+
+38:41.040 --> 38:43.040
+ I think the second disadvantage is of course,
+
+38:43.040 --> 38:45.040
+ anytime you fly, you have to put up
+
+38:45.040 --> 38:49.040
+ with the peculiarities of aerodynamics
+
+38:49.040 --> 38:51.040
+ and their complicated environments.
+
+38:51.040 --> 38:52.040
+ How do you negotiate that?
+
+38:52.040 --> 38:54.040
+ So that's always a problem.
+
+38:54.040 --> 38:57.040
+ Do you see a time in the future where there is...
+
+38:57.040 --> 39:00.040
+ You mentioned there's agriculture applications.
+
+39:00.040 --> 39:03.040
+ So there's a lot of applications of flying robots.
+
+39:03.040 --> 39:07.040
+ But do you see a time in the future where there is tens of thousands
+
+39:07.040 --> 39:10.040
+ or maybe hundreds of thousands of delivery drones
+
+39:10.040 --> 39:14.040
+ that fill the sky, a delivery of flying robots?
+
+39:14.040 --> 39:18.040
+ I think there's a lot of potential for the last mile delivery.
+
+39:18.040 --> 39:21.040
+ And so in crowded cities,
+
+39:21.040 --> 39:24.040
+ I don't know if you go to a place like Hong Kong,
+
+39:24.040 --> 39:27.040
+ just crossing the river can take half an hour.
+
+39:27.040 --> 39:32.040
+ And while a drone can just do it in five minutes at most.
+
+39:32.040 --> 39:38.040
+ I think you look at delivery of supplies to remote villages.
+
+39:38.040 --> 39:41.040
+ I work with a nonprofit called Weave Robotics.
+
+39:41.040 --> 39:43.040
+ So they work in the Peruvian Amazon,
+
+39:43.040 --> 39:47.040
+ where the only highways are rivers.
+
+39:47.040 --> 39:49.040
+ And to get from point A to point B
+
+39:49.040 --> 39:52.040
+ may take five hours.
+
+39:52.040 --> 39:56.040
+ While with a drone, you can get there in 30 minutes.
+
+39:56.040 --> 39:59.040
+ So just delivering drugs,
+
+39:59.040 --> 40:04.040
+ retrieving samples for testing vaccines.
+
+40:04.040 --> 40:06.040
+ I think there's huge potential here.
+
+40:06.040 --> 40:09.040
+ So I think the challenges are not technological.
+
+40:09.040 --> 40:12.040
+ The challenge is economical.
+
+40:12.040 --> 40:16.040
+ The one thing I'll tell you that nobody thinks about
+
+40:16.040 --> 40:21.040
+ is the fact that we've not made huge strides in battery technology.
+
+40:21.040 --> 40:22.040
+ Yes, it's true.
+
+40:22.040 --> 40:24.040
+ Batteries are becoming less expensive
+
+40:24.040 --> 40:27.040
+ because we have these mega factories that are coming up.
+
+40:27.040 --> 40:29.040
+ But they're all based on lithium based technologies.
+
+40:29.040 --> 40:34.040
+ And if you look at the energy density and the power density,
+
+40:34.040 --> 40:39.040
+ those are two fundamentally limiting numbers.
+
+40:39.040 --> 40:41.040
+ So power density is important because for a UAV
+
+40:41.040 --> 40:43.040
+ to take off vertically into the air,
+
+40:43.040 --> 40:47.040
+ which most drones do, they don't have a runway,
+
+40:47.040 --> 40:52.040
+ you consume roughly 200 watts per kilo at the small size.
+
+40:52.040 --> 40:54.040
+ That's a lot.
+
+40:54.040 --> 40:58.040
+ In contrast, the human brain consumes less than 80 watts,
+
+40:58.040 --> 41:00.040
+ the whole of the human brain.
+
+41:00.040 --> 41:04.040
+ So just imagine just lifting yourself into the air
+
+41:04.040 --> 41:08.040
+ is like two or three light bulbs, which makes no sense to me.
+
+41:08.040 --> 41:12.040
+ Yeah, so you're going to have to at scale solve the energy problem
+
+41:12.040 --> 41:19.040
+ then charging the batteries, storing the energy and so on.
+
+41:19.040 --> 41:21.040
+ And then the storage is the second problem.
+
+41:21.040 --> 41:23.040
+ But storage limits the range.
+
+41:23.040 --> 41:30.040
+ But you have to remember that you have to burn a lot of it
+
+41:30.040 --> 41:32.040
+ for a given time.
+
+41:32.040 --> 41:33.040
+ So the burning is another problem.
+
+41:33.040 --> 41:35.040
+ Which is a power question.
+
+41:35.040 --> 41:36.040
+ Yes.
+
+41:36.040 --> 41:39.040
+ And do you think just your intuition,
+
+41:39.040 --> 41:45.040
+ there are breakthroughs in batteries on the horizon?
+
+41:45.040 --> 41:47.040
+ How hard is that problem?
+
+41:47.040 --> 41:52.040
+ Look, there are a lot of companies that are promising flying cars,
+
+41:52.040 --> 42:00.040
+ that are autonomous, and that are clean.
+
+42:00.040 --> 42:02.040
+ I think they're over promising.
+
+42:02.040 --> 42:05.040
+ The autonomy piece is doable.
+
+42:05.040 --> 42:08.040
+ The clean piece, I don't think so.
+
+42:08.040 --> 42:12.040
+ There's another company that I work with called Jatatra.
+
+42:12.040 --> 42:16.040
+ They make small jet engines.
+
+42:16.040 --> 42:20.040
+ And they can get up to 50 miles an hour very easily and lift 50 kilos.
+
+42:20.040 --> 42:22.040
+ But they're jet engines.
+
+42:22.040 --> 42:24.040
+ They're efficient.
+
+42:24.040 --> 42:26.040
+ They're a little louder than electric vehicles.
+
+42:26.040 --> 42:29.040
+ But they can build flying cars.
+
+42:29.040 --> 42:33.040
+ So your sense is that there's a lot of pieces that have come together.
+
+42:33.040 --> 42:39.040
+ So on this crazy question, if you look at companies like Kitty Hawk,
+
+42:39.040 --> 42:45.040
+ working on electric, so the clean, talking as the bashing through.
+
+42:45.040 --> 42:52.040
+ It's a crazy dream, but you work with flight a lot.
+
+42:52.040 --> 42:58.040
+ You've mentioned before that manned flights or carrying a human body
+
+42:58.040 --> 43:01.040
+ is very difficult to do.
+
+43:01.040 --> 43:04.040
+ So how crazy is flying cars?
+
+43:04.040 --> 43:11.040
+ Do you think there will be a day when we have vertical takeoff and landing vehicles
+
+43:11.040 --> 43:17.040
+ that are sufficiently affordable that we're going to see a huge amount of them?
+
+43:17.040 --> 43:21.040
+ And they would look like something like we dream of when we think about flying cars.
+
+43:21.040 --> 43:23.040
+ Yeah, like the Jetsons.
+
+43:23.040 --> 43:26.040
+ So look, there are a lot of smart people working on this.
+
+43:26.040 --> 43:32.040
+ And you never say something is not possible when you're people like Sebastian Thrun working on it.
+
+43:32.040 --> 43:35.040
+ So I totally think it's viable.
+
+43:35.040 --> 43:38.040
+ I question, again, the electric piece.
+
+43:38.040 --> 43:40.040
+ The electric piece, yeah.
+
+43:40.040 --> 43:42.040
+ For short distances, you can do it.
+
+43:42.040 --> 43:46.040
+ And there's no reason to suggest that these all just have to be rotor crafts.
+
+43:46.040 --> 43:50.040
+ You take off vertically, but then you morph into a forward flight.
+
+43:50.040 --> 43:52.040
+ I think there are a lot of interesting designs.
+
+43:52.040 --> 43:56.040
+ The question to me is, are these economically viable?
+
+43:56.040 --> 44:02.040
+ And if you agree to do this with fossil fuels, it instantly immediately becomes viable.
+
+44:02.040 --> 44:04.040
+ That's a real challenge.
+
+44:04.040 --> 44:09.040
+ Do you think it's possible for robots and humans to collaborate successfully on tasks?
+
+44:09.040 --> 44:18.040
+ So a lot of robotics folks that I talk to and work with, I mean, humans just add a giant mess to the picture.
+
+44:18.040 --> 44:22.040
+ So it's best to remove them from consideration when solving specific tasks.
+
+44:22.040 --> 44:24.040
+ It's very difficult to model.
+
+44:24.040 --> 44:26.040
+ There's just a source of uncertainty.
+
+44:26.040 --> 44:36.040
+ In your work with these agile flying robots, do you think there's a role for collaboration with humans?
+
+44:36.040 --> 44:43.040
+ Is it best to model tasks in a way that doesn't have a human in the picture?
+
+44:43.040 --> 44:48.040
+ I don't think we should ever think about robots without human in the picture.
+
+44:48.040 --> 44:54.040
+ Ultimately, robots are there because we want them to solve problems for humans.
+
+44:54.040 --> 44:58.040
+ But there's no general solution to this problem.
+
+44:58.040 --> 45:02.040
+ I think if you look at human interaction and how humans interact with robots,
+
+45:02.040 --> 45:06.040
+ you know, we think of these in sort of three different ways.
+
+45:06.040 --> 45:09.040
+ One is the human commanding the robot.
+
+45:09.040 --> 45:13.040
+ The second is the human collaborating with the robot.
+
+45:13.040 --> 45:19.040
+ So for example, we work on how a robot can actually pick up things with a human and carry things.
+
+45:19.040 --> 45:21.040
+ That's like true collaboration.
+
+45:21.040 --> 45:26.040
+ And third, we think about humans as bystanders, self driving cars.
