Ray Le Maistre, TelecomTV (00:08:34):
It is now time a little bit early, maybe by a minute, but that's give us time to get our speakers up on the stage because it's time to get back to the debates and motions. So I would like to invite our next lineup of expert witnesses to the stage for our path to autonomous network session. So please give them a round applause as they come up to the stage. Look at that. We're going to get an extra minute, but we have got a lot of speakers of course, to get through in the next hour. So the path to autonomous networks, I mean automation has been pitched as the path to cost cutting to faster deployment and reduced human error. But without strong governance strategy and an understanding of hidden dependencies, could automation erode resilience, inflate risk, and even cost more in failures than it saves? What if automating too fast or automating the wrong things introduces compliance gaps or stifles innovation? Innovation because rigid automated processes become brittle, maybe the promise of a fully autonomous network is just an illusion. Well, there we go, plenty to discuss in the next session. So as usual, we're going to come to our expert witnesses first. Then we'll be going over to Graham and Chris to hear the motion. But let's get cracking on with our expert witnesses. And the first, with the three minutes on the clock is Ian Milligan, director of network development and infrastructure at VodafoneThree. Ian, the podium is yours.

Iain Milligan, VodafoneThree (00:10:23):
Okay, afternoon everybody. So as part of that introduction, the thing that I think is actually very exciting actually and quite relevant to the context I'm in is obviously we've just completed our merger and we have that challenge of you need to get the cost savings, bringing two networks together, tech you've got to deliver. We have a legal obligation to build 26,000 sites and get them fully upgraded across an eight year window. Deployment has to be fast, and there's the whole risk component. We need to do it as we change our culture and really our new technology. So we are in an actual sweet spot of having to address this, and if we don't address it using what we're talking about and actually using autonomous networks, using the ai, we will never succeed. So we have to embrace it. There's no choice. It doesn't mean that we're reckless and that's a very important point, but what I think is a real important aspect to understand is, and I've seen it many times where we talk about use cases where it's let's operate, do this within operations, faster ticket resolution, it's more than that.

(00:11:30):
You have to look at the component of the network from the very bottom to the very top. So what's in the radio units, what's in the base band cards, what's in the optical layer? And you have to get the right technology connected in the right way to make sure that you have the environment to do it in a controlled and safe way. And it's something that we're quite proud of. What we've managed to do really fast in our first six months, we've actually given a good foundation in of our architecture or design to do that exact same thing to hit the pace at which we need to go at. So we're on that journey of how we're going to do it, and it is very much in that modularized approach where you can have your controlled automation at a product and solution level before you start talking about your big use cases to enable your differentiation.

(00:12:14):
So that's how we see it, that's how we've started to see the benefit of it across all the way, like I said, from the lowest level of radio component to the core network to the OSS. And that's what we've started on in terms of our journey. Now, the benefits we talk about is an engineer. My big fan is okay, I always want to do stuff easier. I want to be able to just be plug and play, plug and play. But the other thing which I see very powerfully, especially bringing the two cultures together within our organization is the X Vodafone team are exceptional at monetizing products and driving that capability. On the three side, we were excellent at making things simple, putting that together as we build the networks, the product and route to market is so much faster because what we can actually enable our businesses customers, the enterprise teams, the consumer teams, they got a good idea and they're so dependent on network Rather than me having to re-engineer all time, we can actually build a quicker route to market to do that. And that's not within any way compromising engineering efficiency. It's actually been exceptional engineering for better customer outcomes. So the risk is yes, you can go too fast, but it's around how well you control that, how well you have the right governance, the right structure and able to do it without constraining that freedom to be able to do. So. That for me is the opportunity that I see. Thanks very much.

Ray Le Maistre, TelecomTV (00:13:36):
Okay, great opening there from somebody who's actually living that on a day-to-day basis. So thanks, Ian. Next up Brandon Larson, senior VP and general manager of Mavenir Cloud AI and IMS business strategy.

Brandon Larson, Mavenir (00:13:56):
Well thank you and good afternoon. And I'm going to follow up with Ian said journey. I think that's really the key part. So the journey that we're looking at is in the area of network operations, specifically lifecycle management operations. So think about where operators have this ongoing work, deployment, upgrades, updates, security patches, so on and so forth. That's where we see a lot of interest in automation. And why do they see that interest in automation? They need schedule compression, right? They need to get those features and fixes and security patches rolled out faster. They need to do it with less people and they want more consistency in those operations. So that means less variability and less error introduced by humans. So what they're doing about that is they're deploying GI ops based LCM automation and that's really what I would consider a very sizable step towards autonomous networking.

(00:14:49):
But a very practical one. GI ops is a very mature technology widely deployed outside of telco and it's a form of automation where the operator still has a lot of control. It's declarative automation. They define the end state of the network. We call that the declarative state. The automation just does what the operator says to do and when they create that desired state, declarative state, it's a single source of truth, it's auditable, version controlled, and if you go to desired state you don't like, you can roll it back, right? So a lot of control there. In addition to the advantages I mentioned before, it's faster, less error, less people. It offers some more advantages in GI ops. One is closed loop if that desired state deviates itself corrects, that's fantastic. It also offers a very framework that can be implemented across domains. PAC core IMS ran different vendors so that harmonization operations, very important to get to autonomous networking and it's also extensible.

(00:15:50):
And by extensible it means that we can add, easily add more things that can be controlled by that automation framework. In fact, we've just extended it to automate AI resources, right? And that's going to set us up for the next big step in the journey, AI driven GI ops. And what that means, we're going to actually have AI agents determining that desire to end state of the network instead of people. And those AI agents could come from different tasks such as demand planning, service fulfillment, service assurance, and that's where we're going to need more governance and control. And that governance is going to be in the form of guardrails. What are we going to allow it to change, not change, and by how much it's going to have to have conflict mitigation. What if I have two A agents, they're asking for different things, how do I resolve that? It's going to have to have risk analysis. AI can determine what is that impact if I make the change and should I do it? And we'll have to be very judicious about what we per human in the loop. So overall I believe that autonomous networks is going to be very successful if it's done incrementally with very well thought out steps along that journey. And a great first place to start that we've seen in the marketplace is with this skid ops LCM automation right at the end of the clock

Ray Le Maistre, TelecomTV (00:17:02):
On the money. Incredible. We have seen some great timekeeping today I have to say, as well as great presentation. So thank you Brandon. Next up is Diego Lopez who is an ETSI fellow as well as being senior technology expert at Telefonica Diego.

