vtt/episode_024_small.vtt

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.