example-transcript.txt

Ray Le Maistre, TelecomTV(00: 00: 10):
Welcome back to the Proceedings. It's time now for our second session of this year's DSP Leaders Well forum. The session that we've got upcoming is enabling the autonomous network with ai. Now, last year's AI native session kicked off our extensive coverage of AI applications within the telecoms network. Investment is well underway across multiple AI use cases, but as demand on the network increases, where should telcos be focusing their efforts and resources to accelerate the value of AI and what additional challenges will they face in their journey towards the fully autonomous network? We've got a fantastic panel lined up for you today to discuss these topics, but we're going to kick off this session with a progress report conversation. Now. We're going to be having a number of these progress report chats during the course of these two days. Many of them with the session co-hosts from last year to get us up to speed on what's been happening in the past 12 months. I'm delighted to say that joining us right now to give us a progress report on Telco AR is Ahmed Hafe, VP of Technology Strategy from Deutsche Telecom. Ahmed, if you'd like to join me on stage,

(00: 01: 45):
Ahmed,

(00: 01: 45):
Great to see you. Thank you. So Ahmed, welcome back to the DSP Leaders World Forum stage.

Ahmed Hafez, Deutsche Telekom(00: 01: 53):
I need to put the mic back because they were taking a picture. They told me

Ray Le Maistre, TelecomTV(00: 01: 58):
That's it. It's well in

Ahmed Hafez, Deutsche Telekom(00: 01: 59):
Place. I'm really delighted to be here. So I really have an appreciation for this event. So it's very cozy. The faces are very familiar and it feels very comfortable to be amongst you.

Ray Le Maistre, TelecomTV(00: 02: 10):
Okay, excellent. Now you were here last year of course as the co-host for this session that last year was titled Creating a Framework for the AI Native Telco and that was an incredibly popular session and really well received at the time and afterwards as well on demand. Now they say that a week is a long time in politics and the same can be said for the AI sector. And as a result, since you were here last, a great deal has happened in terms of AI and telecom. What for you have been the key trends in the telecom sector during the past few in the last year with regards to ai?

Ahmed Hafez, Deutsche Telekom(00: 02: 54):
Right, so that has been a number of developments. I will, with the byproducts of what happened. I mean generative AI have took us all with surprise. Last year was the hype, I would say it was on the peak of the hype. Now it's actually going down a little bit. So getting more to realities and expectations and right expectations. But what happened is that this hype created a huge momentum not only for generative AI but for AI itself. So for any traditional ai, predictive ai. So you see companies now have a lot of focus on ai. So suddenly AI become a thing. I mean we've been working on AI for many years, so this momentum helped us significantly. So actually the accelerated things that are even not really generative ai. The other thing that happened also that we started going into a generative ai, getting our hands dirty with it if it's correct to say, to understand what are its strengths and weaknesses, and we started to realize that maybe the expectations were inflated a little bit, at least at this stage of development.

(00:03:58):
And there are things that it can fit in and things that we have to be cautious on. To give you a precise example, we are very comfortable to use predictive AI today in closed loop automation, but it's not the same when it comes to generative ai. We cannot see the precision and accuracy coming out generative AI that would actually give us this confidence of having closed loop automation yet. And I'm saying yet because developments are happening as we speak, but there are many things that we thought could be very challenging became, it's not that challenging things are developing. So we give you some examples. Computers improving. I mean we discussed compute last time and there has been a bit of, I think I was showing a slide where 500 more times of demand on compute. I think you have seen the surgeon compute. Unfortunately, I did realize myself that I should buy shares in Nvidia, but that was too late.

(00:04:50):
But in any case, this is happening and it's still a problem. So the other part was actually we expected also the foundation models would get improved and they got significantly improved. But also we see now small language models and other models and other techniques even for training and fine tuning that would enhance the whole story of the needs of a compute and how we actually employ the technology. So yeah, things have been to a large extent clarified, but we figured out other problems. So as you go you develop the knowledge and of course we had the AI act meanwhile, which was an important step. I mean I think Europe, one of the leading areas is regulations, so it's good. It's good to say that we are leading in some aspects in Europe, but these were essentially the most important changes that we have seen last year.

Ray Le Maistre, TelecomTV (00:05:39):
Okay, so is it possible for the telecom sector, I mean we heard just a little bit earlier in the panel about innovation that sometimes the pace of things doesn't keep up with expectations. Can the telecom sector keep pace with AI developments? Is that in any way possible?

Ahmed Hafez, Deutsche Telekom (00:06:00):
It's a good question, but let me maybe take a step back and talk about technology development in general. Then I come to ai, there is a term now called digital dexterity, which is actually describing the gap between the huge transformation in technology and the level of absorption we can absorb and actually it's happening everywhere. So yes, the pace of technology is much faster than what we can absorb and what we can take and what we can implement. But this will continue to be the case. Maybe this curves goes down a little bit because if there are no adoption, there will be no further innovation, so it'll slow down. But nevertheless, coming to ai, the point is with AI depends on the role you want to play. If you want to play as an AI provider, then you want to create models for industries, do things like that, train models and take that, then you would have to keep up with that pace.

(00:06:52):
Then the challenge is much higher. If you decide, no, I will actually use that. These technologies for the use cases I would employ in my network or for my customers or my customer service, BSS and so on, your level of pace would change because you would rely on partnering and you would rely maybe on buy more than make. So this would balance. So I think we should consider, I mean of course every operator is different, but you can consider a mix between the make versus buy in order to make sure that you can sustain the pace because we are in the heart of the changes happening there, but as we go, things will actually stabilize more and more and then you can take decisions or make much easier. But today maybe it's a little bit immature unless you go into you have huge business, you see an outlook of the great opportunities business wise and you decide that you want to hunt for that, then it'll be a different game.

Ray Le Maistre, TelecomTV (00:07:47):
We also heard earlier on in that innovation session about the importance of collaboration and I guess this is an area where it's absolutely paramount. What kind of pressure does the pace of development and the requirements around AI put on companies like Deutsche Telecom when everybody else in the telecom sector?

Ahmed Hafez, Deutsche Telekom (00:08:08):
Okay, so let me take let's say four layers of, so to structure a little bit the answer because it's a very big challenge. If you take the data layer on top of it, the infrastructure, then we have the foundational models. I will focus a little bit on gen AI for now and then the tools and the AI platforms. If I start from data, we are already facing a big challenge that even though we are working with standards, the packet gateways generate data for every supplier in a different manner. Every packet gateway or every function in the network core ran, you name it, have roughly about 10 different types of files of streams that are generated per function. I mean how many functions we have in the networks and how many replications of functions we have. And the very challenging part is that the data models are different.

(00:08:53):
Even the definition of the attribute value pairs are different. And then if you want to take a use case, take anomaly detection, you say, okay, we can do anomaly detection, basic use case easy can easily be done by generative AI or predictive ai. And then you start, you do it in one operator and you are group, you have 11 operators and you want to roll that out. You would figure out if there are different supplies, you have to go again and again and repeat big part of this chain again, it's not easily copyable. So if we want to go to scale, we need certain level of commonality, at least on the data layer. We need to make sure that we have this data definitions properly done. And it's something that of course suppliers can help a lot there because we can align on at least common baseline.