Diego R Lopez, Telefónica (00:17:22):
Thank you. Well, let me insist on the journey because one of the idea was about to talk about our autonomous network journey. It's formally called like that in Telefonica, not necessarily what we're doing there. We are almost at level three, approaching level four in some areas, et cetera. But something about what we're doing for building trustworthy autonomous networks that I guess is one of the themes here and how we are coining some essential concepts and translating into something that can be used for real services in the framework of a project that is called I trust six G. I mean because we expect to have trustworthy six G networks in the future and around three essential aspects. One is a modeling, second, we call it enactment. And third is about explicability because I learned that explainability is argued as a right English word, that's something that I read somewhere.

(00:18:26):
Explicability know that today we are dealing with schedules and schedules and things like that. So the point is that when building these MEE or me things, we are trying to do it by doing graphs and not loops. We are trying to go beyond the idea of the closed loop because of two reasons. One is basically because then you don't need a human in the loop for increasing trustworthiness, but you can have a human at the graph. So taking control of it. Second is because we believe that it is not only about the assurance phase of the service is about controlling it, but it's important as well to have autonomy. For example, when you're talking about the fulfillment phase is about how you express what you want and how you get it for this. When talking about models, what we're trying is to precisely modeling intent, intent that make sense and translating those intents that make sense once we evaluate the sense of those intents into something that is a smart contract that can be automatically processed.

(00:19:33):
Second, we're talking about enactments about how we apply. First is we try to apply two things when it's environments, what if so we can evaluate the impact of a particular intent or a particular contract on the network. And second for sure translating into policies that are evaluated in a distributed way. And third about the applicability is the idea of relating whatever it happens back to the original intents. So it can be understood by the operators, by the users, by whoever got a particular status or result from the network. And for this, we are working in three directions, new directions I would say to be applied. This is about identity for all humans and machines and relating them ontologies and trustworthy data, putting trust on the data that has changed. And for this just to finish, we are trying to make it agent enabled but not necessarily or trying to go beyond LLMs. That's all. Thank you.

Ray Le Maistre, TelecomTV (00:20:36):
Okay, thank you Diego. Next up for three minutes is Fernando Rionegro, who's VP of cloud and network services for Europe at Nokia. Fernando .

Fernando Rionegro, Nokia (00:20:48):
Thank you Ray and good afternoon. So as Ray was pointing in the introduction, the telecom industry is in a critical dilemma. It can be either moved towards intelligent autonomous networks or stay in an improvement, gradual improvement process as it has been done during the last years. We see significant improvements in AI autonomous networks, but I think that the industry is not yet there. According to TM four, only 4% of operators are at level four as the was commenting, most of them are between level two and level three of autonomy. And it's not until 2030 when we are expecting that most of them will be at level four. So there is a balance between what the operators need to do in innovation and the operators has as facing the reality in their operations. There are silo solutions, complex architectures, regulatory pressures, but when is the right time to start with this?

(00:21:50):
Expecting or waiting sometime to do the decisions about autonomous networks is also a problem. It's a high risk in mistake. What is in stake? Mainly three topics. The first one is about customer experience, autonomous networks, print service differentiation as an unique customer experience. For example, for quarantine performance or zero error networks. The second one is network efficiency that has one was commented by brand network efficiency in autonomous networks enable intelligent decision making systems in processes that before were done usually in a manual fashion. And last but not least, savings in CapEx opex. That is what really moves the needle in this industry. That is quite flat. Industry being able to reduce cap opex is a key factor. Network automation is not an all in process. It's as has been commenting four times already a journey differentiation. Parts of the network were on different levels of automation or autonomy.

(00:23:02):
One example is for instance security. Security can be a good example where you can apply autonomous in the moment that there is two sides where you can apply security and is becoming very relevant. But also adjustment consent and being able to observe what's happening in the network under errors. It's a key factor in order to implement autonomous networks to benefit from the full potential of autonomous networks would require a focus plan. Setting the goal, defining the path to get there, executing and measuring the progress. Autonomy is not a certification. You have to have the key understanding which are the biggest impact that you want to do in your network. So focus on the progress and not in the perfection. Thank you.

Ray Le Maistre, TelecomTV (00:23:54):
Great. Excellent. Thank you very much for Fernando. Next up for three minutes is Timo Jokiaho, who is Telco field CTO, at SUSE. Timo.

Timo Jokiaho, SUSE (00:24:09):
Thank you Ray. So we already heard some good topics here and of course we as suse, we are supporting these guys. We are building the networks ourselves. Nobody is. So we are supporting now my telco journey. I have grown to be passionate on few things. I've been working for three different major equipment providers. I don't remember their names anymore, but I have, and this is my third open source company I'm working for. I'd like to share a couple of passions I have. Telco is of course given, open source is given as well. I've been an open source passionate since the famous Leal email in the autumn of 1991. Now the third passion is, which I've really grown to like, which is going to the topic of today, is that I believe that common consistent, high quality, high performant, secure platform, software platform across the network is going to help achieving the autonomous network goals.

(00:25:22):
That's something, and I'm passionate on breaking the vertical integrated silos by vendors. I think that's not a good idea. Industry has been on this journey for, I dunno, 15 years at least since the publication of network function virtualization white paper 2012, 13 years ago. So if we have a more common foundation to build the networks, that would help in many aspects, end-to-end management would be easier and end-to-end automation would be easier. And then this would lead into more or fully autonomous network at the end. And if we have all these things, fantastic platforms and horizontal platforms, whatever you might want to call them, that also helps in the AI space. Like it would be easier to collect data for AI to be used to use and also to build agent AI workflows across the network. So again, very highly passionate on doing what I just explained. And of course I don't think there can be a one platform. There has to be industry collaboration, which is one of the topics in telco industry anyways, and I'd like to mention the Linux Foundation Suba project, which is exactly trying to achieve this one, not just in the networks but outside of the networks as well. And it's reasonably successful already as we speak. And I trust, I believe, I hope my fellow software platform vendors in the room would agree with me to some extent. Okay,

Ray Le Maistre, TelecomTV (00:27:26):
Thank you Timo. And finally our sixth speak on this session is Prashant Agarwal, who's business development manager for Telco EMEA at Intel Prashant.