+
+45:26.040 --> 45:33.040
+ What's the human's role and how do self driving cars acknowledge the presence of humans?
+
+45:33.040 --> 45:36.040
+ So I think all of these things are different scenarios.
+
+45:36.040 --> 45:39.040
+ It depends on what kind of humans, what kind of tasks.
+
+45:39.040 --> 45:45.040
+ And I think it's very difficult to say that there's a general theory that we all have for this.
+
+45:45.040 --> 45:52.040
+ But at the same time, it's also silly to say that we should think about robots independent of humans.
+
+45:52.040 --> 45:59.040
+ So to me, human robot interaction is almost a mandatory aspect of everything we do.
+
+45:59.040 --> 46:00.040
+ Yes.
+
+46:00.040 --> 46:05.040
+ But to wish to agree, so your thoughts, if we jump to autonomous vehicles, for example,
+
+46:05.040 --> 46:10.040
+ there's a big debate between what's called level two and level four.
+
+46:10.040 --> 46:13.040
+ So semi autonomous and autonomous vehicles.
+
+46:13.040 --> 46:19.040
+ And sort of the Tesla approach currently at least has a lot of collaboration between human and machine.
+
+46:19.040 --> 46:24.040
+ So the human is supposed to actively supervise the operation of the robot.
+
+46:24.040 --> 46:33.040
+ Part of the safety definition of how safe a robot is in that case is how effective is the human in monitoring it.
+
+46:33.040 --> 46:43.040
+ Do you think that's ultimately not a good approach in sort of having a human in the picture,
+
+46:43.040 --> 46:51.040
+ not as a bystander or part of the infrastructure, but really as part of what's required to make the system safe?
+
+46:51.040 --> 46:53.040
+ This is harder than it sounds.
+
+46:53.040 --> 47:01.040
+ I think, you know, if you, I mean, I'm sure you've driven before in highways and so on,
+
+47:01.040 --> 47:10.040
+ it's really very hard to relinquish controls to a machine and then take over when needed.
+
+47:10.040 --> 47:18.040
+ So I think Tesla's approach is interesting because it allows you to periodically establish some kind of contact with the car.
+
+47:18.040 --> 47:24.040
+ Toyota, on the other hand, is thinking about shared autonomy or collaborative autonomy as a paradigm.
+
+47:24.040 --> 47:31.040
+ If I may argue, these are very, very simple ways of human robot collaboration because the task is pretty boring.
+
+47:31.040 --> 47:34.040
+ You sit in a vehicle, you go from point A to point B.
+
+47:34.040 --> 47:42.040
+ I think the more interesting thing to me is, for example, search and rescue, I've got a human first responder, robot first responders.
+
+47:42.040 --> 47:44.040
+ I got to do something.
+
+47:44.040 --> 47:45.040
+ It's important.
+
+47:45.040 --> 47:47.040
+ I have to do it in two minutes.
+
+47:47.040 --> 47:48.040
+ The building is burning.
+
+47:48.040 --> 47:50.040
+ There's been an explosion.
+
+47:50.040 --> 47:51.040
+ It's collapsed.
+
+47:51.040 --> 47:52.040
+ How do I do it?
+
+47:52.040 --> 47:58.040
+ I think to me, those are the interesting things where it's very, very unstructured and what's the role of the human?
+
+47:58.040 --> 47:59.040
+ What's the role of the robot?
+
+47:59.040 --> 48:02.040
+ Clearly, there's lots of interesting challenges.
+
+48:02.040 --> 48:05.040
+ As a field, I think we're going to make a lot of progress in this area.
+
+48:05.040 --> 48:07.040
+ Yeah, it's an exciting form of collaboration.
+
+48:07.040 --> 48:08.040
+ You're right.
+
+48:08.040 --> 48:15.040
+ In the autonomous driving, the main enemy is just boredom of the human as opposed to the rescue operations.
+
+48:15.040 --> 48:23.040
+ It's literally life and death and the collaboration enables the effective completion of the mission.
+
+48:23.040 --> 48:24.040
+ So it's exciting.
+
+48:24.040 --> 48:27.040
+ Well, in some sense, we're also doing this.
+
+48:27.040 --> 48:37.040
+ You think about the human driving a car and almost invariably the human is trying to estimate the state of the car, the state of the environment, and so on.
+
+48:37.040 --> 48:40.040
+ But what is the car where to estimate the state of the human?
+
+48:40.040 --> 48:47.040
+ So for example, I'm sure you have a smartphone and the smartphone tries to figure out what you're doing and send you reminders.
+
+48:47.040 --> 48:53.040
+ And oftentimes telling you to drive to a certain place, although you have no intention of going there, because it thinks that that's where you should be.
+
+48:53.040 --> 48:59.040
+ Because of some Gmail calendar entry or something like that.
+
+48:59.040 --> 49:02.040
+ And it's trying to constantly figure out who you are, what you're doing.
+
+49:02.040 --> 49:06.040
+ If a car were to do that, maybe that would make the driver safer.
+
+49:06.040 --> 49:14.040
+ Because the car is trying to figure out there's a driver paying attention, looking at his or her eyes, looking at circadian movements.
+
+49:14.040 --> 49:16.040
+ So I think the potential is there.
+
+49:16.040 --> 49:21.040
+ But from the reverse side, it's not robot modeling, but it's human modeling.
+
+49:21.040 --> 49:23.040
+ It's more in the human, right?
+
+49:23.040 --> 49:30.040
+ And I think the robots can do a very good job of modeling humans if you really think about the framework that you have.
+
+49:30.040 --> 49:39.040
+ A human sitting in a cockpit surrounded by sensors, all staring at him, in addition to be staring outside, but also staring at him.
+
+49:39.040 --> 49:41.040
+ I think there's a real synergy there.
+
+49:41.040 --> 49:48.040
+ Yeah, I love that problem because it's the new 21st century form of psychology actually, AI enabled psychology.
+
+49:48.040 --> 49:54.040
+ A lot of people have sci fi inspired fears of walking robots like those from Boston Dynamics.
+
+49:54.040 --> 49:59.040
+ If you just look at shows on Netflix and so on, or flying robots like those you work with.
+
+49:59.040 --> 50:03.040
+ How would you, how do you think about those fears?
+
+50:03.040 --> 50:05.040
+ How would you alleviate those fears?
+
+50:05.040 --> 50:09.040
+ Do you have inklings, echoes of those same concerns?
+
+50:09.040 --> 50:23.040
+ Any time we develop a technology meaning to have positive impact in the world, there's always a worry that somebody could subvert those technologies and use it in an adversarial setting.
+
+50:23.040 --> 50:25.040
+ And robotics is no exception, right?
+
+50:25.040 --> 50:29.040
+ So I think it's very easy to weaponize robots.
+
+50:29.040 --> 50:31.040
+ I think we talk about swarms.
+
+50:31.040 --> 50:38.040
+ One thing I worry a lot about is, for us to get swarms to work and do something reliably, it's really hard.
+
+50:38.040 --> 50:44.040
+ But suppose I have this challenge of trying to destroy something.
+
+50:44.040 --> 50:49.040
+ And I have a swarm of robots where only one out of the swarm needs to get to its destination.
+
+50:49.040 --> 50:53.040
+ So that suddenly becomes a lot more doable.
+
+50:53.040 --> 51:00.040
+ And so I worry about this general idea of using autonomy with lots and lots of agents.
+
+51:00.040 --> 51:04.040
+ I mean, having said that, look, a lot of this technology is not very mature.
+
+51:04.040 --> 51:12.040
+ My favorite saying is that if somebody had to develop this technology, wouldn't you rather the good guys do it?
+
+51:12.040 --> 51:21.040
+ So the good guys have a good understanding of the technology so they can figure out how this technology is being used in a bad way or could be used in a bad way and try to defend against it?
+
+51:21.040 --> 51:23.040
+ So we think a lot about that.
+
+51:23.040 --> 51:28.040
+ So we're doing research on how to defend against swarms, for example.
+
+51:28.040 --> 51:29.040
+ That's interesting.
+
+51:29.040 --> 51:36.040
+ There is, in fact, a report by the National Academies on counter UAS technologies.
+
+51:36.040 --> 51:38.040
+ This is a real threat.
+
+51:38.040 --> 51:47.040
+ But we're also thinking about how to defend against this and knowing how swarms work, knowing how autonomy works is, I think, very important.
+
+51:47.040 --> 51:49.040
+ So it's not just politicians?
+
+51:49.040 --> 51:51.040
+ You think engineers have a role in this discussion?
+
+51:51.040 --> 51:52.040
+ Absolutely.
+
+51:52.040 --> 51:59.040
+ I think the days where politicians can be agnostic to technology are gone.
+
+51:59.040 --> 52:05.040
+ I think every politician needs to be literate in technology.
+
+52:05.040 --> 52:09.040
+ And I often say technology is the new liberal art.
+
+52:09.040 --> 52:18.040
+ Understanding how technology will change your life, I think, is important and every human being needs to understand that.
+
+52:18.040 --> 52:22.040
+ And maybe we can elect some engineers to office as well on the other side.
+
+52:22.040 --> 52:25.040
+ What are the biggest open problems in robotics in UV?
+
+52:25.040 --> 52:27.040
+ You said we're in the early days in some sense.
+
+52:27.040 --> 52:30.040
+ What are the problems we would like to solve in robotics?
+
+52:30.040 --> 52:32.040
+ I think there are lots of problems, right?
+
+52:32.040 --> 52:36.040
+ But I would phrase it in the following way.
+
+52:36.040 --> 52:46.040
+ If you look at the robots we're building, they're still very much tailored towards doing specific tasks in specific settings.
+
+52:46.040 --> 52:59.040
+ I think the question of how do you get them to operate in much broader settings where things can change in unstructured environments is up in the air.
+
+52:59.040 --> 53:02.040
+ So think of the self driving cars.