(00:09:37):
Then you go to infrastructure and then looking at infrastructure, you have public clouds of course, and hyperscalers and also the on-prem part, and I believe we'll need a hybrid of that and hybrid of that and where you put the data and then how you access the data. Then portability of data and then the cost of moving data. So this is also hindering this kind of cooperation and openness. So this is something that needs to be solved. You go layer above on the foundational models and you find out either you have open source when you have some freedom, but it's not as strong as the proprietary ones. But if you look at the proprietary ones, they're not portable, so you have to use them on a specific platform. You can't take them on-prem, very, very minor exceptions, but mostly you're stuck with them where they are and then you need to deal with them.

(00:10:23):
But what about confidential information? What solution can I do to deal with confidential information that I don't want to put in public cloud, for example? So this is another issue to be resolved. If we go to the tools layer, and this is a very interesting one because I mean suppliers are here, and I talk also to many suppliers about AI and every supply come with a very nice use case, fantastic use case, but comes with the tools and everything down to, not to the infrastructure but including the models. And I come say, oh, this is a solution we have, but I cannot have 100 set of different tools for every use case. If I have some tools for sun and another set of tools for energy efficiency, I mean this has to be one. I mean they cannot be fragmented and this is another challenge. So we need some kind of interoperability and an understanding that you can plug into an API platform, sorry, an AI platform, and you can still work with different use cases and implement your software. Okay,

Ray Le Maistre, TelecomTV (00:11:23):
Very quickly, because we're going to move on to the panel, but a quick word on skills. What have you seen in terms of the developments around skills requirements in the past year?

Ahmed Hafez, Deutsche Telekom (00:11:33):
Right, we have been running huge upskilling programs, but the point that I think, I don't know if we're all missing but probably we're missing there is the skills on the what. So learning new things, learning how to deal with ai, how to manage ai, but there is a skill on the how do you approach it? The part because we often miss the spot because people also have some resistance to change. People are used to do things, our employees are used to do things in certain manner. You want to tell them that this will change. And eventually in order for us to read the value of ai, we should not be using AI for very, very narrow views. So today we have a process. The process consists of 10 steps. You take step three, four and six, and then you put ai. But we need to rethink the entire process by ai. This requires that we think in a different way. So this is a behavioral element. It's not about scaling on the what, it's scaling on the how. So you think from a different perspective and this is something that we need to work more on.

Ray Le Maistre, TelecomTV (00:12:34):
Okay, excellent. So Ahmed, it's great that you're going to be staying with us for the panel and for the panel you are going to be sitting. Yes,

Ahmed Hafez, Deutsche Telekom (00:12:44):
I'm not going to Absolutely

Ray Le Maistre, TelecomTV (00:12:45):
There. You're staying right

Ahmed Hafez, Deutsche Telekom (00:12:46):
There. He told me multiple times I should not move.

Ray Le Maistre, TelecomTV (00:12:49):
So thanks very much for joining us, giving us an update for what you've seen in the past here. So a round of applause please for Ahmed. Thank you very much. So while we transition from the interview to the panel, we're going to play our next highlights video during which our panel speakers are going to join us here on stage. This highlights video appropriately is from our AI native telco summit with some footage also from our December event telcos and the AI event.

Juan Manuel Caro, Telefonica (00:13:30):
Now, we in tele for example, have more than a hundred use cases with AI that we use across the life cycle of the network and across our operations from the beginning of the process of assurance to the end. For example, even before the process begins or even before you monitor the network, we use it to do predictive maintenance. For example, as we are getting all this data and trying to predict faults before they even happen,

Mark Henry, BT (00:13:56):
We're starting to develop a business case for centralized ran Cloudified brand. If we were to put some benefits in that, we would gain through the application AI and machine learning from that platform, we're going to have to do a really good job on characterizing the cost benefit analysis

Akira Tada, SoftBank (00:14:13):
In that at software bank we are investing massively to do generative ai. So we are building our own data center optimize for ai, and also we are bringing more resources to develop our own A RM. I can tell the details so far, but we are making much investment to optimize and also to utilize our data more efficiently for telecommunication business.

Enrique Blanco, Telefonica (00:14:43):
The point is can I talk with my network? Can I send gene AI and tell him for instance, at this moment, which are the number of cells that they're saturated in Barcelona and around the ferra? So which is the behavior of the network, what I need to do trying to improve it. And we are using gene AI technology trying to manage the network in this way.

Abdu Mudesir, Deutsche Telekom (00:15:06):
The way we view artificial intelligence for us is as a continuum of the predictive artificial intelligence we've been doing the last years and the generative, which is coming, which is really bringing a lot of UX capability and gives us the chance to even go to the next level.

Ray Le Maistre, TelecomTV (00:15:31):
Okay, great. It's time to get stuck into our panel enabling the autonomous network with ai. As guy mentioned earlier, each session has its own poll. One question, three answer choices. You can just pick one here is the question that we've got for this session. When will AI enable telco networks to be truly zero touch without any human intervention? So please go ahead and vote on the telecom t TV website. You'll find all of the polls in the DSP leaders world forum section on the agenda page, and we'll take a look at how the voting has gone at the end of this session. So at this point I want to introduce our co-host for this session. Delighted to be joined here by Amal Fad K. He is the EVP and group Chief Technology officer at the Tylenol Group. Amal, you've been to DSP leaders before. Thanks very much for returning as a co-host. Great to see you here.

Amol Phadke, Telenor (00:16:32):
Thank you, Ray. Thank you for having me. Hello everybody. Yes, I'm back wearing a different T-shirt, but yes, I am really looking forward to this panel. Let's meet the panelists first.

Ray Le Maistre, TelecomTV (00:16:45):
Oh yeah, absolutely. Yeah, you're going to give your opening address in just a couple of minutes, but first let's find out who is on our panel. So if we can start at the far end and get our panelists to introduce themselves.

Juan Manuel Caro, Telefonica (00:16:57):
Thank you. Hello everybody. This is my first time here, so thank you for having me. I'm Juan Manuel Carro. I'm director of autonomous network at Telefonica Group.

Nik Willets, TM Forum (00:17:05):
Hello, great to be here. I'm Nick Willets. I'm the CEO of the TM Forum.

Manish Singh, Dell Technologies (00:17:10):
Hello Manishh, CTO, Dell Technologies.

Ahmed Hafez, Deutsche Telekom (00:17:14):
Great to be back here. So I won't give you

Mabel Pous-Fenollar, Vodafone (00:17:19):
Hi. Hello everyone. Nice to be here my first time. So I'm pose in Vodafone and I head the digital and zero touch operations.

Madhukiran Medithe, Rakuten Mobile (00:17:28):
Hello, this is du. I'm chief lead officer for the Rakuten Mobile and this is first time for me and delighted, excited for this panel.

Ray Le Maistre, TelecomTV (00:17:35):
Okay, great. Thank you very much everybody for joining us. So Amal, without further ado, if you'd like to make your way up to our lectin for your DSP leaders address. Thank you very much.

Amol Phadke, Telenor (00:17:46):
Thank you Ray. So as I was saying, I asked Ray during the break if I can talk about something interesting. The topic I had with mine was general elections in the US general elections in the uk and then if you had time general elections in India. But Ray sort of vetoed me and said, no, no, we have a much more interesting topic of ai. So I'm going to talk about that today sir. But first of all, great to see all of you. As Ahmed said in the beginning of our panel today, AI continues to drive a huge amount of news in the media. In fact, if there is one technology that I see a lot more over and over again every week, every day, that's ai. And it's obvious why that is because all of the analysts are sort of unequivocal in their view of the trillions of dollars of benefits that AI is supposed to sort of offer us.