Prashant Agarwal, Intel (00:27:40):
Good afternoon everyone. As the last speaker, I was hoping that panel will automate this part for me, but I think we are still not there where AI can fully replace the humans. If this panel was truly autonomous, I may have been replaced by an algorithm by now. So let's talk about automation. So automation is sold as a superhero. It can cut the cost, speed the things up and eliminate the human error. Sounds fantastic, but remember superheroes can also have weaknesses. So what can go wrong? Quite a lot to be honest. Let's see, without proper checks, automation can bypass the compliance. Just think about automation system rolling out the configuration which can violate the regulations not because it's malicious, it just because there are no proper checks. Also automation at a speed, it can be very risky. Rolling out thousands of network changes overnight without proper validation. Even a single error can cost it through the whole infrastructure.

(00:28:57):
And also don't forget the networks are interconnected. Automating one part without considering the impact on other side can be catastrophic. Just think about removing that unused route. On one side you may accidentally break the fall over mechanism on other side. So what about this dream of fully autonomous network? It sounds amazing, but to be honest, reality is a little bit messy. We still need the human intervention for the complicity, for compliance and maybe to deal with those unexpected scenarios. So what is the way forward? I will say maybe start with low risk repetitive tasks. For example, automating configuration audits, scheduling patches, and keep humans in the loop for those strategic decisions.

(00:29:56):
We need to make sure that automation deliver the business goals and not just the technology trends. At the end of the day, automation is not magic, it's just a tool and any other tool, it works best when works with the clear guidelines and human insights. So we need to make sure that automation can be used like a copilot. That's an analogy, but not the pilot. It can use the repetitive task, risky tasks, but we still need to put the humans in the loop for any critical decisions. Thank you very much and let's make sure automation works for us and not the other way around.

Ray Le Maistre, TelecomTV (00:30:47):
Okay, excellent. All right. A good lineup of expert witness statements there, but of course we have a motion. We need a motion to vote on Graham. What is the motion?

Graham Wilde, Three Group Solutions (00:30:57):
We do.

Ray Le Maistre, TelecomTV (00:30:58):
Tell us About motion.

Graham Wilde, Three Group Solutions (00:30:59):
We've got one. Good here it is network automation. It's a Trojan horse. Promising efficiency but delivering risk.

Ray Le Maistre, TelecomTV (00:31:08):
Outrageous.

Graham Wilde, Three Group Solutions (00:31:10):
Outrageous speaking for is Chris. And I'll be speaking against,

Chris Lewis, Lewis Insight (00:31:16):
I think we've already heard in the six witnesses that are presented to us now, the mixture. There's the promise of automation and there's the threat of automation. Now obviously from an employee point of view, we have this challenge of what people will do in the future, but there's also the problem of losing that knowledge, using that human knowledge of how things work within an organization. So talking theoretically about the ability to automate is fine, but actually making sure that the knowledge is embedded into that automation is one of the challenges that we really struggle with as an industry. In addition, and Intel just mentioned this, that actually we're not just talking about one part of the network, we're not even talking about one network. In order to automate everything, we're bringing together all of the networks, the local, the long distance, the mobile, the fixed, and of course increasingly satellite as well.

(00:32:05):
What this means is that to deliver that service to everybody, if we automate, it takes away that human oversight. And certainly in much of the agent AI work, we saw early on this notion of at least two, not just one pair of eyes. And obviously I'm not the right one to talk about pairs of eyes, but two pairs of eyes in every cycle because we do need that human overseeing to make sure that we deliver things to the highest possible standard so that because when things do go wrong and they have been going wrong, we need to make sure the fallback position as was already mentioned, we can rewind, we can go back to the previous state, but we need to make sure that there is that oversight, that human oversight. And of course, as we've heard throughout the rest of the day, that automation and the service, it delivers out to ecosystem partners to ultimately to customers.

(00:32:52):
The threat landscape of what may be brought into the network is certainly extending. So we need to make sure that also covers everything in there. So the potential for bringing that Trojan horse in of any different shape or size is significant. And finally, this notion of data and the data we've heard many times during the day already how data is still in silos. And of course as we get towards true automation and longer term and bringing together the ecosystem partners to deliver that end-to-end service that we are looking for, that we're seeking, we need to make sure that the data that is included is clean, it's coherent and fits into the organizational structure and the delivery that we're trying to achieve. At the end of the day, automation has to support the delivery of the ultimate service, the outcome of the network, not just automating the inner workings of the network. So ladies and gentlemen, I would ask you to vote for the motion. Sometimes it's tough, right? It's all right for you. You got there, you got the right

Graham Wilde, Three Group Solutions (00:33:54):
Answer, that's good,

Chris Lewis, Lewis Insight (00:33:55):
Right? You're looking at the Screen.

Graham Wilde, Three Group Solutions (00:33:58):
No, very good. Well done. So look ladies and gentlemen, in the last 12 months I've got quite sort of obsessed with counting my steps because you can count steps on your phone and you're supposed to do like 10,000 steps a day, aren't you? And I've got really into this and if I feel like I'm going to get to the end of the day and my step count's a bit low, I'll get off the bus one stop earlier or get off the underground, one stop earlier and I walk the extra distance to get my steps up. And I was explaining this to my daughter the other day and I was going, oh look, and I'm showing her on my phone, look at my step count, look over the last two weeks, over 10,000 steps every single day. And she said, that's good, but you are smoking 20 cigarettes a day

(00:34:53):
And fair enough. And that kind of speaks to a little bit about where is our perception of risk here? And it is our perception of risk really correct. The same way that my perception of risk is that I'm not doing enough steps. Her perception of risk is actually you're smoking too much. And I put to you ladies and gentlemen, that the risk around networking automation is that you don't do it. That you don't do enough of it, okay? Because as Ian's been in three for a long time, I've worked at the telecom industry in a long time and I'm sure we can both cite many, many examples of manual interventions on networks that have gone horribly wrong. So the real risk is manual intervention and mistakes that come from that network or information actually promises to improve that situation substantially. Second point is networks are so complicated now, they are so complicated that you cannot possibly achieve the efficient management of a network, let alone the merger of two networks like Ian was talking about. Without a degree of automation, it's just impossible, right? We've already moved away from that. And the third thing is that we're not talking about just kind of randomly introducing some LLM model that controls the network. The way it's done is by taking the low risk things first and then building on those and building on those and building on those and making sure that we understand what's going on and we can roll 'em back. So ladies and gentlemen, this motion is absolutely fucking hotwash vote against. Thank you.