+
+53:02.040 --> 53:05.040
+ Today, we can build a self driving car in a parking lot.
+
+53:05.040 --> 53:09.040
+ We can do level five autonomy in a parking lot.
+
+53:09.040 --> 53:17.040
+ But can you do a level five autonomy in the streets of Napoli in Italy or Mumbai in India?
+
+53:17.040 --> 53:18.040
+ No.
+
+53:18.040 --> 53:27.040
+ So in some sense, when we think about robotics, we have to think about where they're functioning, what kind of environment, what kind of a task.
+
+53:27.040 --> 53:32.040
+ We have no understanding of how to put both those things together.
+
+53:32.040 --> 53:36.040
+ So we're in the very early days of applying it to the physical world.
+
+53:36.040 --> 53:39.040
+ And I was just in Naples, actually.
+
+53:39.040 --> 53:46.040
+ And there's levels of difficulty and complexity depending on which area you're applying it to.
+
+53:46.040 --> 53:47.040
+ I think so.
+
+53:47.040 --> 53:51.040
+ And we don't have a systematic way of understanding that.
+
+53:51.040 --> 54:00.040
+ Everybody says just because a computer can now beat a human at any board game, we suddenly know something about intelligence.
+
+54:00.040 --> 54:01.040
+ That's not true.
+
+54:01.040 --> 54:04.040
+ A computer board game is very, very structured.
+
+54:04.040 --> 54:11.040
+ It is the equivalent of working in a Henry Ford factory where things, parts come, you assemble, move on.
+
+54:11.040 --> 54:14.040
+ It's a very, very, very structured setting.
+
+54:14.040 --> 54:15.040
+ That's the easiest thing.
+
+54:15.040 --> 54:18.040
+ And we know how to do that.
+
+54:18.040 --> 54:23.040
+ So you've done a lot of incredible work at the UPenn University of Pennsylvania Grass Club.
+
+54:23.040 --> 54:26.040
+ You're now Dean of Engineering at UPenn.
+
+54:26.040 --> 54:34.040
+ What advice do you have for a new bright eyed undergrad interested in robotics or AI or engineering?
+
+54:34.040 --> 54:37.040
+ Well, I think there's really three things.
+
+54:37.040 --> 54:45.040
+ One is you have to get used to the idea that the world will not be the same in five years or four years whenever you graduate, right?
+
+54:45.040 --> 54:46.040
+ Which is really hard to do.
+
+54:46.040 --> 54:53.040
+ So this thing about predicting the future, every one of us needs to be trying to predict the future always.
+
+54:53.040 --> 55:01.040
+ Not because you'll be any good at it, but by thinking about it, I think you sharpen your senses and you become smarter.
+
+55:01.040 --> 55:02.040
+ So that's number one.
+
+55:02.040 --> 55:09.040
+ Number two, it's a callery of the first piece, which is you really don't know what's going to be important.
+
+55:09.040 --> 55:15.040
+ So this idea that I'm going to specialize in something which will allow me to go in a particular direction.
+
+55:15.040 --> 55:22.040
+ It may be interesting, but it's important also to have this breadth so you have this jumping off point.
+
+55:22.040 --> 55:25.040
+ I think the third thing, and this is where I think Penn excels.
+
+55:25.040 --> 55:30.040
+ I mean, we teach engineering, but it's always in the context of the liberal arts.
+
+55:30.040 --> 55:32.040
+ It's always in the context of society.
+
+55:32.040 --> 55:35.040
+ As engineers, we cannot afford to lose sight of that.
+
+55:35.040 --> 55:37.040
+ So I think that's important.
+
+55:37.040 --> 55:43.040
+ But I think one thing that people underestimate when they do robotics is the importance of mathematical foundations,
+
+55:43.040 --> 55:47.040
+ the importance of representations.
+
+55:47.040 --> 55:56.040
+ Not everything can just be solved by looking for ROS packages on the Internet or to find a deep neural network that works.
+
+55:56.040 --> 56:06.040
+ I think the representation question is key, even to machine learning, where if you ever hope to achieve or get to explainable AI,
+
+56:06.040 --> 56:09.040
+ somehow there need to be representations that you can understand.
+
+56:09.040 --> 56:16.040
+ So if you want to do robotics, you should also do mathematics, and you said liberal arts, a little literature.
+
+56:16.040 --> 56:19.040
+ If you want to build a robot, you should be reading Dostoyevsky.
+
+56:19.040 --> 56:20.040
+ I agree with that.
+
+56:20.040 --> 56:21.040
+ Very good.
+
+56:21.040 --> 56:24.040
+ So Vijay, thank you so much for talking today. It was an honor.
+
+56:24.040 --> 56:47.040
+ Thank you. It was just a very exciting conversation. Thank you.
+

vtt/episode_039_small.vtt

@@ -0,0 +1,1994 @@
+WEBVTT
+
+00:00.000 --> 00:02.800
+ The following is a conversation with Colin Engel.
+
+00:02.800 --> 00:05.840
+ He's the CEO and cofounder of iRobot,
+
+00:05.840 --> 00:09.680
+ a robotics company that for 29 years has been creating robots
+
+00:09.680 --> 00:12.600
+ that operate successfully in the real world.
+
+00:12.600 --> 00:15.600
+ Not as a demo or on a scale of dozens,
+
+00:15.600 --> 00:18.880
+ but on a scale of thousands and millions.
+
+00:18.880 --> 00:24.320
+ As of this year, iRobot has sold more than 25 million robots
+
+00:24.320 --> 00:28.200
+ to consumers, including the Roomba vacuum cleaning robot,
+
+00:28.200 --> 00:30.000
+ the Bravo floor mopping robot,
+
+00:30.000 --> 00:34.000
+ and soon the Terra lawn mowing robot.
+
+00:34.000 --> 00:37.680
+ 29 million robots successfully operating autonomously
+
+00:37.680 --> 00:39.680
+ in real people's homes.
+
+00:39.680 --> 00:42.120
+ To me, it's an incredible accomplishment
+
+00:42.120 --> 00:45.120
+ of science, engineering, logistics,
+
+00:45.120 --> 00:48.800
+ and all kinds of general entrepreneurial innovation.
+
+00:48.800 --> 00:51.360
+ Most robotics companies fail.
+
+00:51.360 --> 00:56.880
+ iRobot has survived and succeeded for 29 years.
+
+00:56.880 --> 01:00.160
+ I spent all day at iRobot, including a long tour
+
+01:00.160 --> 01:03.520
+ and conversation with Colin about the history of iRobot,
+
+01:03.520 --> 01:06.720
+ and then sat down for this podcast conversation
+
+01:06.720 --> 01:08.560
+ that would have been much longer
+
+01:08.560 --> 01:10.760
+ if I didn't spend all day learning about
+
+01:10.760 --> 01:13.960
+ and playing with the various robots in the company's history.
+
+01:13.960 --> 01:17.480
+ I'll release the video of the tour separately.
+
+01:17.480 --> 01:20.720
+ Colin, iRobot, its founding team,
+
+01:20.720 --> 01:23.200
+ its current team, and its mission
+
+01:23.200 --> 01:26.200
+ has been and continues to be an inspiration to me
+
+01:26.200 --> 01:28.880
+ and thousands of engineers who are working hard
+
+01:28.880 --> 01:33.000
+ to create AI systems that help real people.
+
+01:33.000 --> 01:35.640
+ This is the Artificial Intelligence Podcast.
+
+01:35.640 --> 01:38.000
+ If you enjoy it, subscribe on YouTube,
+
+01:38.000 --> 01:41.240
+ give it five stars on iTunes, support it on Patreon,
+
+01:41.240 --> 01:43.280
+ or simply connect with me on Twitter
+
+01:43.280 --> 01:47.120
+ at Lex Freedman, spelled F R I D M A N.
+
+01:47.120 --> 01:51.080
+ And now, here's my conversation with Colin Engel.
+
+01:52.120 --> 01:55.120
+ In his 1942 short story, Run Around,
+
+01:55.120 --> 01:58.520
+ from his iRobot collection, Asimov,
+
+01:59.400 --> 02:02.840
+ proposed the three laws of robotics in order,
+
+02:02.840 --> 02:06.800
+ don't harm humans, obey orders, protect yourself.
+
+02:06.800 --> 02:07.640
+ So two questions.
+
+02:07.640 --> 02:11.640
+ First, does the Roomba follow these three laws?
+
+02:11.640 --> 02:14.760
+ And also, more seriously,
+
+02:14.760 --> 02:17.120
+ what role do you hope to see robots take
+
+02:17.120 --> 02:20.280
+ in modern society and in the future world?
+
+02:20.280 --> 02:25.280
+ So the three laws are very thought provoking
+
+02:25.720 --> 02:30.720
+ and require such a profound understanding
+
+02:31.360 --> 02:36.280
+ of the world a robot lives in,
+
+02:36.280 --> 02:38.360
+ the ramifications of its action
+
+02:38.360 --> 02:40.040
+ and its own sense of self,
+
+02:40.040 --> 02:45.040
+ that it's not a relevant bar,
+
+02:46.640 --> 02:48.360
+ at least it won't be a relevant bar
+
+02:48.360 --> 02:50.160
+ for decades to come.
+
+02:50.160 --> 02:54.560
+ And so, if Roomba follows the three laws,
+
+02:54.560 --> 02:56.840
+ and I believe it does,
+
+02:58.040 --> 03:00.920
+ it is designed to help humans not hurt them,
+
+03:00.920 --> 03:03.120
+ it's designed to be inherently safe,
+
+03:03.120 --> 03:05.960
+ and we design it to last a long time.
+
+03:07.200 --> 03:11.600
+ It's not through any AI or intent on the robot's part.
+
+03:11.600 --> 03:14.960
+ It's because following the three laws
+
+03:14.960 --> 03:18.200
+ is aligned with being a good robot product.