(00:18:44):
But at the same time, we also have to be mindful of the fact that a lot of that is also hype. And I was looking at the previous panel and you had this interesting poll Ray where we sort of were distinguishing between where innovation comes from and majority of that innovation was coming from our vendor partners. And I must also say that a lot of this hype has also been generated by our vendor partners and actually when I was one of those vendor partners, we call it business development. And I think we need to just also be conscious of that a little bit because obviously a lot more investment has gone into building the compute infrastructure and building the hardware layers and the LLMs on top and the cloud platforms that support all of that. So it's natural that there'll be a lot of interest that all of those things get used.

(00:19:33):
Otherwise we are going to have a much bigger problem on return, on investment on all of that billions of dollars of investments that's happening. So it's worth keeping that in mind as we look at this problem and how we actually benefit from AI has a lot to do with how we actually work with the ecosystem. So at Telenor, the way we sort of looked at it was we looked at our entire span of operations in all of our markets and we came to the conclusion that there's a huge amount of use cases that a telco could look at, but in order for us to sort of attack all of that immediately would not be feasible. And so when you look at where the biggest benefit is about 80% of the benefit that we see in Telenor comes in four areas, customer service, marketing, sales network and autonomous network and operations, which is a topic of the panel today.

(00:20:22):
And then product development, IT development, software development. So these are the four areas that Telenor is planning to sort of address. And we have already started, we have about 250 use cases that we as Telenor have already started to pilot. Again, going back to the comment you made at the beginning, I would say 95% of them are AI and 5% of them are gen ai. But we of course call all of it gen AI because we want to support the hype that's happening. I would also say that within network itself, I don't think we have really gone into the depth yet of exactly which areas would benefit the most. There are obviously four areas and we will talk about it in our panel. There is the whole design and building of the networks. Then there is of course a provisioning of the network and then how do you assure those networks?

(00:21:08):
And then finally, how do you sort of capacity plan and optimize those networks. So each of those big areas requires us to sort really concretely look at how AI can truly give us a benefit. But then we also came up with a program inside Telenor because it could be network, but it could be customer service and regardless of which area we focus into for us to really benefit and not get caught up in the hype, there are five dimensions that often get overlooked because we get obsessed on delivering use cases. But there are these five other things that I'm quite passionate about and we are driving that in Telenor. First is the people dimension. There is still a huge knowledge gap inside the industry on people actually getting trained on ai. At Telenor we have 12,000 people and I would, 70 10% of them are actually AI aware today.

(00:21:59):
So there's no point us building thousands of use cases if none of our Telenor employees can actually use it. So that would be something to just deep dive on in the panel. The second big important dimension is the ecosystem partnerships, Telenor and all other large colleagues of mine in the telco audience today, I would have a question which I ask myself every day, how would you do an RFP on ai? I struggle to understand how you would actually do a commercial negotiation because it's a completely different ecosystem play. You really need to build an ecosystem with five or six constituents to build a solution. And that's a very different mindset. So ecosystem is another area to look at. Third important dimension is responsible ai. You mentioned the AI act. When it starts coming in, it'll be good because it's going to get us discipline, it'll be bad because they're going to come up with a stick and we'll need to have that kind of tools to sort of show that we are compliant to that.

(00:22:55):
So responsible AI is about being bold but being thoughtful at the same time. And I think that would be another area to just tease into the panel. And then the final two things are, one is really around building that data, that underlying data to avoid the whole garbage and garbage out capability that might happen with ai. So that underlying data platform, we are very passionate in Telenor, unless you get that right, a lot of the benefits you want in the network space is simply not going to come. And then the final point is the thought leadership piece. I go back to what I said at the beginning as telcos and as this community, we need to start to celebrate some of the successes. Yes, there's a lot of hype, yes, gen AI is still immature and not as accurate as we would like it to be and deterministic AI is actually much better. But let's show some thought leadership even on that. I think if we can get these five dimensions working well, then the use cases start to benefit in the areas that we believe there will be a lot of benefit. So that's all I wanted to just set the scene for you Ray. And of course the two of us are now going to look at the panel to give us a little bit more wisdom. Thank you all for listening.

Ray Le Maistre, TelecomTV (00:23:59):
I'm all thank you very much. So obviously we can come to the elections during the lunchtime. Yes, I'm sure plenty of opinions on that as well. But I'd like to start off by giving our panelists the opportunity just to pick up on anything that you mentioned in your opening speech. Has anybody, Ahmed, you've got a question

Ahmed Hafez, Deutsche Telekom (00:24:24):
I like very much the points that you raised. I'll just maybe take one of them, otherwise we will consume all the stuff. So it's the point on where is the network, is AI more valuable or would have more impact? So what we did in Deutsche Telecom, we looked into lifecycle of network. So from plan design, build runs or operate cross-functional and security. And we also discussed with a lot of partners and we figured out that number one by far is the operational part and next to that is the planning and security. So even though security as such, security will become a huge topic and I think we discussed that today. The other part, there are other parts like for example, design, which we felt that it's only an assistant because luckily the level of intelligence needed in design is not yet, AI is not there yet. Happily that we still have some jobs for some time. So design is still not a contended area, but this is how we've seen it. If this

Amol Phadke, Telenor (00:25:24):
Answer, yeah, no, no, that makes total sense. And operation is the most obvious because it's the most suited for ai. There's a huge amount of tele data we get from our networks and AI is very well built in order to sort of process that very quickly and give us insights to be proactive and predictive. So totally agree with you.

Madhukiran Medithe, Rakuten Mobile (00:25:43):
Yep. Adding to that, to what Hamed said, I just like to add my experience when we started rollout telecom in Japan, Rakuten Mobile. So we started in 2019 and we know that we are the first in the industry applying the solutions in the cloud native functions and virtualized functions. So definitely we need to go with the ai. We are a greenfield operator, we don't have the data, but we start capturing the data from the day when itself and from the probes from the milliseconds to the aggregated minutes. So that's how we started. I mean we follow that. AI is our DNA, Rakuten mobile is one of the group companies in the Rakuten ecosystem. So we have the ecosystem data, 70 plus groups are there in the Rakuten mobile including e-commerce, FinTech. And what I'm trying to say is we have a lot of subscriber data it.

(00:26:29):
So adding to that, we have data which we started day one with capturing it on the top. We built our AI models as talking about planning because that's main thing because we went in the domestic roaming initially. So we know where the is happening, where the data, where the users are. And we have done according to that our operations and building the air models on the top. Definitely AI plays a crucial role at this kind of era. What I can see that because of the data is more viable today. That's what we have used and that's how we have built our network. The span of three years, we have rolled out more than three 50 K cells across Pan Japan covering around 99.2% of population, which is a massive achievement for us.

Ray Le Maistre, TelecomTV (00:27:11):
So I'm sure that the collection of data, obviously records and mobile started in a different place. You are starting at a point where you've already got an awful lot of data to deal. We've

Amol Phadke, Telenor (00:27:23):
Got about 180 years worth of data. We are in a slightly different place, but yes, I agree with you in terms of fragmentation and how we create a consolidation would be key.