Ray Le Maistre, TelecomTV (00:36:41):
And there goes our stream. So there's the motion, there is the four and the against. We have six experts on the stage to discuss this. We have a room of people full of lunch dying to ask questions. Please put up your hands and wave them very wildly so that my colleagues can see. Now I can see we've got, okay, have, we've got a nice stream. So we're going to start over there and we'll move this way. But please say who you are, where you are from, and what your question is for the panel please.

Ben Hickey, IBM (00:37:19):
Yep. Thanks everyone. My name is Ben Hickey. I'm at IBM. Let me maybe start with to posit a view. So I think as network people, as network engineers, we're kind of inherently risk averse people. So we tend to operate at the bottom of the stack. No one wants to be on the front page of the news. People don't want career limiting sort of outages. So over time as networks have come out, many vendors have released lots of bills and whistles in their platforms. There's lots of advanced features you could turn on. Many of them have gone unused because network engineers want to have a deterministic network. So if that's what we're dealing with, and I'd argue that's probably one of the reasons that's held back automation right now is that people are worried about the butterfly effect or the things that can go wrong. How are you thinking about this from an AI perspective with LLMs, which are inherently non-deterministic systems?

Ray Le Maistre, TelecomTV (00:38:16):
Okay, great question to start, who wants to go first?

Brandon Larson, Mavenir (00:38:20):
I can go first. Yep,

Ray Le Maistre, TelecomTV (00:38:21):
Sure. Brandon.

Brandon Larson, Mavenir (00:38:22):
Yeah, that's a great question. I mean when I consider ai, it's kind of like I think of my kid, all the potential in the world but never does what you want it to do and needs a hell of a lot of oversight to make sure. Right? And I also talk about the industry in being very risk averse, right? I'm sure the cowboys back when they rode horses, were afraid of cars when they first came out, right? Boy, it's going to go fast. What if it runs out of gas? And really the question is are we going to do what everybody does outside of telecom embracing ai, embracing cloud technologies and truly innovating and moving the ball forward? Or are we going to be so risk averse that we're going to stop innovating? And I really think that's the point we have to make, right? Are you going to move forward with the future or are you going to live in the fast what you're used to?

(00:39:09):
And we have this debate all the time, but we also heard a lot through this stage about the journey and about doing the journey. And I've heard it from a couple of my colleagues up here about proper planning, working our way back to what the solution should be. And part of that is putting the proper governance controls, understanding the outcomes as we had mentioned, right? Knowing where that journey is going to end and making sure it's a very well thought out. I think this concept of AI is just a magic genie that's going to solve everything we all know that's bs, right? So really the point is upskilling so we do it right and changing our culture so we have a mindset that we're willing to embrace these innovative technologies that are helping everybody outside of Telco and bringing 'em into telco.

Ray Le Maistre, TelecomTV (00:39:51):
Maybe we can ask Iain.

Iain Milligan, VodafoneThree (00:39:53):
Yeah, it just depends because I've got a slightly different take. I actually think we're quite far down the road of where we need to be and we don't give ourselves credit as an industry. And it's something that Graham said, we don't just switch on a large language model and see what happens. We don't just turn on a bit of ai. What we've got is a really good understanding of products, work with our vendors, with the products themselves. And the fact is, and it's something we should be proud of, it's not just engineering, it's exceptional engineering. Be proud ourselves in getting our teams up to speed to understand how that works. And I'd say at about 2017 and three when we first launched Telco cloud core, we learned very hard how that can go wrong. But we also made it work and from that the engineers themselves started to understand the benefits and the risk. Definitely if you don't try it, you're never going to go. So it is how you build that up. But we're a long way down there and I'd say that we don't give ourselves enough credit in working with the brilliant vendors that create these great products is work with them to actually really understand and upskill our teams as much as possible. So I'm in a more positive mood about it.

Ray Le Maistre, TelecomTV (00:41:00):
Do you think that in general the teams that you've worked with are maybe less risk averse now than they were say 10 years ago? Do you think there's a greater embracing of potential and trying stuff out? Maybe because the tools are there to help that now?

Iain Milligan, VodafoneThree (00:41:16):
Definitely. But I'd also say there's the transparency. So it's not just a matter of buying a black box with the automate button on it, the fact that you get lots of products which show you how it's been made and what the decision making factors are that allows people to get comfortable with it. Good practice around fos or sorry, introductions of new products and the services and features. It's good robust engineering and transparency with the products and the vendors with the use cases you're trying to achieve, the confidence is definitely there.

Diego R Lopez, Telefónica (00:41:44):
May I start with a probably provocative statement? AI is no different from any other software that we have been running in the past and that we will run in the future. The only difference is how it is programmed and how we perceive it, which is very important. And this is something that is, the second part is about the big smoke of the LLMs are making us forget that there are many other AI mechanisms that are not LLMs, that are predictable. That probably is the same. When you have an algorithm, you have a computer, a computer program that is complex enough, you cannot prove it correct either. It is something that is essentially, well theoretically you can, but in practice you cannot. So at the end you're entering a different degree of uncertainty or risk when you're using ai fine because you are abstracting level at another level.

(00:42:49):
And that's why I was trying to assess very much when I was making my introduction on this idea of having trustworthy data, having everything registered so you can backtrack and identify the whys and the ways in your programming and your programming will be the intent and the data that you were feeding on the ai. So as long as we have these safe mechanisms, I don't see any further risk that when we move from assembler to sea and from sea to python and from Python to whatever, Julia is what is now in fashion. So we are simply changing the levels of obstruction no more, no less Fernando.