+
+03:19.600 --> 03:23.120
+ So I guess it does,
+
+03:23.120 --> 03:27.240
+ but not by explicit design.
+
+03:27.240 --> 03:28.800
+ So then the bigger picture,
+
+03:28.800 --> 03:31.560
+ what role do you hope to see robotics,
+
+03:31.560 --> 03:36.560
+ robots take in what's currently mostly a world of humans?
+
+03:37.360 --> 03:42.360
+ We need robots to help us continue
+
+03:42.360 --> 03:44.960
+ to improve our standard of living.
+
+03:46.160 --> 03:51.160
+ We need robots because the average age of humanity
+
+03:52.840 --> 03:55.040
+ is increasing very quickly,
+
+03:55.040 --> 03:59.760
+ and simply the number of people young enough
+
+03:59.760 --> 04:02.480
+ and spry enough to care for the elder
+
+04:03.880 --> 04:07.880
+ growing demographic is inadequate.
+
+04:08.800 --> 04:11.440
+ And so what is the role of robots?
+
+04:11.440 --> 04:15.000
+ Today, the role is to make our lives a little easier,
+
+04:15.000 --> 04:17.400
+ a little cleaner, maybe a little healthier.
+
+04:18.520 --> 04:22.160
+ But in time, robots are gonna be the difference
+
+04:22.160 --> 04:25.720
+ between real gut wrenching declines
+
+04:25.720 --> 04:28.360
+ in our ability to live independently
+
+04:28.360 --> 04:30.280
+ and maintain our standard of living,
+
+04:30.280 --> 04:34.920
+ and a future that is the bright one
+
+04:34.920 --> 04:37.520
+ where we have more control of our lives,
+
+04:37.520 --> 04:40.560
+ can spend more of our time focused
+
+04:40.560 --> 04:43.280
+ on activities we choose.
+
+04:44.680 --> 04:48.840
+ And I'm so honored and excited to be
+
+04:48.840 --> 04:50.520
+ playing a role in that journey.
+
+04:50.520 --> 04:51.800
+ So you've given me a tour,
+
+04:51.800 --> 04:54.280
+ you showed me some of the long histories now,
+
+04:54.280 --> 04:57.280
+ 29 years that iRobot has been at it,
+
+04:57.280 --> 04:59.280
+ creating some incredible robots.
+
+04:59.280 --> 05:01.280
+ You showed me PacBot,
+
+05:01.280 --> 05:03.200
+ you showed me a bunch of other stuff
+
+05:03.200 --> 05:04.600
+ that led up to Roomba,
+
+05:04.600 --> 05:08.760
+ that led to Brava and Tara.
+
+05:08.760 --> 05:13.760
+ So let's skip that incredible history
+
+05:14.080 --> 05:15.080
+ in the interest of time,
+
+05:15.080 --> 05:16.160
+ because we already talked about it,
+
+05:16.160 --> 05:18.080
+ I'll show this incredible footage.
+
+05:18.080 --> 05:22.680
+ You mentioned elderly and robotics and society.
+
+05:22.680 --> 05:26.320
+ I think the home is a fascinating place for robots to be.
+
+05:26.320 --> 05:29.840
+ So where do you see robots in the home?
+
+05:29.840 --> 05:31.680
+ Currently, I would say, once again,
+
+05:31.680 --> 05:34.560
+ probably most homes in the world don't have a robot.
+
+05:34.560 --> 05:36.200
+ So how do you see that changing?
+
+05:36.200 --> 05:39.880
+ Where do you think is the big initial value add
+
+05:39.880 --> 05:41.960
+ that robots can do?
+
+05:41.960 --> 05:46.960
+ So iRobot has sort of over the years narrowed in on the home,
+
+05:48.160 --> 05:53.160
+ the consumer's home as the place where we want to innovate
+
+05:53.160 --> 05:58.160
+ and deliver tools that will help a home be
+
+06:00.840 --> 06:04.280
+ a more automatically maintained place,
+
+06:04.280 --> 06:06.840
+ a healthier place, a safer place
+
+06:06.840 --> 06:11.520
+ and perhaps even a more efficient place to be.
+
+06:11.520 --> 06:16.520
+ And today vacuum we mop, soon we'll be mowing your lawn,
+
+06:17.000 --> 06:21.520
+ but where things are going is,
+
+06:22.760 --> 06:27.120
+ when do we get to the point where the home,
+
+06:27.120 --> 06:29.160
+ not just the robots that live in your home,
+
+06:29.160 --> 06:32.200
+ but the home itself becomes part of a system
+
+06:32.200 --> 06:36.000
+ that maintains itself and plays an active role
+
+06:36.000 --> 06:40.800
+ in caring for and helping the people live in that home.
+
+06:40.800 --> 06:43.240
+ And I see everything that we're doing
+
+06:43.240 --> 06:46.200
+ as steps along the path toward that future.
+
+06:46.200 --> 06:47.760
+ So what are the steps?
+
+06:47.760 --> 06:51.760
+ So if we can summarize some of the history of Roomba,
+
+06:53.280 --> 06:55.560
+ you've mentioned, and maybe you can elaborate on it,
+
+06:55.560 --> 06:57.280
+ but you mentioned that the early days
+
+06:57.280 --> 07:02.280
+ were really taking a robot from something that works
+
+07:02.360 --> 07:04.920
+ either in the lab or something that works in the field
+
+07:04.920 --> 07:09.920
+ that helps soldiers do the difficult work they do
+
+07:10.240 --> 07:12.680
+ to actually be in the hands of consumers
+
+07:12.680 --> 07:15.680
+ and tens of thousands, hundreds of thousands of robots
+
+07:15.680 --> 07:18.520
+ that don't break down over how much people love them
+
+07:18.520 --> 07:21.480
+ over months of very extensive use.
+
+07:21.480 --> 07:22.880
+ So that was the big first step.
+
+07:22.880 --> 07:26.040
+ And then the second big step was the ability
+
+07:26.040 --> 07:29.920
+ to sense the environment, to build a map, to localize,
+
+07:29.920 --> 07:32.600
+ to be able to build a picture of the home
+
+07:32.600 --> 07:34.640
+ that the human can then attach labels to
+
+07:34.640 --> 07:38.440
+ in terms of giving some semantic knowledge
+
+07:38.440 --> 07:40.920
+ to the robot about its environment.
+
+07:40.920 --> 07:44.880
+ Okay, so that's like a huge, two big, huge steps.
+
+07:46.280 --> 07:47.560
+ Maybe you can comment on them,
+
+07:47.560 --> 07:51.080
+ but also what is the next step
+
+07:51.080 --> 07:54.760
+ of making a robot part of the home?
+
+07:54.760 --> 07:55.600
+ Sure.
+
+07:55.600 --> 08:00.600
+ So the goal is to make a home that takes care of itself,
+
+08:01.320 --> 08:03.720
+ takes care of the people in the home
+
+08:03.720 --> 08:07.880
+ and gives a user an experience of just living their life
+
+08:07.880 --> 08:10.880
+ and the home is somehow doing the right thing,
+
+08:10.880 --> 08:14.160
+ turning on and off lights when you leave,
+
+08:14.160 --> 08:17.280
+ cleaning up the environment.
+
+08:17.280 --> 08:22.280
+ And we went from robots that were right in the lab,
+
+08:22.280 --> 08:26.760
+ but were both too expensive and not sufficiently capable
+
+08:26.760 --> 08:30.480
+ to ever do an acceptable job of anything
+
+08:30.480 --> 08:34.080
+ other than being a toy or a curio in your home
+
+08:34.080 --> 08:39.080
+ to something that was both affordable
+
+08:38.880 --> 08:42.440
+ and sufficiently effective to drive,
+
+08:42.440 --> 08:45.360
+ be above threshold and drive purchase intent.
+
+08:47.520 --> 08:50.000
+ Now we've disrupted the intent of the work
+
+08:50.000 --> 08:54.400
+ and now we've disrupted the entire vacuuming industry.
+
+08:55.480 --> 08:59.840
+ The number one selling vacuums, for example, in the US
+
+08:59.840 --> 09:02.480
+ are Roombas, so not robot vacuums,
+
+09:02.480 --> 09:05.160
+ but vacuums and that's really crazy and weird.
+
+09:05.160 --> 09:08.080
+ We need to pause that, I mean, that's incredible.
+
+09:08.080 --> 09:10.520
+ That's incredible that a robot
+
+09:10.520 --> 09:15.520
+ is the number one selling thing that does something.
+
+09:15.560 --> 09:16.400
+ Yep.
+
+09:16.400 --> 09:18.240
+ Something as essential as vacuuming.
+
+09:18.240 --> 09:20.120
+ So we're... Congratulations.
+
+09:20.120 --> 09:20.960
+ Thank you.
+
+09:20.960 --> 09:22.440
+ It's still kind of fun to say,
+
+09:22.440 --> 09:26.600
+ but just because this was a crazy idea
+
+09:26.600 --> 09:30.920
+ that just started in a room here,
+
+09:30.920 --> 09:33.800
+ we're like, do you think we can do this?
+
+09:35.000 --> 09:36.160
+ Hey, let's give it a try.
+
+09:38.000 --> 09:40.440
+ But now the robots are starting
+
+09:40.440 --> 09:42.840
+ to understand their environment.
+
+09:42.840 --> 09:45.280
+ And if you think about the next step,
+
+09:45.280 --> 09:47.880
+ there's two dimensions.
+
+09:48.720 --> 09:53.040
+ I've been working so hard since the beginning of iRobot
+
+09:53.040 --> 09:55.080
+ to make robots are autonomous,
+
+09:55.080 --> 09:59.120
+ that they're smart enough and understand their task enough
+
+09:59.120 --> 10:02.480
+ that they can just go do it without human involvement.