Ray Le Maistre, TelecomTV (00:27:33):
Okay. Yeah. Nick

Nik Willets, TM Forum (00:27:35):
Just wanted to build on the use case point that Ahed and Amal both made and it's pretty consistent with what we see. We've surveyed in the TM forum over 20 different operators and where they're focusing their automation efforts, pretty consistent picture. The question I think we need to keep asking ourselves is not to let this just become a cost reduction exercise. We see a lot of the emphasis on solving the challenges of retirement Cliff solving the around 30% of opex that goes into network operations, cost rightful targets, important things to do. We're all concerned about free cashflow, but at the same time there's not enough conversation in our view on the customer experience impact. And to do that, actually looking at individual use cases or use case areas, kind of rat holes you in an area, it may be the right place to start an experimentation, but if you're thinking about how our customers experience us, quite often a network fault requires multiple different teams to get engaged from frontline support all the way through to a deep specialist.

(00:28:34):
And there's ways to leverage AI in those use cases to problem solve across and join up what would typically be an email chain internally, these kind of faults. There's also a link to the hot topic at the moment, which is network monetization, which is that in order to stand up some of the open gateway kamara use cases, the TM forum is very close to we need to automate elements of the network. So I think when we're talking about use cases, we always have to bear in mind what's the business driver. If it's just cost reduction, it's going to push us down a path. We'll probably be pretty happy with the results, but you can't cut your way to growth,

Ray Le Maistre, TelecomTV (00:29:07):
Hence the large number of use cases that you were able to identify. So chatting to Ahmed before there, we were looking back at the past year, a lot has changed, but I just want to get a sense from the panel now. Are we more or less confused about the role that AI will play in telecom compared with a year ago? Because the hype circle might be on the way down, but it's still enormous. So Juan Mal, if we can start with you, are things clearer for you now than they were a year ago in terms of the role of ai?

Juan Manuel Caro, Telefonica (00:29:46):
Yeah, I think here would be on the optimistic side, no, I think we are less confused, although maybe tomorrow something appears and we confuse us again. But I would say we started a few years ago with ai, with the traditional ai, with the predictive things, trying to understand that, applying them to use cases, looking that it really works, it can really help. We set up strategy and we were working with that and I think a year and a half ago chat, TPT came and gen ai, all these kind of things. And then I think that was a bit of a confusion. I think everybody that was to save the world, we could do everything with that or not or whatever. And it took us some time to figure out that covid is an amazing thing, have a lot of possibilities, but it's not going to replace the traditional air for a lot of use cases.

(00:30:32):
It is another world. And now I think we have had the time, or at least in Telefonica to settle up this hype to really understand that technology with talking with all the vendors and all the market and put that where it really adds more value and setting again the strategy. And I think we now, at least in Telefonica, I think we are now in the execution mode. We have settled this up again, I think this goes as fast at least Myer was super surprised of JG, BT and leading the AI in Telefonica in the networks for four one. And even me, when I saw that and I started playing with that, I said, this is really amazing. I wasn't expecting this right now. So maybe tomorrow it's another thing and we get a bit confused. But I would say right now we understand better traditional AI and AI and still a lot of challenges to do that now and to execute. And we'll talk about the product here, but I would say I'm more confident we know where we're sitting and what we'll have to do. Not easy to do it.

Ray Le Maistre, TelecomTV (00:31:28):
Okay, Manish, we'll come to you and then Mabel, but Manish, how do you feel compared to a year ago?

Manish Singh, Dell Technologies (00:31:33):
So I think Amol accused us as vendors that we create the hype

Juan Manuel Caro, Telefonica (00:31:41):
That's job

Manish Singh, Dell Technologies (00:31:42):
Taking that I think my view is as an industry, I mean if you just look a little bit back in the rear view, right? I mean we've talked so much about the OTD cycle coming and eating the lunch, we've had the cloud cycle which came and went and the telcos in many ways struggled to catch up with these cycles. In certain cases, one could say even monetization opportunities were missed here. Now we again find ourselves at a very interesting crossroad where the tech is moving and the tech's not going to slow down because the demand is enormous and it's not dependent just on the telco verticals. So let me also say this, there are other industry verticals that are already making huge demands on AI and just as a vendor, as a supplier, I can tell you to lay our hands on the GPUs today is a challenge, right?

(00:32:42):
Just the demand is outpacing supply. It's a very rare situation you get to see in the industry. So we are at this juncture, the question now becomes is what do the telcos do? I tend to break this down into three big buckets. Number one, AI in the network. How are you going to use AI in the network and are there opportunities to create differentiated networks? And we had this discussion earlier around innovation where you go innovate, et cetera. And this is an area where the service providers at telcos should think, are there areas where they can create differentiated networks? We have had the networks where we're differentiated on coverage and capacity and what differentiation can you create. So AI in the network. Second thing is what I would call is AI on the network where what can you enable with AI as a set of technologies on your network, including things that Nick touched on, even if you go in the realm of billing care, customer care, et cetera, where generative AI starts to play a very key role.

(00:33:48):
And then the third bucket, which I think is the most obvious one, at least here and now is what we call as AI factories. And we've seen that in certain markets where the telcos and for that matter governments are also taking note of this because at the end of the day, what's an AI factory? Its data in intelligence out. And the question is, can any society, can any country wait to outsource that intelligence and what do you need to do to create that intelligence? And I'm not talking telco right now, I'm saying broadly across the ecosystems and there is a demand for AI capabilities to be developed in many sovereign markets. So we are also seeing in certain areas, especially I would say partly in Europe, partly in Asia Pacific, where we are seeing some of the telcos who are already moving to create these AI factories. And so if I sum all of this up, I think there are opportunities in the network, on the network and even from a monetization perspective. And now I don't think there is one cookie cutter approach for every telco. Each one needs to think through, work on their strategies and see where do they want to play, do they want to monetize, do they want to work on the opex customer care, et cetera, et cetera.

Ray Le Maistre, TelecomTV (00:35:00):
Yeah, it's nice to carve it up like that and to get a perspective. Mabel a Vodafone, is there a clearer picture or does it seem even more overwhelming than a year ago about the

Mabel Pous-Fenollar, Vodafone (00:35:15):
I agree with the co-panelists. So it's clearer now. I'll touch on what Nick said. It is super essential and I often we're very, very focused on what is the outcome that you want to achieve with that. So it needs to be AI and engine. AI is yet another technology, great technologies, but we need to look at the outcome. For us, customer simplicity are the two ones that are closer to us. We're looking at growth, it's not as close, but we are exploring some use cases on that as well. So it is clearer. It is clearer the technology as well as the opportunities and the risk it brings, right? So we see a lot of the partners that went really, really big on this, the risk on cost because the cpu. So we really believe that it's clearly both the opportunities and the things that we need to watch out as an industry to get it right. So definitely think it's clear air and we use that as one of the many other technologies that we can do because for autonomous networks is AI is gen ai, but also a lot of the autonomous technologies like RPAs, VPMs and so on and so forth.

Ray Le Maistre, TelecomTV (00:36:30):
So it's not just about ai, it's about adding AI to the

Mabel Pous-Fenollar, Vodafone (00:36:33):
Mix of already there. I would say yeah,

Madhukiran Medithe, Rakuten Mobile (00:36:38):
Lemme add a few more points and mention the same, but in my opinion, basically there is no confusion. We know how we have to use Jena and stuff, but the point is where exactly it's going to be used. That's the main picture. What I can see that. So we are all coming from the data science background. We have GPUs already secured in our network. We are running petabytes of data every day. So the point, what we're trying to figure it out is some of the models which we run, maybe the forecasting models, the prediction models or anomalies which are constructed to telecom sector, which are not available in LLMs. So I have to do my own intent learning there, meaning I have to invest my time, my resources over there to train the model in the LLM as well. So it's a trade off what exactly you want to do with the general, there is a lot of hype pressure on data scientists.