Fernando Rionegro, Nokia (00:43:32):
So the games are too good to miss it, right? We don't have a choice to be very honest. The point is how we implement it and my colleagues were exactly commenting this first automation as we were discussing before is a journey in the sense that you will select exactly the use cases that you want to implement and you will implement those cases that would have more impact for you. That's the first statement. Once that you have to have the right tools and define the process very clearly. And it doesn't mean that without human intervention, it's not like running artificial intelligence or LLM at their own and see which is the result. The first thing is that they in intelligence servicing networks, the first thing that you define is the what. And then the solution will determine the how. But then at the same time there could be a human supervising of the whole process first guiding and correcting and let's say fine tuning the solution and then further like a supervisor meaning that the solution is already more developed, more secure and then that use case is done and then you go to the next one.

(00:44:42):
So if we think in a full revolution of the network, I completely agree with you, I couldn't go there, but when we go case by case understanding that we have a clear path to go and then we are going in the parts that we are creating an impact, then that's a risk that I could take in order to get the benefits that it's bringing.

Ray Le Maistre, TelecomTV (00:45:03):
Timo, did you want to come in?

Timo Jokiaho, SUSE (00:45:04):
Yes. So I'd like to bring observability into this discussion like collecting raw telemetry data and use observability to contain hallucination of AI in many cases. So it's a very important thing to have a in-depth observability tools in the network, maybe even AI driven observability tool, which is a little bit controversial, but observability is getting more and more important also because what I mentioned earlier, the networks are typically now built by in a multi-vendor fashion and it's sometimes hard to figure out what's happening in the network. So observability is kind of gaining more and more momentum and becoming more and more important. And myself, my intelligence has always been very artificial and I need some observability tools so people can contain my hall.

Ray Le Maistre, TelecomTV (00:46:11):
Prashant, did you want to comment?

Prashant Agarwal, Intel (00:46:12):
I agree with my panelist that we cannot stop and we have to go on this journey in AI ML. So what we can do, how can we gain more confident confidence? So a couple of things I think like traceability and visibility. We need to able to see what the decisions are being made. It's not just what we made but why it has been made. So we can learn from there how these decisions are being made. And I will again say keeping the human also in the loop for the bigger decisions. So AI can maybe propose some changes or some recommendation, but ultimately for the big decisions you need to have a human in the loop. And also maybe we should have a rollover mechanism also if something goes wrong, which will sometime it'll go wrong how soon or how only we can recover, how long it take us to recover Or maybe in the automation side also can it recover without any intervention? What is those time factor? So I think those things will give us more confidence because one thing is for sure, it's not like we can just sit idle and say okay, we not implement, we have to go in that direction. But the main point is how can we make that journey better and how can we get more confidence?

Ray Le Maistre, TelecomTV (00:47:25):
Okay, great. Now I know we've got somebody the back, there's got the microphone if you can say who you are and what company you're from and what your question is, please.

Sherif Sedkey, Virgin Media O2 (00:47:35):
Thanks all everyone. I'm Sharifs from Virgin Media, UTU ran architecture and strategy. I have a question which is really resonating with all of you have said and a certain internal work I was doing on which is automation and AI with a more focus on radio access network. We have a great work from 3G PP or an alliance then in the cloudification by itself and a V and then we recently as mentioned Silva of automation and then TM forum, intent based orchestration. All of these are great things, but my question with all of this work, it creates a lot of permutations why we don't create more effort for the conflict mitigation. It's one term I miss versus thousands of great work of the automation and ai.

Ray Le Maistre, TelecomTV (00:48:26):
Okay, so essentially

Sherif Sedkey, Virgin Media O2 (00:48:27):
The question is

Ray Le Maistre, TelecomTV (00:48:28):
A lot of complexity

Sherif Sedkey, Virgin Media O2 (00:48:29):
About complexity. It's one, but the conflict management to ensure a conflict mitigation with a multi-vendor. Okay, thank you Brandon.

Brandon Larson, Mavenir (00:48:41):
I can start again. No, I mean conflict mitigation, you're spot on. In fact as we move to autonomous networks and you think about you're going to have more things making a decision about a certain thing. So for one RAN specifically, we started looking at doing KP optimization, we'll give it some few knobs. Atune a three offset histories is time to trigger. And even in that small little thing across a cluster of nine cells, we found it was like five times 10 to the power of ninth. Different combinations that we had to come up with very complex. And then if we said, okay, I want to optimize that, then I want to optimize this, we could have the same AI agent coming up going, well I need to turn it this way and that way. And part of that strategy of implementation was putting in that conflict mitigation through policy.

(00:49:30):
We will take the first change and then we'll say we'll wait 10 minutes, we'll see its effect before we make the next change. If we have conflict, we'll back 'em both off. So absolutely, and I think a lot of my colleagues mentioned the implementation of these when we start with the use case working our way back to the solution, it gets very much into the nitty gritty about what we're going to change and what are the guardrails we put around it and if we do have conflict or conflict, it was explicit policy of how we deal with that. And that was well thought out and that was the only way it worked. We actually had to have those guardrails specifically program and says do not do this if this condition happens. That's the only way it's going to work.

Iain Milligan, VodafoneThree (00:50:09):
And I'd say it's an exciting real world example and it's one we work on with your colleagues Moran network between Vodafone and VM O2 already complex mocking now enabled on 8,000 sites, huawe, Samsung, Erickson, Nokia, and possible to do manually automated it 8,000 sites done in six months. And that is through about 10,000 use cases that fundamentally been tested, policy defined, rules defined and then excellent execution. It is all about the policy, the control, knowing your environment, understand, again, I can't emphasize more understanding the technology you're playing with and getting the right people working on it, but it is doable then the complexity is part of the fun of it I'd say. But it's fundamentally, it's achievable and doable. You need obviously the right tool sets to go at the pace that you need to go at, but it's policy driven, it's the control, your testing, your validation and doing it in steps and it makes a difference.

Ray Le Maistre, TelecomTV (00:51:12):
And is that fed back into the industry then? Is everybody learning from this?