+
+10:04.160 --> 10:07.440
+ Now what I'm really excited and working on
+
+10:07.440 --> 10:09.520
+ is how do I make them less autonomous?
+
+10:09.520 --> 10:14.520
+ Meaning that the robot is supposed to be your partner,
+
+10:15.080 --> 10:20.080
+ not this automaton that just goes and does what a robot does.
+
+10:20.240 --> 10:22.440
+ And so that if you tell it,
+
+10:22.440 --> 10:27.160
+ hey, I just dropped some flour by the fridge in the kitchen.
+
+10:27.160 --> 10:29.000
+ Can you deal with it?
+
+10:29.000 --> 10:31.840
+ Wouldn't be awesome if the right thing just happened
+
+10:31.840 --> 10:35.240
+ based on that utterance.
+
+10:35.240 --> 10:37.960
+ And to some extent that's less autonomous
+
+10:37.960 --> 10:40.120
+ because it's actually listening to you,
+
+10:40.120 --> 10:44.400
+ understanding the context and intent of the sentence,
+
+10:44.400 --> 10:49.400
+ mapping it against its understanding of the home it lives in
+
+10:50.040 --> 10:52.680
+ and knowing what to do.
+
+10:52.680 --> 10:56.360
+ And so that's an area of research.
+
+10:56.360 --> 10:59.360
+ It's an area where we're starting to roll out features.
+
+10:59.360 --> 11:02.880
+ You can now tell your robot to clean up the kitchen
+
+11:02.880 --> 11:05.880
+ and it knows what the kitchen is and can do that.
+
+11:05.880 --> 11:09.360
+ And that's sort of 1.0 of where we're going.
+
+11:10.440 --> 11:12.480
+ The other cool thing is that
+
+11:12.480 --> 11:14.640
+ we're starting to know where stuff is.
+
+11:14.640 --> 11:16.040
+ And why is that important?
+
+11:16.040 --> 11:21.040
+ Well, robots are supposed to have arms, right?
+
+11:21.520 --> 11:24.200
+ Data had an arm, Rosie had an arm,
+
+11:24.200 --> 11:25.240
+ Robbie the robot had an arm.
+
+11:25.240 --> 11:26.680
+ I mean, robots are, you know,
+
+11:26.680 --> 11:29.560
+ they are physical things that move around in an environment
+
+11:29.560 --> 11:31.280
+ they're supposed to like do work.
+
+11:31.280 --> 11:34.120
+ And if you think about it,
+
+11:34.120 --> 11:35.440
+ if a robot doesn't know anything,
+
+11:35.440 --> 11:38.760
+ where anything is, why should it have an arm?
+
+11:38.760 --> 11:43.760
+ But with this new dawn of home understanding
+
+11:44.320 --> 11:47.680
+ that we're starting to go enjoy,
+
+11:47.680 --> 11:49.320
+ I know where the kitchen is.
+
+11:49.320 --> 11:52.040
+ I might in the future know where the refrigerator is.
+
+11:52.040 --> 11:55.280
+ I might, if I had an arm, be able to find the handle,
+
+11:55.280 --> 11:58.440
+ open it and even get myself a beer.
+
+11:58.440 --> 12:01.920
+ Obviously that's one of the true dreams of robotics
+
+12:01.920 --> 12:03.520
+ is to have robots bringing us a beer
+
+12:03.520 --> 12:05.280
+ while we watch television.
+
+12:05.280 --> 12:10.280
+ But I think that that new category of tasks
+
+12:11.000 --> 12:14.240
+ where physical manipulation, robot arms
+
+12:14.240 --> 12:19.240
+ is just a popery of new opportunity and excitement.
+
+12:20.200 --> 12:23.800
+ And you see humans as a crucial part of that.
+
+12:23.800 --> 12:26.280
+ So you kind of mentioned that.
+
+12:26.280 --> 12:28.960
+ And I personally find that a really compelling idea.
+
+12:28.960 --> 12:33.960
+ I think full autonomy can only take us so far,
+
+12:33.960 --> 12:35.360
+ especially in the home.
+
+12:35.360 --> 12:38.920
+ So you see humans as helping the robot understand
+
+12:38.920 --> 12:42.600
+ or give deeper meaning to the spatial information.
+
+12:43.600 --> 12:45.680
+ Right, it's a partnership.
+
+12:46.800 --> 12:51.800
+ The robot is supposed to operate according to descriptors
+
+12:52.800 --> 12:55.620
+ that you would use to describe your own home.
+
+12:57.040 --> 13:02.040
+ The robot is supposed to in lieu of better direction
+
+13:02.040 --> 13:03.840
+ kind of go about its routine,
+
+13:03.840 --> 13:07.520
+ which ought to be basically right
+
+13:07.520 --> 13:12.200
+ and lead to a home maintained
+
+13:12.200 --> 13:14.840
+ in a way that it's learned you like,
+
+13:14.840 --> 13:19.840
+ but also be perpetually ready to take direction
+
+13:21.440 --> 13:26.440
+ that would activate a different set of behaviors or actions
+
+13:26.880 --> 13:28.880
+ to meet a current need
+
+13:28.880 --> 13:32.320
+ to the extent it could actually perform that task.
+
+13:32.320 --> 13:33.560
+ So I gotta ask you,
+
+13:33.560 --> 13:36.960
+ I think this is a fundamental and a fascinating question
+
+13:36.960 --> 13:39.760
+ because iRobot has been a successful company
+
+13:39.760 --> 13:42.320
+ and a rare successful robotics company.
+
+13:42.320 --> 13:46.720
+ So Anki, Gibo, Mayfield Robotics with a Robot Curry,
+
+13:46.720 --> 13:49.160
+ SciFi Works, Rethink Robotics.
+
+13:49.160 --> 13:51.320
+ These were robotics companies that were founded
+
+13:51.320 --> 13:52.920
+ and run by brilliant people.
+
+13:54.000 --> 13:58.760
+ But all very unfortunately, at least for us roboticists,
+
+13:58.760 --> 14:02.120
+ and all went out of business recently.
+
+14:02.120 --> 14:05.160
+ So why do you think they didn't last longer?
+
+14:05.160 --> 14:07.000
+ Why do you think it is so hard
+
+14:07.000 --> 14:10.680
+ to keep a robotics company alive?
+
+14:10.680 --> 14:14.160
+ You know, I say this only partially in jest
+
+14:14.160 --> 14:16.760
+ that back in the day before Roomba,
+
+14:17.840 --> 14:22.840
+ you know, I was a high tech entrepreneur building robots,
+
+14:23.920 --> 14:26.440
+ but it wasn't until I became a vacuum cleaner salesman
+
+14:26.440 --> 14:29.420
+ that we had any success.
+
+14:29.420 --> 14:34.240
+ So I mean, the point is technology alone
+
+14:34.240 --> 14:37.680
+ doesn't equal a successful business.
+
+14:37.680 --> 14:42.680
+ We need to go and find the compelling need
+
+14:43.600 --> 14:45.920
+ where the robot that we're creating
+
+14:47.640 --> 14:52.640
+ can deliver clearly more value
+
+14:52.800 --> 14:55.440
+ to the end user than it costs.
+
+14:55.440 --> 14:59.040
+ And this is not a marginal thing
+
+14:59.040 --> 15:01.800
+ where you're looking at the skin, like, it's close.
+
+15:01.800 --> 15:04.380
+ Maybe we can hold our breath and make it work.
+
+15:04.380 --> 15:09.380
+ It's clearly more value than the cost of the robot
+
+15:11.560 --> 15:13.920
+ to bring, you know, in the store.
+
+15:13.920 --> 15:17.360
+ And I think that the challenge has been finding
+
+15:17.360 --> 15:22.360
+ those businesses where that's true
+
+15:22.360 --> 15:26.400
+ that's true in a sustainable fashion.
+
+15:28.240 --> 15:33.240
+ You know, when you get into entertainment style things,
+
+15:34.600 --> 15:38.240
+ you could be the cat's meow one year,
+
+15:38.240 --> 15:42.440
+ but 85% of toys, regardless of their merit,
+
+15:43.400 --> 15:45.640
+ fail to make it to their second season.
+
+15:45.640 --> 15:47.720
+ It's just super hard to do so.
+
+15:47.720 --> 15:52.720
+ And so that that's just a tough business.
+
+15:53.800 --> 15:58.800
+ And there's been a lot of experimentation around
+
+15:59.200 --> 16:02.640
+ what is the right type of social companion?
+
+16:02.640 --> 16:05.840
+ What is the right robot in the home
+
+16:05.840 --> 16:10.840
+ that is doing something other than tasks people do every week
+
+16:14.560 --> 16:16.400
+ that they'd rather not do?
+
+16:16.400 --> 16:20.880
+ And I'm not sure we've got it all figured out right.
+
+16:20.880 --> 16:22.960
+ And so that you get brilliant roboticists
+
+16:22.960 --> 16:25.680
+ with super interesting robots
+
+16:25.680 --> 16:29.800
+ that ultimately don't quite have
+
+16:29.800 --> 16:34.040
+ that magical user experience and thus the,
+
+16:36.480 --> 16:40.560
+ that value benefit equation remains ambiguous.
+
+16:40.560 --> 16:43.960
+ So you as somebody who dreams of robots,
+
+16:43.960 --> 16:45.720
+ you know, changing the world,
+
+16:45.720 --> 16:46.880
+ what's your estimate?
+
+16:47.840 --> 16:52.840
+ Why, how big is the space of applications
+
+16:53.200 --> 16:55.800
+ that fit the criteria that you just described
+
+16:55.800 --> 16:58.080
+ where you can really demonstrate
+
+16:58.080 --> 17:00.520
+ an obvious significant value
+
+17:00.520 --> 17:04.620
+ over the alternative non robotic solution?