(00:37:28):
Basically everybody wants to use AI stuff. But some of these things like sentiment analysis, like NLP things and conation models, which image processing things, those are readily available, which we can take directly from the generic prospect. But the contents which are related only telecom sector side where you have own defined formulas, own defined KPIs, which are not available in the lms, that's where you have to invest your trade off. Whether I have to go with the G AI or I have to go with the traditional AI going with my deep learning techniques. So it's not the resource cost, it's the time efforts which you need to spend across. So that's what we're still looking into that we are still the standard has to evolve into more into that how we can use gene open brand is one of the segment which actually evolving towards that.

(00:38:15):
Some of the cases like so is standard defined if you go to that spec. So it's been different already. So those are all, we can directly go with the lms, but how we implement in the network, because dynamics is keep changing the rfs, keep changing the subscriber profiles, keep changing. That cannot be done to the L lm. And one more challenge, what I can see is the governance, the security, how much amount of data I have to send to if I go with third party LLMs or you have to use any cloud source LLMs. So that comes into the security about the compliance prospect, every organization, the governance side. So these are all the things which we need to be closely watched. And then definitely we can use Gene and I don't think so there is a confusion but the point is how to use is a question.

Ray Le Maistre, TelecomTV (00:39:01):
Yeah. And Amal, you kind of hit the ground running at Telenor, didn't you? Joined towards the end of last year, right? When that wave was?

Amol Phadke, Telenor (00:39:10):
I think I agree with everything the panel has said. I would add one thing, we as a telco group are very good at building something and then launching it and then keeping it stable. This is not going to be stable.

Madhukiran Medithe, Rakuten Mobile (00:39:27):
Exactly. This

Amol Phadke, Telenor (00:39:27):
Is going to change every week, every month. So how do you then use, as Madu said, the models that we know will not be mature, that will change over time. And that would be the big challenge for us. It's not the clarity, we have clarity, but I think the issue is going to be how do you launch a AI based service to what money said and start building a factory and then the underlying foundation changes because it's obsolete and now I have a new model and a new infrastructure, how do I create that delaying that is a big challenge.

Nik Willets, TM Forum (00:39:57):
And can I jump in on that because I think your question was are we more or less confused? I think a panel and a room of technology experts, I'm not sure we were ever super confused. We were sort of questioning where it could go. I think where there is confusion is at the top of the house and we see a massive difference right now between CEOs and boards who get the potential of AI and are removing all the roadblocks internally and are seeing top line and bottom line impact from AI who are unfortunately in the minority and a majority who are still a bit skeptical. And as we start to go, and I think you're absolutely right Ray, we're just tipping over the top of the hype curve. That'd be how we'd position it. It's a hell of a hype curve, but it's going to be a hell of a downward run on the ski slope.

(00:40:40):
Those boards may just get bored and move on to the next thing and miss the profound nature of what we're talking about. So when you're talking about governance, those groups need to be bought in. And to what Amal just said of this is a moving template, we are not used to governing business in this way. It was said in the previous panel, by the time you've deployed a technology and telcos move forward two or three generations. So unless we bring the top of the house on that journey and understand the fundamental operating model impact, I think we'll struggle no matter how clear we are at a technology level.

Ray Le Maistre, TelecomTV (00:41:09):
Yeah, no that's a great point. And I guess the thing that needs to be done is that you need to get on and show some concrete steps to the higher level the people at the top of the stack. So we talked last year about being AI native, but the industry is a long way from being that. So maybe we can look now and have a quick chat about what are some of the concrete first steps in relation to automating network operations. What are the things that can be proven and shown to be working? So Nick, if we can come to you first, I know you've got a couple of examples, but I think everybody's going to have something to say here.

Nik Willets, TM Forum (00:41:54):
So I think first of all, it's come up I think in the panel already sets and understanding, particularly when we talk about the network could be a challenge. You've often got individuals who spent their whole career, so you're talking 20 or 30 years in the network space. And when we make this a cost centric argument, we see a lot of resistance from those teams of what are you telling me? My 20 or 30 years of expertise can now be automated. So you sort of start the conversation on the wrong footing in many cases. But we do see fantastic progress where you can actually bring the team on that journey. And we're starting to see a pattern of the companies who've started that journey earlier. So this isn't actually really about what you've done in the last 12 months reacting to gen ai. We think the roots of it, and I think several of the companies on the stage have experienced this started in open and good data governance programs that started probably five or six years ago.

(00:42:45):
So the business was getting used to the privacy, security, regulatory implications of opening that up, be able to train the models with the right data, the right data being available. We also see a trend between that top of the house support and practical impact. And you're right that you need to actually bring that into real examples, show the goods. The challenge can be that can be a bit of a vicious cycle. So if you look industry-wide as we have the privilege of doing, you can see very meaningful impacts. We've got new services and standards being used to stand up, new services, 90% created by Gen AI based on microservices, network scenarios and other scenarios. So this is real, this is happening, this is having an impact where we just see there's fast differences in the readiness of the business to get there. Okay.

Ray Le Maistre, TelecomTV (00:43:33):
And well from the Telefonica perspective, are there any real first notable steps that the company has taken to show internally and to show externally as well?

Juan Manuel Caro, Telefonica (00:43:47):
Yeah, yeah, for sure. I think I say we have been working with it for a long time and there are practical, we use AI in the network today and for last years and it works. I think we are far from in AI native I think in the industry and telephonic and as it says now we have a structural program to go there and it will take years and I am happy. I'm keen to see the poll if it's 2030 or never. That's a funny one. But I think we have a program structuring eight projects. One is about use cases but the other are about fundamentals on data skills. We was working tooling, n oss, whatever. So there's a lot to change drastically to become AI company, even the network itself now we have to go to ization of the network clarification, we have still the two G there and that's not going to be governed by ai, probably never, I dunno.

(00:44:36):
So that's a huge transformation. The base is slow. You have to build from bottom on half. Setting the basis well to be able to progress towards that autonomous network and AI native operation and a sneak set. I think there's a resistance to change the way you work. But I would say we know how to do that and AI is quite new. But automation, I been in tele franca for 20 years, 23 and it started doing automation but rule-based whatever it was. But we have been doing that for a lot of time and you have to involve the people that know and show to them. He talk about the hype before. If we think AI is going to be like this Terminator or Skynet that is going to govern the network and nobody else is going to touch, that's not the way. So we are thinking more in AI for example, assistance to the people making the job easier, helping them to be more productive, to focus on things that we need really, I don't know.

(00:45:33):
I think we have to set up this message clear and transform. But on the other hand, while we are transforming, and this is going to take a huge transformation with Telefonica, we are 110,000 employees, a hundred year company this year, last month. So I have to say it takes a while. On the other hand you can use it. Yeah, we're using that across all the whole cycle as dt, we did the same, we call the whole life cycle and then put the domains radio and we did this metrics and we try to understand well do we use that? We have a project for each one and we are aiming to reach level for the team for end of next year. So that's a lot to say for project for some processes. So we are using that in planning, so to forecast traffic, that kind of things.