Iain Milligan, VodafoneThree (00:51:17):
I think they soon will do. I think what we've achieved is the first mocking over on multi-vendor network ever. And that's feeding into product development with our vendors and including our tooling vendors as well in terms of the lessons learned on it. And it's something which sets a really good standard if it is possible and you need to do it, it'd been possible to do the volume of what we've done in six months with the customer benefits and the complexity of the network without being able to do it in an automated manner.

Ray Le Maistre, TelecomTV (00:51:46):
Okay, great. Any other comments from Yeah, Timo?

Timo Jokiaho, SUSE (00:51:50):
Yeah, so I agree what Sherif was saying about the complexity of OR alliance for example or architecture or alliance and to an extent or extreme case, if you look at the OR alliance architectural diagram, there can be 10 technology vendors in one RAN implementation and I think that's crazy. Two complex, but that's what we do. It's life unfortunately that's what we do. Now, what I wanted to say is that industry, we as telco industry, we need way more collaboration still and we need to be more trust each other more than we have used to. It's been a long journey already, but we need to still deepen the collaboration and trust between the vendors and we need to be able, I know it's going to be difficult, but we need to be able to even share some maybe sensitive information of the technology components to make the integration more seamless. The collaboration and sharing information is given for open source companies and open source people. Of course it's not easy even in that space, but that's what we do and we should increase the trust and collaboration between all the technology vendors in the networks.

Diego R Lopez, Telefónica (00:53:26):
Does a brief mention to something that we have started to do in the ITF with other colleagues that is about defining to identify whether you have a conflict and whether the conflict lies when you need it is a common language or a common set of concepts across the whole thing. And this has a name has been around for a long time now. It is being, well, taking some more momentum that is about ontologies. We're working on that on how you can build ontologists that are common and can allow us to share this. And this is an essential work starting if you want to join, you're more than welcome.

Prashant Agarwal, Intel (00:54:13):
I think I agree what my fellow panelist staff said, but just want to highlight maybe restarting. When we design we have to be open standard designed and we need to focus about multi-vendor interoperability because we can't just see the automation on its own. And as a question asked that there are various domain like RAN and telcon, those things. And there are a lot of other initiatives going on like this hardware, software, disaggregation. There are also multi-vendor things come into place. So I think what we need to make sure when we are designing, so we are using the open standards, we are not just binding to a particular vendor and then from the starting from the design, if we put all this principle in place, it's much easier to implement the new features and do if you have to change intent based changes.

Ray Le Maistre, TelecomTV (00:55:02):
Alright, thank you. Okay. We're going to come to, I think we have a question over here. Yep. If you could say who you are again and what your question is.

Salim Khodri, SUSE (00:55:09):
Yeah. Hi everyone. I'm REM from suse. My question is how far the telco can move toward autonomous networks while staying compliant with the European AI Act regulation. Especially when it comes down to traceability and transparency.

Diego R Lopez, Telefónica (00:55:31):
That's precisely what we're trying to build. When I was talking about this trustworthy autonomous networks, we are trying to build a consistent set of evidence that can be used to analyze and to provide full accountability of what's happening. We are trying to do so well. I believe it's something that is frankly doable. There is something that probably is more risky when it comes to this, this about that things networks are identified as critical infrastructures than the AI or whatever in the algorithms that are around, are identified as well as critical mechanisms because that would pose some additional requirements on those systems that would be difficult to achieve. And there is a struggle between need for fair regulation and well the risk of our regulation. That is something that, well we are, we have been suffering and we're still suffering in the industry.

Fernando Rionegro, Nokia (00:56:49):
So one of the key pillars of autonomous networks is precisely observability. So having being able to extract the information in a transparent way, in an organized way, have the data from a single point of contact, curated the story and then shared with the rest of the systems has already been thought as part of internal part of the autonomous networks journey. And it's not by chance, it's because we have to learn in the moment that we have that we are failing one of these risky use cases that we were commenting before. How we can learn, how can we improve? It has to be with very clear and objective data. And that data has to have points that are being able to be audited by human in order to bring evolution. So this is not by chance that the observability is an integral part of this. If you have a bad data or a pure data, the outcome of the application of artificial intelligence or any algorithms will be bad or not optimal. So the results could not be there is not just only linked with regulation itself, but it's linked with the proper outcome of the autonomous network application.

Ray Le Maistre, TelecomTV (00:58:05):
Okay. Thank you. Okay. And also I see there's a number of hands up. My colleagues at the back who can see better, who's got their hands up, we'll bring the microphone to you, but please do that on the audit as much as you can. And then my colleagues will be able to see. But next question over here somewhere you can say who you are. Yeah,

Sam Iskander, Dell Technologies (00:58:26):
Yeah. This is Sam from Dell Technologies, I guess you guys, the panelists, we have a mix of two operators and the rest are vendors and vendor as well. Now we see around autonomous network, we have the standard bodies from GSMA and TM four trying to help where operators is part of that and how to reach autonomous level four. GSMA have been also very thoughtful by also sponsoring the benchmark of different LLM. But the challenge is the moment that operator is trying to save CapEx and opex and having a better efficiency, but at the same time monetize and go up in the value chain. If they can monetize the gene AI vendors like everybody just trying to sell more from both sides. What is the sweet spot that you guys believe we all win?

Iain Milligan, VodafoneThree (00:59:15):
Very easy one. Put some skin in the game,

Diego R Lopez, Telefónica (00:59:18):
Lower prices from the vendors. That's a sweet spot.

Fernando Rionegro, Nokia (00:59:22):
Definitely. No, actually it is a very, very interesting question. And I think that win is for both because the cost that we vendors have in the moment that we don't implement autonomy in the network is huge, is a huge cost in services. So the autonomy is also bringing a reduction in cost and probably distributing the value chain in a different way. Maybe there is no growth, but having the same level of investment with different balance of cost, it's also bringing more revenue. So for us it's not trying to sell something new, implement autonomous networks because it's good for you. Yeah, it's good for you and for me. So in this case, I think it's already a win-win in the way that this is thought.