+
+17:05.720 --> 17:08.680
+ Well, I think that we're just about none of the way
+
+17:10.000 --> 17:13.360
+ to achieving the potential of robotics at home.
+
+17:13.360 --> 17:18.360
+ But we have to do it in a really eyes wide open,
+
+17:20.400 --> 17:21.920
+ honest fashion.
+
+17:21.920 --> 17:25.400
+ And so another way to put that is the potential is infinite
+
+17:25.400 --> 17:27.040
+ because we did take a few steps
+
+17:27.040 --> 17:29.640
+ but you're saying those steps are just very initial steps.
+
+17:29.640 --> 17:32.560
+ So the Roomba is a hugely successful product
+
+17:32.560 --> 17:34.440
+ but you're saying that's just the very, very beginning.
+
+17:34.440 --> 17:36.520
+ That's just the very, very beginning.
+
+17:36.520 --> 17:37.960
+ It's the foot in the door.
+
+17:37.960 --> 17:40.840
+ And, you know, I think I was lucky
+
+17:40.840 --> 17:45.120
+ that in the early days of robotics,
+
+17:45.120 --> 17:48.360
+ people would ask me, when are you gonna clean my floor?
+
+17:48.360 --> 17:52.240
+ It was something that I grew up saying,
+
+17:53.560 --> 17:54.800
+ I got all these really good ideas
+
+17:54.800 --> 17:58.080
+ but everyone seems to want their floor clean.
+
+17:58.080 --> 18:02.480
+ And so maybe we should do that.
+
+18:02.480 --> 18:03.320
+ Yeah, your good ideas.
+
+18:03.320 --> 18:05.840
+ Earn the right to do the next thing after that.
+
+18:05.840 --> 18:10.160
+ So the good ideas have to match with the desire of the people
+
+18:10.160 --> 18:13.280
+ and then the actual cost has to like the business,
+
+18:13.280 --> 18:16.640
+ the financial aspect has to all match together.
+
+18:16.640 --> 18:20.160
+ Yeah, during our partnership
+
+18:20.160 --> 18:22.200
+ back a number of years ago with Johnson Wax,
+
+18:22.200 --> 18:27.200
+ they would explain to me that they would go into homes
+
+18:29.480 --> 18:32.520
+ and just watch how people lived
+
+18:32.520 --> 18:35.560
+ and try to figure out what were they doing
+
+18:35.560 --> 18:39.960
+ that they really didn't really like to do
+
+18:39.960 --> 18:42.440
+ but they had to do it frequently enough
+
+18:42.440 --> 18:47.440
+ that it was top of mind and understood as a burden.
+
+18:51.720 --> 18:53.000
+ Hey, let's make a product
+
+18:54.480 --> 18:58.480
+ or come up with a solution to make that pain point
+
+18:58.480 --> 19:02.040
+ less challenging.
+
+19:02.040 --> 19:07.040
+ And sometimes we do certain burdens so often as a society
+
+19:07.400 --> 19:09.400
+ that we actually don't even realize,
+
+19:09.400 --> 19:10.680
+ like it's actually hard to see
+
+19:10.680 --> 19:13.200
+ that that burden is something that could be removed.
+
+19:13.200 --> 19:15.760
+ So it does require just going into the home
+
+19:15.760 --> 19:19.560
+ and staring at, wait, how do I actually live life?
+
+19:19.560 --> 19:21.080
+ What are the pain points?
+
+19:21.080 --> 19:26.080
+ Yeah, and it getting those insights is a lot harder
+
+19:26.400 --> 19:29.360
+ than it would seem it should be in retrospect.
+
+19:29.360 --> 19:33.120
+ So how hard on that point,
+
+19:33.120 --> 19:37.480
+ I mean, one of the big challenges of robotics
+
+19:37.480 --> 19:40.240
+ is driving the cost to something,
+
+19:40.240 --> 19:42.240
+ driving the cost down to something
+
+19:42.240 --> 19:45.680
+ that consumers, people would afford.
+
+19:45.680 --> 19:48.880
+ So people would be less likely to buy a Roomba
+
+19:48.880 --> 19:52.280
+ if it costs $500,000, right?
+
+19:52.280 --> 19:55.840
+ Which is probably sort of what a Roomba would cost
+
+19:55.840 --> 19:58.040
+ several decades ago.
+
+19:58.040 --> 20:02.200
+ So how do you drive, which I mentioned is very difficult.
+
+20:02.200 --> 20:04.200
+ How do you drive the cost of a Roomba
+
+20:04.200 --> 20:07.920
+ or a robot down such that people would want to buy it?
+
+20:07.920 --> 20:09.720
+ When I started building robots,
+
+20:09.720 --> 20:12.240
+ the cost of the robot had a lot to do
+
+20:12.240 --> 20:15.480
+ with the amount of time it took to build it.
+
+20:15.480 --> 20:18.400
+ And so that we would build our robots out of aluminum,
+
+20:18.400 --> 20:21.160
+ I would go spend my time in the machine shop
+
+20:21.160 --> 20:22.840
+ on the milling machine,
+
+20:23.840 --> 20:28.000
+ cutting out the parts and so forth.
+
+20:28.000 --> 20:29.720
+ And then when we got into the toy industry,
+
+20:29.720 --> 20:34.520
+ I realized that if we were building at scale,
+
+20:34.520 --> 20:36.000
+ I could determine the cost of the Roomba
+
+20:36.000 --> 20:38.920
+ instead of adding up all the hours to mill out the parts,
+
+20:38.920 --> 20:40.360
+ but by weighing it.
+
+20:42.080 --> 20:44.200
+ And that's liberating.
+
+20:44.200 --> 20:49.200
+ You can say, wow, the world has just changed
+
+20:49.560 --> 20:53.160
+ as I think about construction in a different way.
+
+20:53.160 --> 20:56.880
+ The 3D CAD tools that are available to us today,
+
+20:56.880 --> 21:01.680
+ the operating at scale where I can do tooling
+
+21:01.680 --> 21:06.680
+ and injection mold, an arbitrarily complicated part,
+
+21:07.080 --> 21:09.800
+ and the cost is going to be basically
+
+21:09.800 --> 21:12.560
+ the weight of the plastic in that part
+
+21:13.920 --> 21:16.360
+ is incredibly exciting and liberating
+
+21:16.360 --> 21:18.560
+ and opens up all sorts of opportunities.
+
+21:18.560 --> 21:23.560
+ And for the sensing part of it, where we are today
+
+21:23.560 --> 21:28.560
+ is instead of trying to build skin,
+
+21:29.280 --> 21:31.440
+ which is like really hard for a long time.
+
+21:31.440 --> 21:36.440
+ I spent creating strategies and ideas
+
+21:38.240 --> 21:42.720
+ around how could we duplicate the skin on the human body
+
+21:42.720 --> 21:45.360
+ because it's such an amazing sensor.
+
+21:47.880 --> 21:49.600
+ Instead of going down that path,
+
+21:49.600 --> 21:53.000
+ why don't we focus on vision?
+
+21:54.000 --> 21:59.000
+ And how many of the problems that face a robot
+
+22:00.080 --> 22:04.880
+ trying to do real work could be solved
+
+22:04.880 --> 22:08.480
+ with a cheap camera and a big ass computer?
+
+22:09.480 --> 22:12.440
+ And Moore's Law continues to work.
+
+22:12.440 --> 22:16.520
+ The cell phone industry, the mobile industry
+
+22:16.520 --> 22:18.800
+ is giving us better and better tools
+
+22:18.800 --> 22:21.120
+ that can run on these embedded computers.
+
+22:21.120 --> 22:26.120
+ And I think we passed an important moment,
+
+22:26.600 --> 22:31.600
+ maybe two years ago, where you could put
+
+22:32.760 --> 22:37.520
+ machine vision capable processors on robots
+
+22:37.520 --> 22:39.640
+ at consumer price points.
+
+22:39.640 --> 22:43.040
+ And I was waiting for it to happen.
+
+22:43.040 --> 22:46.600
+ We avoided putting lasers on our robots
+
+22:46.600 --> 22:51.600
+ to do navigation and instead spent years researching
+
+22:51.840 --> 22:54.640
+ how to do vision based navigation
+
+22:54.640 --> 22:58.440
+ because you could just see it
+
+22:58.440 --> 23:01.640
+ where these technology trends were going
+
+23:01.640 --> 23:05.880
+ and between injection molded plastic
+
+23:05.880 --> 23:08.040
+ and a camera with a computer
+
+23:08.040 --> 23:10.840
+ capable of running machine learning
+
+23:10.840 --> 23:12.560
+ and visual object recognition,
+
+23:12.560 --> 23:15.560
+ I could build an incredibly affordable,
+
+23:15.560 --> 23:18.760
+ incredibly capable robot
+
+23:18.760 --> 23:21.240
+ and that's gonna be the future.
+
+23:21.240 --> 23:23.440
+ So you know, on that point with a small tangent
+
+23:23.440 --> 23:25.000
+ but I think an important one,
+
+23:25.000 --> 23:27.600
+ another industry in which I would say
+
+23:27.600 --> 23:31.880
+ the only other industry in which there is automation
+
+23:31.880 --> 23:34.840
+ actually touching people's lives today
+
+23:34.840 --> 23:36.520
+ is autonomous vehicles.
+
+23:37.640 --> 23:42.360
+ What the vision is described of using computer vision
+
+23:42.360 --> 23:44.520
+ and using cheap camera sensors,
+
+23:44.520 --> 23:48.320
+ there's a debate on that of LiDAR versus computer vision
+
+23:48.320 --> 23:53.320
+ and sort of the Elon Musk famously said
+
+23:53.320 --> 23:58.320
+ that LiDAR is a crutch that really in the long term,
+
+23:58.440 --> 24:00.880
+ camera only is the right solution
+
+24:00.880 --> 24:03.520
+ which echoes some of the ideas you're expressing.