(00:46:19):
We all are using that and it works very well. We are now piloting digital twins and things more complex that will take more time in operations. As you said, we are using predictive maintenance, we use that Germany in some operations. It works for certain scenarios and it predicts things. Anoma detection, we are now launching a huge observability project across the group and normally detection is very useful and as he said, once you have the data, it works, the algorithm works, there's no data and it's not rocket science or something, we have to wait five years and it's getting results right now. We use it for energy efficiency for example. It's powering features boost by ai. We have deployed across the networks and one that we are most are, or at least me excited is customer experience really. So we have created that. That's something we have done internally, AI models to be able to understand the experience of our customers.

(00:47:11):
So based on all the network KPIs, we have trained models for mobile broadband for wifi customer experience indexes. Now we used to launch a poll a thousand customers and say are you satisfied or not? And with a thousand thing we tried to understand the quality of the demographic of a country and we couldn't do more and you said no, then you ask why and then you dunno what. So now we have trained these models with all these answers and we are able to predict if a customer would answer satisfied, neutral or dissatisfied based on and not only what this cluster of customers would answer, but why? Because the model is explainable. And so we can really manage the network based on that. And that's ai and that's a machine learning and that's a trained model and that works. And that's been working for five years in our networks in our DNA. And so I would say yes, we have a huge things coming. It is super difficult. We talk probably about data, there are a lot of challenges, skills, but we've been working on AI all of us for the last five years and there are AI real cases working in the network. So we don't have to be frightened of this super hype. It's a valuable tool and tool that we have. Another tools, this probably more powerful, but it's not something of the future, it's here.

Ray Le Maistre, TelecomTV (00:48:31):
And Mabel, from vodafone's perspective, is this something you've taken on a piece by piece basis or maybe even try out in a particular market first?

Mabel Pous-Fenollar, Vodafone (00:48:40):
Yeah, so going back to the initial question, I think it is super important and some of the panelists touch on that, that AI is a change. So it's a transformation. So it's all about the technology really, really important, but also the people and the processes. So I think it's super, super essential that at the first step you really consider those three circles because otherwise it's very difficult that you'll achieve an outcome. So that will be one of the things that we will advise. And then the other thing is that when we started the journey, similar to what you guys have said is you look at the data and you look at versus what is the quick wins in operations, in the assurance side of things, we done quite a lot of the stuff and yes, we usually pick a market, sometimes we pick a large market because there is more of it and then you see the results bigger.

(00:49:33):
Sometimes a smaller market give us more the flexibility and the agility. So depending on the use case, we use a market to start with and then the challenge comes into the scale across the markets. So I think this is a challenge that we see. So start small is what we will say and then see how you can scale. Sometimes we don't believe that certain areas can be scaled across all the markets because it might not make business sense. Some automation or AI is very particularly for certain markets because it's a challenge for that. Others are very much a scale. So we look at the opportunity on the scale and how much is a completely redo and reuse side of things. So yeah, I will say focus on the quick wins but also keep an eye on the scale and keeping up to date as I'm always saying. Right? Because the reality is that, and we've seen it from learnings, you put it one of the tool sets change and then the whole thing gets destroyed. So it's not something

Ray Le Maistre, TelecomTV (00:50:39):
Spinning plates. Exactly.

Mabel Pous-Fenollar, Vodafone (00:50:41):
So it's not a technology that you live there and it say okay, it goes on its own. It has to live in the lifecycle of everything. So as you scale is that maintainance and that run that needs to be very much a scale. And the last thing I will say is that usually when operators like us look at the outcomes you want to reach the outcome for customers, simplicity, the bigger and sometimes we dismiss the innovative part of the new technologies. They need to come automated by design. So yes, focus on the four Gs and the three Gs and the two Gs, but all the new ones, they have beautiful technology embedded, some that we can put on. But let's start bringing that through the lifecycle already automated. And I always joke like then my team doesn't exist because digital is embedded with everyone. And that's the ambition that we also want to have is that focus on that as well.

Ray Le Maistre, TelecomTV (00:51:36):
Okay. ied, I had a quick

Ahmed Hafez, Deutsche Telekom (00:51:38):
Thing. I agree with all what's said and use cases, but also go back to AL'S point when in your speech you mentioned that the use case perspective is one perspective, but let's come from another perspective. There are three steps that we can take one on data. And what happens is that with predictive ai, we did not have that demand on data because it was very specific. So if you get one piece right, you can do ai, but with generative AI it becomes very different because we need to look into data consistently from data catalog to the ontology of data and relationships. So as Nick was saying, to connect the dots and have a root cause analysis that cross domain, then you need completely different level of data, cleanness and data preparedness. So data, we have to work on data if we want to be serious about ai.

(00:52:23):
The second part is the combination of the skills and the tools. Also, we recognize that also your point, Nick is absolutely correct. If I have people who have 20, 30 years of experience in networks, I'm not expecting them to become data scientists. And I'm also expecting them to go to a data scientist to do anything on ai. We need a mix. We need them to give them tools to empower them to, with no code, low code, manage ai, create a new AI application. So you need the right tools to have that. Then you can empower the organization underneath, then you have everybody working on it. And then with that, finally the whole organization becomes centralized and decentralized. You centralize a governance, but you decentralize the implementation. And then I always say that AI as a marathon is not a sprint. So this is something that we will continue to work on until we retire. It'll not stop.

Ray Le Maistre, TelecomTV (00:53:11):
Okay. Manish, I want to come to you and get what you are hearing from your customers. And then Amal, I know you've got a couple. Yeah,

Manish Singh, Dell Technologies (00:53:18):
I think a lot of the points the panelists have already, and I kind of really agree with them. I'll say just one additional thing. I think one of the things that's very clear with AI is the number of use cases. And I think it's easier to imagine a new use case every day. You can just imagine more use cases than you can think of. So the question really becomes is how do you pick and find what's the right use case or use cases to attack? And one of the simple two by two matrix that I typically use in the discussions is, Hey, look at your CapEx optic spend, sort it out, where is your biggest spend happening? And then look at another axis of your data layer, exactly the point where is your data readiness. And between those two, you're going to identify the sweet spots of use cases you can get initiated. Getting initiated is absolutely essential. And I can't agree more with the panelists around people and process. I think the technology again is the easier was said earlier as well. It's the people in the process and upskilling that Amal touched on. That's the challenge. That's a real challenge. So getting initiated I think is important. It's a few set of use cases.

Ray Le Maistre, TelecomTV (00:54:30):
Okay. And Amal, I know that you've been working on a number of particular step points at lenor. If you can tell us a little bit about those and

Amol Phadke, Telenor (00:54:42):
Yeah, I mean I think I'm just trying to reflect on what the panel has been saying. There are a few use cases that we see that tick most of the boxes the panel mentioned in terms of are they really going to give us a benefit? Are they pervasive enough that everybody would get interested in them, would it actually matter to the board and so on. And so one area that we have now started to look at quite earnestly is energy consumption in the networks because we believe that is actually ticking a lot of different boxes from sustainability, from customer experience, from cost and just from overall being good citizens in the industry. And it's actually something that, again, AI can help us a lot. I mean we now are partnering with a lot of our vendor partners in terms of asking them to become more intelligent in their AI solutions.