Iain Milligan, VodafoneThree (01:00:08):
I know I was being flippant on this again and the game, but a very important point is we completed the communicated we had done like our RAN equipment deals with Noia Ericsson, we've also looked at our new core networks. It wasn't a procurement exercise for the sake of a procurement exercise, it was an actual proper partnership deal. And within 20 weeks we got in, got new core network, new transport, optical ran. And it was a matter of we need to hit this at this pace. If you help us, what we've achieved by in the UK getting the merger approved, going from four to three, if we can then deliver it to the obligations we've legally done, you're helping consolidation of the market to be more effective for you as a vendor. You wouldn't be able to sell more if you help us deliver this, if you actually help us do it efficiently, effectively achieve our synergies. So you put a foot forward, we put a foot forward, we balance it, and it's a proper partnership. It's not just a beating with a big stick to the lowest unitary cost that doesn't work. And if it does, it's the exact problem. But you're going to describe. So it's how we get into that mentality of proper partnership. And again, we're just on that journey, but already we're bearing good fruits.

Ray Le Maistre, TelecomTV (01:01:20):
I guess the challenge with autonomous networks is that nobody really knows how this is going to play out. So does that require a different kind of relationship over time?

Iain Milligan, VodafoneThree (01:01:31):
Time's the important part. So again, you always hear about three to five year deals. That doesn't work in our industry. I know that pain from my previous role in three, we're talking about eight year deals. Investments long-term, whether it's with civil companies who are helping us build it to the equipment vendors and say, long-term investment you put in, we put in, if we deliver, it'll be success. So it is a commitment on both sides. And again, it's that basis is once we deliver, we get our revenue, you get paid, you get success as well. And the quicker we deliver then we a reference case, you can use this. And I think that is a very important part of the partnership.

Diego R Lopez, Telefónica (01:02:10):
I would say that the relationship has been quite dynamic for quite a long time. We started with probably specific devices built per operator in the good old days of the monopolies. And now we have different, because we have different technologies, we have different ways of approaching the technologies and that implies that the relationships, the relationships are going to change. We cannot find a sweet spot. The sweet spot is going to be moving according to, on the one hand, the evolution of technology, the evolution of the market, and well new opportunities that can arise. So it's difficult to say there is a one single spot where we should be. It will be, it will be and it should be, I would say a moving target.

Brandon Larson, Mavenir (01:02:59):
Okay. And Brandon? Yeah, and I'll bring up another dynamic to this. I think the question is who is the vendor? When we get into ai, that's a little bit different because we have AI groups and our customers and they're like ER. And some of us, they're like, you guys are plumbers. You do that type of stuff. We want to do the ai. They go knock on an open AI store. Can we do that hosted and start bringing these things together? So we see in the marketplace in developing is like do we use an open AI hosted model? Do we bring it in network? If we bring in network, we got to work with cloud providers to bring that infrastructure in. When we look at the cost performance of models, you can use these big generic LLMs, billions, trillions of parameters, but they are costly and they are overkill.

(01:03:46):
And what we found is that for a lot of these use cases, even bringing in smaller models, and I think you mentioned it's even not LLMs. Do you use a random forest? Do you use A DQM? Do you use other types of technology? You can actually run on a CPU architecture on premise. So when we look at this ecosystem, it's really getting crazy in terms of what's the model we use? Is it on premise, is it cloud? And it's all going to start from that solution and working its way back. So even as a vendor, we're now in competition with the open AI world. We have AI startups knocking at the door saying, we have these wonderful technologies, but we don't know the telco plumbing side. We need to partner with you on doing that. So then there's starting collaboration there. So I do think it's going to be a shake out pick time to shake out a little bit in terms of how all these different players are going to add value into that ecosystem or development. And I guess there's more

Ray Le Maistre, TelecomTV (01:04:37):
Unknown, or not unknown, but companies previously not in the telecom space that are starting to come in now with a RAN apps that would run on a RAN intelligence cloud on a Rick. I guess that changes the dynamic there as well and new partnerships have to be made. How do those small companies get a piece of that and what testing is there to make sure that they are not a mini Trojan horse in themselves? It's an open question.

Brandon Larson, Mavenir (01:05:17):
You want to take that or

(01:05:19):
No, I could jump in here. That was actually one of the promises of O brand, right? One was supply chain diversity solving the app problem. The other was opened up the ecosystem. So we did say like Intel Flex ran and we saw Red Hat getting involved and AWS and everybody kind of wanted to be a piece of the solution to coming back in. But what I've always said is that all these different companies that come from the outside, whether it be a cloud provider or a provider, one of the things they soon learned is that you need the telco experience. I've always said we started solving problems and we just put a data scientist at it, they could never figure it out. They needed to have the telco guys that understood the problem, what data was relevant and if the answer was right. And so these guys that come up with these great ideas, they're always going to need some help from an NEP on how to get this solution from a concept or what they do into that larger macro ecosystem where you have to be able to insert it into the telco operational, I call it the bloodstream, which is hard to do.

Iain Milligan, VodafoneThree (01:06:16):
Can I give a very boring answer?

Brandon Larson, Mavenir (01:06:18):
Yeah,

Ray Le Maistre, TelecomTV (01:06:18):
Absolutely. Please.

Iain Milligan, VodafoneThree (01:06:20):
It's the benefit and the downside of the telecom security act. Fundamentally these types of vendors, they provide network oversight functions. If you as a provide a network oversight function, you can't demonstrate certain levels of product maturity, company maturity, capability to scale. These things unfortunately just don't get in the door. So there's an element of if you can prove that, then you should be good enough to at least get testing and trials within the UK in any case, if you do, then unfortunately there is a barrier to entry for the good reasons, the bad reason, but it's the reality upside downside of the TSA.

Ray Le Maistre, TelecomTV (01:06:54):
Yep. Okay. Absolutely. Now where is the,

Beth Cohen, Luth Computer (01:06:58):
Okay. Hi. So I'm just going to throw a little mud in this whole thing. We've been automating and going toward autonomous networks for probably decades at this point. I dunno, AI machine learning seems to be kind of the better way to approach it. I think that we are mostly there. We're probably 80 to 90% there in many of the telcos already. So why are we still discussing it 20 years later?