+
+24:03.520 --> 24:05.120
+ Of course, the domain
+
+24:05.120 --> 24:07.720
+ in terms of its safety criticality is different.
+
+24:07.720 --> 24:10.720
+ But what do you think about that approach
+
+24:10.720 --> 24:13.480
+ in the autonomous vehicle space?
+
+24:13.480 --> 24:15.200
+ And in general, do you see a connection
+
+24:15.200 --> 24:18.560
+ between the incredible real world challenges
+
+24:18.560 --> 24:20.800
+ you have to solve in the home with Roomba
+
+24:20.800 --> 24:23.000
+ and I saw a demonstration of some of them,
+
+24:23.000 --> 24:27.920
+ corner cases literally and autonomous vehicles.
+
+24:27.920 --> 24:31.720
+ So there's absolutely a tremendous overlap
+
+24:31.720 --> 24:35.520
+ between both the problems, you know,
+
+24:35.520 --> 24:38.680
+ a robot vacuum and autonomous vehicle are trying to solve
+
+24:38.680 --> 24:41.880
+ and the tools and the types of sensors
+
+24:41.880 --> 24:46.720
+ that are being applied in the pursuit of the solutions.
+
+24:48.040 --> 24:53.040
+ In my world, my environment is actually much harder
+
+24:54.680 --> 24:57.320
+ than the environment in automobile travels.
+
+24:57.320 --> 25:02.320
+ We don't have roads, we have t shirts, we have steps,
+
+25:02.720 --> 25:07.120
+ we have a near infinite number of patterns and colors
+
+25:07.120 --> 25:10.200
+ and surface textures on the floor.
+
+25:10.200 --> 25:12.560
+ Especially from a visual perspective.
+
+25:12.560 --> 25:14.760
+ Yeah, visually it's really tough.
+
+25:14.760 --> 25:18.880
+ Is an infinitely variable.
+
+25:18.880 --> 25:22.560
+ On the other hand, safety is way easier on the inside.
+
+25:22.560 --> 25:26.440
+ My robots, they're not very heavy,
+
+25:26.440 --> 25:28.280
+ they're not very fast.
+
+25:28.280 --> 25:31.480
+ If they bump into your foot, you think it's funny.
+
+25:32.720 --> 25:36.920
+ And, you know, and autonomous vehicles
+
+25:36.920 --> 25:39.400
+ kind of have the inverse problem.
+
+25:39.400 --> 25:44.400
+ And so that for me saying vision is the future,
+
+25:45.920 --> 25:47.800
+ I can say that without reservation.
+
+25:49.480 --> 25:54.480
+ For autonomous vehicles, I think I believe what Elon's saying
+
+25:55.360 --> 25:59.040
+ about the future is ultimately gonna be vision.
+
+25:59.040 --> 26:01.000
+ Maybe if we put a cheap lighter on there
+
+26:01.000 --> 26:03.280
+ as a backup sensor, it might not be the worst idea
+
+26:03.280 --> 26:04.120
+ in the world.
+
+26:04.120 --> 26:04.960
+ So the stakes are much higher.
+
+26:04.960 --> 26:05.800
+ The stakes are much higher.
+
+26:05.800 --> 26:08.200
+ You have to be much more careful thinking through
+
+26:08.200 --> 26:10.720
+ how far away that future is.
+
+26:10.720 --> 26:14.240
+ Right, but I think that the primary
+
+26:16.080 --> 26:19.320
+ environmental understanding sensor
+
+26:19.320 --> 26:21.760
+ is going to be a visual system.
+
+26:21.760 --> 26:23.000
+ Visual system.
+
+26:23.000 --> 26:25.560
+ So on that point, well, let me ask,
+
+26:25.560 --> 26:29.440
+ do you hope there's an iRobot robot in every home
+
+26:29.440 --> 26:30.920
+ in the world one day?
+
+26:31.880 --> 26:34.880
+ I expect there to be at least one iRobot robot
+
+26:34.880 --> 26:36.440
+ in every home.
+
+26:36.440 --> 26:41.160
+ You know, we've sold 25 million robots.
+
+26:41.160 --> 26:44.600
+ So we're in about 10% of US homes,
+
+26:44.600 --> 26:45.760
+ which is a great start.
+
+26:47.120 --> 26:51.080
+ But I think that when we think about the numbers
+
+26:51.080 --> 26:55.840
+ of things that robots can do, you know,
+
+26:55.840 --> 26:58.560
+ today I can vacuum your floor, mop your floor,
+
+26:58.560 --> 27:01.240
+ cut your lawn or soon we'll be able to cut your lawn.
+
+27:01.240 --> 27:06.240
+ But there are more things that we could do in the home.
+
+27:06.720 --> 27:11.520
+ And I hope that we continue using the techniques
+
+27:11.520 --> 27:14.480
+ I described around exploiting computer vision
+
+27:14.480 --> 27:18.680
+ and low cost manufacturing that we'll be able
+
+27:18.680 --> 27:22.680
+ to create these solutions at affordable price points.
+
+27:22.680 --> 27:25.600
+ So let me ask, on that point of a robot in every home,
+
+27:25.600 --> 27:26.880
+ that's my dream as well.
+
+27:26.880 --> 27:28.400
+ I'd love to see that.
+
+27:28.400 --> 27:31.880
+ I think the possibilities there are indeed
+
+27:31.880 --> 27:34.560
+ infinite positive possibilities.
+
+27:34.560 --> 27:39.560
+ But in our current culture, no thanks to science fiction
+
+27:39.800 --> 27:44.720
+ and so on, there's a serious kind of hesitation
+
+27:44.720 --> 27:47.120
+ and anxiety, concern about robots
+
+27:47.120 --> 27:50.080
+ and also a concern about privacy.
+
+27:51.480 --> 27:54.040
+ And it's a fascinating question to me
+
+27:54.040 --> 27:59.040
+ why that concern is amongst a certain group of people
+
+27:59.600 --> 28:02.880
+ is as intense as it is.
+
+28:02.880 --> 28:04.280
+ So you have to think about it
+
+28:04.280 --> 28:05.520
+ because it's a serious concern,
+
+28:05.520 --> 28:08.040
+ but I wonder how you address it best.
+
+28:08.040 --> 28:09.840
+ So from a perspective of a vision sensor,
+
+28:09.840 --> 28:14.200
+ so robots that move about the home and sense the world,
+
+28:14.200 --> 28:19.200
+ how do you alleviate people's privacy concerns?
+
+28:19.720 --> 28:22.880
+ How do you make sure that they can trust iRobot
+
+28:22.880 --> 28:25.360
+ and the robots that they share their home with?
+
+28:26.640 --> 28:28.160
+ I think that's a great question.
+
+28:28.160 --> 28:33.160
+ And we've really leaned way forward on this
+
+28:33.760 --> 28:38.760
+ because given our vision as to the role the company
+
+28:38.880 --> 28:40.720
+ intends to play in the home,
+
+28:43.000 --> 28:45.480
+ really for us, make or break is,
+
+28:45.480 --> 28:50.480
+ can our approach be trusted to protecting the data
+
+28:50.480 --> 28:53.560
+ and the privacy of the people who have our robots?
+
+28:53.560 --> 28:56.920
+ And so we've gone out publicly
+
+28:56.920 --> 29:00.440
+ with a privacy manifesto stating we'll never sell your data.
+
+29:00.440 --> 29:05.440
+ We've adopted GDPR, not just where GDPR is required,
+
+29:05.520 --> 29:10.520
+ but globally, we have ensured that images
+
+29:16.640 --> 29:18.080
+ don't leave the robot.
+
+29:18.080 --> 29:22.120
+ So processing data from the visual sensors
+
+29:22.120 --> 29:23.680
+ happens locally on the robot
+
+29:23.680 --> 29:28.680
+ and only semantic knowledge of the home
+
+29:29.520 --> 29:32.720
+ with the consumer's consent is sent up.
+
+29:32.720 --> 29:34.480
+ We show you what we know
+
+29:34.480 --> 29:39.480
+ and are trying to go use data as an enabler
+
+29:41.560 --> 29:44.120
+ for the performance of the robots
+
+29:44.120 --> 29:49.120
+ with the informed consent and understanding
+
+29:50.040 --> 29:52.440
+ of the people who own those robots.
+
+29:52.440 --> 29:56.880
+ And we take it very seriously.
+
+29:56.880 --> 30:01.880
+ And ultimately, we think that by showing a customer
+
+30:01.880 --> 30:06.880
+ that if you let us build a semantic map of your home
+
+30:07.360 --> 30:09.000
+ and know where the rooms are,
+
+30:09.000 --> 30:11.760
+ well, then you can say clean the kitchen.
+
+30:11.760 --> 30:13.720
+ If you don't want the robot to do that,
+
+30:13.720 --> 30:14.560
+ don't make the map,
+
+30:14.560 --> 30:17.000
+ it'll do its best job cleaning your home,
+
+30:17.000 --> 30:18.640
+ but it won't be able to do that.
+
+30:18.640 --> 30:20.280
+ And if you ever want us to forget
+
+30:20.280 --> 30:22.080
+ that we know that it's your kitchen,
+
+30:22.080 --> 30:26.680
+ you can have confidence that we will do that for you.
+
+30:26.680 --> 30:31.680
+ So we're trying to go and be a sort of a data 2.0
+
+30:34.520 --> 30:37.680
+ perspective company where we treat the data
+
+30:37.680 --> 30:40.800
+ that the robots have of the consumer's home
+
+30:40.800 --> 30:43.200
+ as if it were the consumer's data
+
+30:43.200 --> 30:47.360
+ and that they have rights to it.