(00:55:33):
So we can sort of selectively turn off networks where we feel that demand isn't there or the usage isn't there, but at the same time be ready to dynamically turn them on to full capacity if needed. And that is already showing us almost 10% saving in power, which if you can imagine TE loss, total power bill, that is something that the board suddenly wakes up and says we must do AI to Nick's point. So I think we need that kind of galvanizing use cases. Other big area for us, which is a massive pain point is obviously in the Nordics we build a lot of fiber, we build a lot of fixed wireless access, we build a lot of 5G. And as all of us in the room know that is a very, very intensive task and is prone to a lot of errors and callbacks and repeat visits.

(00:56:20):
Again, huge amount of data that we have. We haven't just used it intelligently enough and now we are starting to look at that and say can we equip our engineers who are actually doing this with all of that repository real time with AI enabled assistant that can sort of avoid any missed appointments or repeat appointments or any kind of revisits that just keep adding costs And that has a customer experience benefit because often the customers are going to call and say when am I going to get connected? And it'll be a wonderful thing to sort of go back to them and say you already are. And we've sort of used AI to do that. So I think we need to think beyond the sort of standard use cases of which there are many, many and be very selective in what we go for because that's where I think the industry will see the most value.

Ray Le Maistre, TelecomTV (00:57:10):
Okay. I'm conscious of time here and there's a number of talking points we still want to get to. So I'm going to come to the audience in a minute and see if there's any burning questions for the panel. But I just want to touch at this point on the topic of the kind of infrastructure support that the network operators need as they consider how to automate their network operations. There's a lot of talk about what can be done, what can be processed in the cloud, and of course there's a data issue there, security issue there. So is there a big infrastructure consideration here apart from the fact that it's really hard to get hold of certain G? So Manish, if we can start with you, and this is really in the news right now, isn't it, about who might be buying what GPUs and then using them for something else.

Manish Singh, Dell Technologies (00:58:08):
Okay. I think infrastructure implications, there are plenty we can address and tackle this topic on multiple dimensions. Let me just take a few. Number one is the silicon ecosystem. You've got to look at the silicon ecosystem, the GP ecosystem, what's happening and I think the good news is there's a lot of innovation that's already happening in there and we are going to see diversification on the silicon ecosystem itself. What I mean is there're going to be more GPUs coming in from different players and that's good news. So we will have diversification in the supply chain. Number two just a little bit, I'll get in the weeds. There are also infrastructure challenges around getting networking because it's not about just the GPU, you've got to get the storage and the networking so you can feed the GPUs effectively. And so there's more work to be done to get the networking piece.

(00:59:07):
And so there is again, good news, there's more work already being initiated in the industry to even get to ethernet based solutions, get 800 gig ethernet et cetera, which is all very critical to again create these open ecosystems, open platforms and create scale diversity choice on the infrastructure side. That's one second way you can just slice and dice the infrastructure aspects is around power and cooling. You've got to look at when you're talking about these systems, these platforms are power hungry, GPUs are power hungry. And so both from a power and then cooling especially. And the implication again there is in many cases we have to start thinking about direct liquid cooling standardization of that needs to happen in the industry. And so there's work that's getting initiated around that. And then I'll stop at this last one which is the third one is where are you going to place this infrastructure as a telco and power cooling puts the constraints.

(01:00:09):
So you have your national and regional data centers, you have your edge locations and where is the right one? And I think where we are right now, you're going to see a lot more of if you're going more GPU heavy, you're almost in many cases limited at least the scale systems I'm talking about inside your data centers on the edge you can have certain lower form factor accelerators, especially on the edge locations. So you can have that because you're going to be constrained by power and cooling air capabilities that you have available. So are, I mean it's a complex puzzle piece that you have to work your way through and then identify what's the right infrastructure, where do I place and then is the ecosystem open enough in terms of the needs? Okay,

Ray Le Maistre, TelecomTV (01:00:57):
Yeah Nick quickly and then I'll come to you Matt,

Nik Willets, TM Forum (01:00:58):
Just one build on what man said, you've got to also look at that outside in as well as inside out. So quite often these conversations end up being where are my workloads going to be? How much are I doing myself versus in cloud we believe AI is also going to change what enterprise customers need in terms of creating demand for edge. So you've also got to think about what does that do to my infrastructure planning where I may need to locate compute close to customer premises And that creates all of the implications Manishh just talked about. So all too often in the industry we're thinking about our network, our needs, there's actually an opportunity here as well to think and plan and that may offset some of the inevitable costs we're talking about.

Madhukiran Medithe, Rakuten Mobile (01:01:37):
So adding to the points mentioned by the banish. So we have our own cloud platform rock having their own cloud platform. So we capture every telematic logs fitting from the system coming from the base station to the subscriber AP N gateway as well. So now when you do the scale and scale out, there are few things which we need to look into that that's what we normally do that that's how our A model does that. So before getting into that, standards are also evolving towards that. NAF is one of the network functions. Basically what subscriber is doing, what kind of application and when you tie up with this brick solution ran digital controller where you can deploy your modules on the top, which you can play on the ran side. So overall when you try to marry these two functions via own app functions, so you will see that where I have to the scale out what application, what cell or what cluster at what time the traffic is getting consumed, that's where we are playing our energy consumptions.

(01:02:38):
That's where we're optimizing our resources. So of course definitely when we talking about when you go with the AI models, definitely the consumption of the power is more expensive. If you see the statistics like 60% of the energy consumption happening in the RAN and the remaining 30 to 40 is coming on the cloud, your basically CS data centers. So the RAN in the trend 60%, hardly 15 to 20% is being used. The remaining are 40 to 50% are only for the coverage. So there you have to play. That's where you have to play very smartly, not only with the automation, that's where your AM ML models will come into picture. That's where you Rick control is giving your platform over there you can deploy your AI models, understand what's happening in the core, what applications being used, and then try to deploy on the ran segment.

(01:03:20):
That's how we have to do stitch by stitch. So now implementing this AI model is one kind of set, but deploying is one more set. So the biggest challenge what we see today is how to test my AI models. I'm not looking to my fund score, I'm not looking to my gradient or precision records when I deploy any model into the network, any change in parameter leads to so many diversions. So to do that we need to have a digital twin kind of application where you can test your model and where you can simulate your model. So we don't have such tools we have to implement ourself. So that is something which we're developing, those kind of things. So any operator or any vendor talking about ai, they should have their own test tool, not as a traditional test tool. When you put the model, you have to simulate the actual traffic in the actual consistency.

(01:04:08):
So that's something which we have to develop inhouse tool. And the second thing is how you take it into the production. So when you go to production, we'll see a lot of issues in the network. So you should have CICD pipelines over there. So when you give a patch, it has to be implemented exclusively aggressively. We cannot go with the pr, you cannot make the sell down for a couple of hours, a couple of minutes. A lot of traffic will be distorted countries like in Japan, every minute is countable. If you see the outage close to two to three minutes, we have to send our apologies to MIC. So these are very restricted complaints out there. So these kind of things needs to be implemented in the network when you're going to them production. So this is a lot of ecosystems involved over there. So it's not resources, but how you use your resources effectively when you put your AI modules, having leveraging these kind of platforms, definitely we have successful solutions in the network.