Diego R Lopez, Telefónica (01:07:36):
Well, because it's an endless journey I guess. I mean, when you impressions that when you make autonomous or you automate a particular process, then you have the opportunity of including that in a bigger, more complex process that can deliver higher value or can address other requirements. And then you need to automate that. And this, I mean sort of a spiral that is growing and growing and growing. I saw it like this. So I don't believe that we will see a moment, then we'll say, no, we have achieved whatever the level is. I don't have in my mind, for my whole network, for all my services, there will be always a next goal to achieve. And probably in 10 years, 20 years, 100 years time, people will keep talking about autonomous whatever. And that's human struggle, I would say

Fernando Rionegro, Nokia (01:08:36):
Yes. And also because the ingredients in order to apply autonomy in the scale that we are discussing today, were not there for two years ago. So there was no cloud activity in the network. So in order to implement automation, you need certain elements as for instance, APIs. Yeah, we have been discussing about APIs for a long time, but APIs develop in a way that can be consumed by internal developers or external developers or to have really the network functions with a maturity in a cloud that is able to be consumed, be handling different angle. So these elements were not there. It doesn't mean, again, I fully agree with you that it will end up because now we have the ingredients and in five years everything is solved, but there are certain conditions that were not met before in my opinion.

Ray Le Maistre, TelecomTV (01:09:21):
Okay. Quick comment from PR and then we'll come to one final question that I'll few minutes

Prashant Agarwal, Intel (01:09:25):
From my perspective, I can also say that it's seen now we are bringing more and more processing power. So maybe initially when you started with those machine learning, there were not enough juice on the hardware to achieve maybe some automation. Now we think we can run a little bit bigger model. Also we can achieve more. So it's always like you have more power coming on with the CPU generations and remote processing power. You think what can you do? Maybe you can run better models here. You can optimize the field better and I think it'll continue with the in future. Also, we have remote processing powers. This will always, in evaluations, you cannot even say it, okay, we have achieved now you have achieved what is possible now with this hardware, what we have, how much power we have, how much memory we have. But it'll keep on evolving.

Ray Le Maistre, TelecomTV (01:10:11):
So we have three minutes left for a question and some very quick answers. So

Robert Curran, Appledore Research (01:10:16):
Hi Robert Curran from Apple Door Research. There's been quite a number of interesting announcements from, for example, Google on Microsoft and Azure talking about their own autonomous frameworks. And I'm just interested to get a view from the panel about the distinction between autonomous networks in a fixed line context and a high growth context, which is their context and a mobile context. If you look at something like Rakuten appreciate different kind of network, but they're running their network with about a 20th of the staff that a commission mobile network takes. So just like to get a sense from you, I know Diego mentioned the kind of incremental never ending journey, but where do we end up with a fractional number of people to run a network? And whether that's different between a fixed line and a mobile context, what kind of saving ultimately if you just take opex, what's the goal here? 10%, 20%, 80%. I mean it's got to be worth something if it's going to be that kind of number

Brandon Larson, Mavenir (01:11:16):
That I'm going to start and flip that a little bit. You're right. A lot of the autonomous networking, a lot of people are always thinking about the opex side of the house. The other side of the house is how about we make some more money and bring that into telecom. So what's really come up recently is conversational ai, which is how can we put AI into communications today? Communications is connectivity play, right? We connect people to people, people to business, business to business. But that's commodity. Very valuable. But it's commoditized. What we've seen now is the ability to say we can use AI to run productivity like translation use cases, AI agents answering calls. We've started talking to product groups that are very, very excited about leveraging AI to now add value for the first time into the networks and start generating revenue and they can actually start adding value added services, which have kind of alluded them and all these different technologies.

(01:12:08):
So I think that's where we're really bullish on is how do we use this not just to continue to deliver the same services and cut cost people right? Race to the bottom on the connectivity commodity play. But how do we actually start flipping the script and start generating more value? Like all the other people that are using ai like a Spotify, like a Netflix, they use AI to generate money. We want to bring that into Telco. So I'll start there and then maybe the cost cutting. Well, we've got a minute left. Anybody want to address

Ray Le Maistre, TelecomTV (01:12:39):
The

Diego R Lopez, Telefónica (01:12:40):
Simply for sure. Right now the driver is about reducing costs, but you never, and this is something that we are trying to force, for example, the use of these data infrastructures to generate more business or most business opportunities. And there is no reason to believe that this is something that is impossible. Things are commoditized and we get very used. Copper hierarchical, all telephone system was totally commoditized till someone decided to put a battery and use the radio to connect. And then we started with the mobile frenzy. This is something that can happen at any time and we are obliged and it's our part of our professional obligation to keep trying these opportunities till we find another one.

Iain Milligan, VodafoneThree (01:13:36):
There's a really good chart, which I seen a couple of weeks ago, and it showed the top five companies investing in AI in the world and it showed revenue growth of the usual stuff. Then it showed staff costs in headcount and it wasn't going down. It's flat. And that for me is the real story on the basis that they're getting far more bang for their buck than there. And I think that's the story and it's around changing the skill sets to leverage that. So they take the boring stuff off the table and actually monetize more. Because the thing that I find interesting when people talk about six G, there's so much still to get out of 5G and see when you actually leverage the technologies around to build more products, more monetization, more capabilities, that's where the money should be going. So I see it less about is cost avoidance and actually repurposing the value to get better out of your capital investment. Again, that's how I see

Ray Le Maistre, TelecomTV (01:14:27):
It. And that's a great point on which to end this discussion because incredibly we are at the end of our hour and five minutes and we have to come back to the motion and we have to find out how the room is going to vote. Is it going to be another 50 50 split? So here's a reminder of the motion network automation is a Trojan horse promising efficiency, but delivering risk. Are you voting for or against? It's a toughie, but let's see. Remember, if you're voting for point the green side to me, if you are voting against the red side to me, let's see it everybody. It's a lot of thought going on here.

Graham Wilde, Three Group Solutions (01:15:05):
Oh no, it's easy.

Ray Le Maistre, TelecomTV (01:15:06):
Oh, I think this is by far the clearest vote of the day it's against. Oh, it's the clearest one of the day. Fantastic panel. Great questions. Let's give him a round of applause.