+
+30:47.360 --> 30:50.960
+ So we think by being the good guys on this front,
+
+30:50.960 --> 30:53.840
+ we can build the trust and thus be entrusted
+
+30:55.120 --> 31:00.120
+ to enable robots to do more things that are thoughtful.
+
+31:00.200 --> 31:04.520
+ You think people's worries will diminish over time?
+
+31:04.520 --> 31:06.840
+ As a society, broadly speaking,
+
+31:06.840 --> 31:09.320
+ do you think you can win over trust,
+
+31:09.320 --> 31:10.640
+ not just for the company,
+
+31:10.640 --> 31:14.880
+ but just the comfort of people have with AI in their home
+
+31:14.880 --> 31:17.040
+ enriching their lives in some way?
+
+31:17.040 --> 31:19.560
+ I think we're an interesting place today
+
+31:19.560 --> 31:22.400
+ where it's less about winning them over
+
+31:22.400 --> 31:26.240
+ and more about finding a way to talk about privacy
+
+31:26.240 --> 31:28.840
+ in a way that more people can understand.
+
+31:28.840 --> 31:30.920
+ I would tell you that today,
+
+31:30.920 --> 31:33.320
+ when there's a privacy breach,
+
+31:33.320 --> 31:37.040
+ people get very upset and then go to the store
+
+31:37.040 --> 31:38.320
+ and buy the cheapest thing,
+
+31:38.320 --> 31:41.000
+ paying no attention to whether or not the products
+
+31:41.000 --> 31:44.640
+ that they're buying honor privacy standards or not.
+
+31:44.640 --> 31:48.600
+ In fact, if I put on the package of my Roomba,
+
+31:50.080 --> 31:53.640
+ the privacy commitments that we have,
+
+31:53.640 --> 31:58.640
+ I would sell less than I would if I did nothing at all
+
+31:58.720 --> 32:00.400
+ and that needs to change.
+
+32:00.400 --> 32:02.880
+ So it's not a question about earning trust.
+
+32:02.880 --> 32:05.000
+ I think that's necessary but not sufficient.
+
+32:05.000 --> 32:08.440
+ We need to figure out how to have a comfortable set
+
+32:08.440 --> 32:13.440
+ of what is the grade A meat standard applied to privacy
+
+32:14.200 --> 32:18.400
+ that customers can trust and understand
+
+32:18.400 --> 32:20.480
+ and then use in the buying decisions.
+
+32:23.040 --> 32:25.520
+ That will reward companies for good behavior
+
+32:25.520 --> 32:29.880
+ and that will ultimately be how this moves forward.
+
+32:29.880 --> 32:32.680
+ And maybe be part of the conversation
+
+32:32.680 --> 32:34.800
+ between regular people about what it means,
+
+32:34.800 --> 32:36.280
+ what privacy means.
+
+32:36.280 --> 32:38.400
+ If you have some standards, you can say,
+
+32:38.400 --> 32:41.080
+ you can start talking about who's following them,
+
+32:41.080 --> 32:42.680
+ who's not, have more.
+
+32:42.680 --> 32:45.440
+ Because most people are actually quite clueless
+
+32:45.440 --> 32:47.320
+ about all aspects of artificial intelligence
+
+32:47.320 --> 32:48.400
+ or data collection and so on.
+
+32:48.400 --> 32:49.920
+ It would be nice to change that
+
+32:49.920 --> 32:52.760
+ for people to understand the good that AI can do
+
+32:52.760 --> 32:56.520
+ and it's not some system that's trying to steal
+
+32:56.520 --> 32:58.760
+ all the most sensitive data.
+
+32:58.760 --> 33:02.640
+ Do you think, do you dream of a Roomba
+
+33:02.640 --> 33:05.240
+ with human level intelligence one day?
+
+33:05.240 --> 33:10.240
+ So you've mentioned a very successful localization
+
+33:10.520 --> 33:11.880
+ and mapping of the environment,
+
+33:11.880 --> 33:14.360
+ being able to do some basic communication
+
+33:14.360 --> 33:16.560
+ to say go clean the kitchen.
+
+33:16.560 --> 33:21.360
+ Do you see in your maybe more bored moments,
+
+33:22.880 --> 33:24.840
+ once you get the beer,
+
+33:24.840 --> 33:27.000
+ just sit back with that beer
+
+33:27.000 --> 33:30.800
+ and have a chat on a Friday night with the Roomba
+
+33:30.800 --> 33:32.840
+ about how your day went.
+
+33:34.120 --> 33:37.560
+ So through your latter question, absolutely.
+
+33:38.640 --> 33:40.720
+ To your former question as to whether Roomba
+
+33:40.720 --> 33:43.680
+ can have human level intelligence, not in my lifetime.
+
+33:45.200 --> 33:48.440
+ You can have you, you can have a great conversation,
+
+33:49.680 --> 33:51.960
+ a meaningful conversation with a robot
+
+33:53.920 --> 33:55.400
+ without it having anything
+
+33:55.400 --> 33:57.760
+ that resembles human level intelligence.
+
+33:57.760 --> 34:02.760
+ And I think that as long as you realize
+
+34:02.800 --> 34:05.280
+ that conversation is not about the robot
+
+34:06.360 --> 34:08.600
+ and making the robot feel good.
+
+34:08.600 --> 34:13.600
+ That conversation is about you learning interesting things
+
+34:14.880 --> 34:18.400
+ that make you feel like the conversation
+
+34:18.400 --> 34:21.240
+ that you had with the robot is
+
+34:23.800 --> 34:27.280
+ a pretty awesome way of learning something.
+
+34:27.280 --> 34:30.800
+ And it could be about what kind of day your pet had.
+
+34:30.800 --> 34:35.040
+ It could be about, how can I make my home
+
+34:35.040 --> 34:36.160
+ more energy efficient?
+
+34:36.160 --> 34:39.600
+ It could be about, if I'm thinking about
+
+34:39.600 --> 34:41.720
+ climbing Mount Everest, what should I know?
+
+34:44.320 --> 34:46.480
+ And that's a very doable thing.
+
+34:48.600 --> 34:51.480
+ But if I think that that conversation
+
+34:51.480 --> 34:53.640
+ I'm gonna have with the robot is,
+
+34:53.640 --> 34:56.760
+ I'm gonna be rewarded by making the robot happy.
+
+34:56.760 --> 34:58.720
+ But I could have just put a button on the robot
+
+34:58.720 --> 35:00.280
+ that you could push and the robot would smile
+
+35:00.280 --> 35:02.120
+ and that sort of thing.
+
+35:02.120 --> 35:04.160
+ So I think you need to think about the question
+
+35:04.160 --> 35:06.680
+ in the right way.
+
+35:06.680 --> 35:11.520
+ And robots can be awesomely effective
+
+35:11.520 --> 35:14.400
+ at helping people feel less isolated,
+
+35:14.400 --> 35:17.520
+ learn more about the home that they live in
+
+35:17.520 --> 35:21.920
+ and fill some of those lonely gaps
+
+35:21.920 --> 35:23.760
+ that we wish we were engaged
+
+35:23.760 --> 35:25.640
+ learning cool stuff about our world.
+
+35:25.640 --> 35:30.640
+ Last question, if you could hang out for a day
+
+35:30.640 --> 35:34.640
+ with a robot from science fiction, movies, books
+
+35:34.640 --> 35:39.640
+ and safely pick, safely pick its brain for that day,
+
+35:39.640 --> 35:41.640
+ who would you pick?
+
+35:41.640 --> 35:42.640
+ Data.
+
+35:42.640 --> 35:43.640
+ Data.
+
+35:43.640 --> 35:44.640
+ From Star Trek.
+
+35:44.640 --> 35:48.640
+ I think that data is really smart.
+
+35:48.640 --> 35:52.640
+ Data's been through a lot trying to go and save the galaxy
+
+35:52.640 --> 35:57.640
+ and I'm really interested actually in emotion and robotics.
+
+35:58.640 --> 36:00.640
+ And I think you'd have a lot to say about that
+
+36:00.640 --> 36:05.640
+ because I believe actually that emotion plays
+
+36:05.640 --> 36:10.640
+ an incredibly useful role in doing reasonable things
+
+36:11.640 --> 36:14.640
+ in situations where we have imperfect understanding
+
+36:14.640 --> 36:15.640
+ of what's going on.
+
+36:15.640 --> 36:18.640
+ In social situations when there's imperfect information.
+
+36:18.640 --> 36:23.640
+ In social situations also in competitive
+
+36:23.640 --> 36:25.640
+ or dangerous situations,
+
+36:26.640 --> 36:30.640
+ that we have emotion for a reason.
+
+36:30.640 --> 36:35.640
+ And so that ultimately, my theory is that as robots
+
+36:35.640 --> 36:36.640
+ get smarter and smarter,
+
+36:36.640 --> 36:38.640
+ they're actually going to get more emotional
+
+36:38.640 --> 36:46.640
+ because you can't actually survive on pure logic.
+
+36:46.640 --> 36:51.640
+ Because only a very tiny fraction of the situations
+
+36:51.640 --> 36:55.640
+ we find ourselves in can be resolved reasonably with logic.
+
+36:55.640 --> 36:57.640
+ And so I think data would have a lot to say about that
+
+36:57.640 --> 36:59.640
+ and so I could find out whether he agrees.
+
+36:59.640 --> 37:02.640
+ What, if you could ask data one question
+
+37:02.640 --> 37:05.640
+ and you would get a deep, honest answer to,
+
+37:05.640 --> 37:06.640
+ what would you ask?
+
+37:06.640 --> 37:08.640
+ What's Captain Picard really like?
+
+37:08.640 --> 37:12.640
+ Okay, I think that's the perfect way to end the call
+
+37:12.640 --> 37:14.640
+ and thank you so much for talking today.
+
+37:14.640 --> 37:16.640
+ I really appreciate it.
+
+37:16.640 --> 37:45.640
+ My pleasure.
+