Ray Le Maistre, TelecomTV (01:04:59):
Okay, Amol, I want to come to you for a quick comment here, but I just want to say I'm going to go to the floor for a question straight after this. I can see one over there. But Amal, in terms of the infrastructure, because

Amol Phadke, Telenor (01:05:14):
I would say that on the infrastructure side we also have to now be cognizant of the fact that we have to start aggregating the infrastructure and find ways by which we can run multiple applications on the same infrastructure. Going back to man's point, these infrastructures are very power hungry and I have now sat in a lot of conversations or many years where radio requires its own custom. Silicon AI requires its own GPU. The processing that we do in the core really needs an intel architecture. But I'm really definitely not going to be able to invest in three different sets of infrastructures going forward. And especially because the packet processing and the latency that we need is now getting to the point where a lot of the chip sets are able to do some of the things we wanted to do. So we as an industry should also need to push now quite strongly on how can we get a same set of compute infrastructure to deliver multiple applications, whether it's ai, whether it's ran, whether it's other things as opposed to building vertically integrated stacks for each of those applications, which then add power and energy and so on.

(01:06:22):
But yeah, that's a call out to the industry and then we should go to the audience.

Ray Le Maistre, TelecomTV (01:06:25):
Yeah, so I did see, yeah, I can see your hand up over there. If you can say who you are, where you're from and your question for the panel please.

Francis Haysom, Appledore Research (01:06:34):
Hello Francis Haysom from Appledore Research. A lot of the discussion I've heard about AI and autonomy, I put it in the sort of picture of being, we are looking to mechanize the processes that we do at the moment. We want to make a better knock process or we want to make a better planning process. One of the things, another example in AI is that when you look at things like visual recognition, for example, the major leap forward in visual recognition using AI processes was when we stopped trying to tell AI how to do a process. And we allowed it to discover how you identify a cat in a picture or a person in a picture. And I wonder whether the challenge that we have as an industry is that we are looking for a sort of perfect AI replacing something that we already do rather than AI giving us a solution which maybe does something in a completely different way than a person does. And the question to the panel is how much are you as part of your AI initiative looking at not just how perfect AI could be, how perfect data could be, but actually the reality of human decision making in your organization and where AI is already better at it by doing things differently. How much are we measuring the human intelligence as it were in human decision making in our organizations?

Ray Le Maistre, TelecomTV (01:08:10):
Okay, Ahmed, I think can

Ahmed Hafez, Deutsche Telekom (01:08:11):
Give a start and then I leave the others. So I like very much the, because when we started looking into use cases, the quick wins, the quick wins have this notion of looking into steps on the way. So exactly what you said is that we will look on existing processes. We're not trying to change the process of just automating part of the process. And that's I would say stage one in ai. We're still in that stage one, the tactical stage where we're trying to get our hands dirty, understand more, get into the problems, but the second stage which we ought to do is actually to rethink the entire process end to end. And that is a third change. So the third stage would be that we let AI decide how to not maybe implement a process but have a diffusion, not a process, something different, which we don't know what would it be, but we are still in the phase one. So there's still two phases to come and I acknowledge what you said,

Francis Haysom, Appledore Research (01:09:05):
But my question is, are you measuring the effectiveness of this human decision making in your organization? Because that seems to be a key thing. Without that knowledge of how effective we are in decisions in planning, in capacity management, I

Ahmed Hafez, Deutsche Telekom (01:09:20):
Think that is reflected in the results of every year. So we can see that as it go. But of course some decisions would only appear over time. There is not such a mechanism easily to actually quantify and measure the decisions. If we can also do that with politicians and politics, that would be fantastic, but it's not yet there. But we are at least trying to measure the AI side of things.

Mabel Pous-Fenollar, Vodafone (01:09:43):
I can give a clear example which and again is not, I think we probably are all working through it. So we have seen that automation, it's good because humans make errors. We all make errors. Anybody who has touched the network made an error, I did as well. So we do measure and we have use cases to bring that error down and our engineers acknowledge that one of the use cases is what we call have the second pair of eyes with an AI agent. So instead of having a lot of really subject matter experts, we all have those subject matter experts, but AI or the gen AI can be that subject matter expert that can drive to the right decision. So we are starting to see that on the change area where when there are not that many human errors, but when they are, they're pretty noisy and it's an area that they might not bring simplicity or cost reduction, but it's an area that we're starting to realize that we as humans should leverage from technology when they have a better driven position. So that's one area that we are exploring at Vodafone where we're using them as the ones helping us to not make errors.

Ray Le Maistre, TelecomTV (01:11:12):
Okay, thank you. Very quick comment from Nick. Yeah,

Nik Willets, TM Forum (01:11:15):
I think it's a great question. I think you can map how ready from a management layer down the organization is as to whether they're asking that question. People are very comfortable in call centers for some reason, just we just think call center people are not valuable or something. But very comfortable to say gen AI will do a much better job call summary of sentiment analysis than a human being. Just I think it has the time it can do more accurately as we get closer and closer and closer to the belly of the beast in terms of network operations, there is more resistance in terms of that. But it tends to start I think as both the other panelists have said in terms of saying, okay, we can see now that more of the time the AI is on parity or making a slightly better decision. And we see for say, field force operations understanding that not all of your field force are experts and everything. So it will deploy depending on the use case and the AI will typically get that better than a human who even knows that field force quite well. So I'd say is embedded in the logic where the politics creeps in is where actually the workforce doesn't really want it to happen. And so it's quite a good metric actually to say if you're not asking that question, maybe you're not really moving towards being AI native.

Ray Le Maistre, TelecomTV (01:12:23):
Okay, thank you. Now I'm getting the red flashing light to say we're running out of time, but I do want a very quick comment from you Amal, before we end on the security implications and about whether the industry really understands the security implications of the greater use of various AI tools. This is something that's the, that you have to spend a lot of time thinking about.

Amol Phadke, Telenor (01:12:49):
Yes,

(01:12:51):
We do spend a lot of time thinking about this and we all are sort of avoiding the nightmare scenario where something comes in the press about something about our customers in a model that something else red. So for me it is about being thoughtful about the choices we make in terms of the architecture. We build the hardware, we buy the models, we train, and ultimately goes back to what I said before. There are two forces here. There is a force that is going to say if you're not fast enough, you're going to miss the innovation and there's a force that's going to say if you're too fast, you're going to mess up. So I think you need to sort of find a balance here to build that secure infrastructure is a must, but at the same time, if you spend two years building a secure infrastructure, you're going to miss the innovation. So I think there is a trade off that we'll have to do, but it's the choices we make and building a combination of LLMs hardware and software with the use cases on top that what's going on in your organization would be the only way to solve it.

Ray Le Maistre, TelecomTV (01:13:53):
Right. Okay. Thanks so much, Amal. We do have to end there. And before we move on, we are going to take a quick look at the poll results. The poll we had at the beginning of the session and the question we asked was, when will AI enable telco networks to truly zero touch without any human intervention? Never has got a pretty big, I thought never might be quicker. So 37% are saying never 42% after 2030 and 21% by 2030. So that's quite a task. Let's take that as an action point. Our vendors

Ahmed Hafez, Deutsche Telekom (01:14:40):
Are 2030,

Ray Le Maistre, TelecomTV (01:14:43):
So fantastic. Thank you much for voting. The polls are not closed so you can continue to vote in these polls. But that brings us to the end of session two. We're going to be back on stage this afternoon at 2:00 PM UK time. That's 1400 hours for session three, which is creating cloud native software engineering teams. In the meantime for us here in the room, we have a leisurely lunch break, which is going to be served in the foyer. And don't forget the pinball and the charity tournament we've got going on over there for our online audience. Don't go away. We're about to start our extra shot program. Charlotte's going to be picking up on the AI themes with her guests, so don't miss that. Stay with us. But for now, let's say thank you to our panelists here. Great.