Guy Daniels, TelecomTV (00:24):
Hello, you are watching the AI Native Telco Summit part of our year-Round DSP Leaders Coverage. I'm Guy Daniels and for our first discussion today, we are going to explore where telcos should focus their gen AI activities. Given that GENAI has the potential to impact a wide range of disciplines, how should telcos prioritize use cases? What are the business objectives and what is required to ensure successful implementation? Well, I'm delighted to say that joining me on the program are Antonio Guzman, Telefonica delivery Director at Telefonica Digital Innovation, Beth Cohen, SDN Network product Strategy with Verizon Business Group. Paul Miller, the Chief Technology Officer for Wind River and Martin Halstead who is Senior Distinguished Technologist at Aruba Telco Solutions HPE. Hello everyone. It is so good to see you. Thanks very much for taking part in our show. First of all, I'd like to ask you, what is the telco view of generative ai? How and where did Telco see it being used within their organizations? And let's just have a brief overview for the moment because we're going to some more detail later in this discussion. Antonio, lemme come across to you first. What are your views here?

Antonio Guzmán, Telefónica (01:57):
Yes, artificial intelligence is a great tool for driving innovation in companies including telco for sure. Telcos use AI for achieve the main three objectives. First of all, improve efficiency internally and externally in terms of the private or the digital services, but also try to increase the value of existing product and services. The third objectives that is more difficult to achieve when we are speaking about gene AI is to create new digital products. In Telefonica, we have been focused on the first two objectives where we are exploring new solutions and we are in a good position to do it because we have been evolving our digital transformation and we have been creating a powerful internal platform to assure the governance of the data. You have to consider that anytime that you are creating this kind of solution based on ai, you have to do it with this control of what is happening with the privacy when it comes into these new AI visit products and services. We evaluate how the technology evolves. That is freaking quickly evolution and how the end user react to this solution. And we are a little bit kosher in that point because we understand that it's not so tightly close all the problems and issues that comes with this new technology.

Guy Daniels, TelecomTV (03:20):
Yeah, appreciate that Antonio, thanks very much for that initial overview. We'll come to Paul in a minute, but Beth, let's get another telco perspective. Where are you seeing it being used?

Beth Cohen, Verizon (03:32):
So it's a mix. We are using it heavily on the operations side, which is kind of the more obvious area that has been proven coming from the large language model successes of gen ai. We're also, I think using it to a certain extent in our marketing efforts. We're still experimenting with it in other areas, particularly around network capacity, predictive analytics and other areas like that. So more to come.

Guy Daniels, TelecomTV (04:14):
Yeah, I'm sure there will be Beth and I'm sure a lot of telcos are viewing this as experimental at the moment and trying a lot of things out. Paul, how are you seeing things from your perspective with the telcos you work with?

Paul Miller, Wind River (04:30):
Yeah, thanks guy. I think that we're seeing, as Beth mentioned, a pretty heavy effort in the operation side of the business. This all started with kind of the transformation to a cloud native telco when we started moving to a more software-defined environment, things like NFV and the core and open ran at the far edge. As we've done those things, we've also created a lot of complexity in the systems ass. A huge amount of software as an integration of a huge number of vendors coming together in the service provider networks. And AI really represents a nice way to be able to automate these things because as you get to tens of thousands of sites running all these different software stacks, the ability to operationally manage it is very challenging. And so using generative AI training and LLM to be able to command APIs for example, and get live network information and be able to manage the network in a much more automated way, really puts the grasp finally within reach to have human beings able to manage such complex networks.

Guy Daniels, TelecomTV (05:25):
Yeah, thanks very much Paul. And we heard this link during last month's cloud native telco summit and the AI side was coming out of that organically through the discussions. Well, let me look at a couple of specific cases perhaps because the use of gen AI in chatbots for call centers and customer service is well documented. It's a bit outside of our area of interest here at telecom tv. So when we look at some other areas, can I first of all perhaps focus on revenue, the all important revenue area? How can gen be used to improve BSS capabilities and is there a compelling case to help here with revenue assurance? Beth, can I come back to you?

Beth Cohen, Verizon (06:08):
Yeah, sure. I think the jury is still out on that. We are, Verizon certainly has some or is in the process of developing some customer facing potential revenue opportunities. I think our customers are looking for it or maybe it's just the buzzword of the week. It's a little hard to say. But in terms of operationally helping with those sorts of things, that's still definitely in the test and experiment stage I would say. So we're still looking, we're still hopeful that there is something there, but as we all know, gen AI is pretty much in maximum hype right now.

Guy Daniels, TelecomTV (07:00):
Yeah, it certainly is. It's top of that curve, isn't it? Well, what about some other areas? How can gen AI perhaps best support network development? And by this I mean going all the way through from network design procurement to construction and rollouts. Paul, have you got any views on this?

Paul Miller, Wind River (07:22):
Yeah, I guess I can share one interesting experiment that we ran tied to the development side of the network, we were able to actually use a gen AI in an experiment to an AI that had been trading on the code base, look at the log file information from a production network, immediately identify a fault within the log, show us in the code where the source of that problem was and then at the same time suggest a correction for that. So what we're seeing there is an incredible efficiency that might've taken days or weeks for software engineers and operational teams to root cause to now be produced in seconds or minutes. And so what that can do for network design and deployment services as well as operational services is pretty much unheard of. It's, it's pretty exciting capabilities.

Guy Daniels, TelecomTV (08:09):
Great. Thanks very much Paul. We'll come back to Beth in a moment, but Martin, I'd like to bring you in at this stage from your position, your activities and the work you are doing, have you got any views on how gene AI can be used on the network side in deployment, procurement, et cetera?

Martin Halstead, HPE (08:29):
Yeah, sure. So there's some pretty interesting work that we've been looking at from the perspective of how do you generate pretty complex configurations, templates, et cetera, for doing things like service orchestration, rollouts, et cetera. And the development work that we've been doing so far has been very promising. The idea being that you would effectively create an AI enabled middleware layer that would sit on top of existing application sets, be there across, in our case for BSS, sorry for OSS across both orchestration and assurance, and then have a centralized method of being able to set up mass configuration across multiple telco networking domains. So results are kind of early stage so far, but promising nevertheless.

Guy Daniels, TelecomTV (09:39):
Oh great. Thanks for sharing that with us Martin. Very interesting. Beth, let's come across to you as well for additional thoughts here. So

Beth Cohen, Verizon (09:47):
I want to make it clear that I think we're a little bit mixing up gen AI and machine learning and they are two different very distinct tools in the AI tool set. And gen AI works quite well with semi-structured models and machine learning works better with structured models and data sets.

(10:13):
And so I think we have had some success with fault correlations as Paul mentioned, and we have discovered Gene AI is quite good at writing code and debugging code, but that's not necessarily how it applies to a live production network, a telco size network. So that's I think where we're still struggling to incorporate that into how do we deploy it to a telco live large telco network successfully? And I've said this before, it has to work. You can't get hallucinations introduced into the network because as a telco we are expected to have pretty close student uptime for these networks. And so we have to be very risk adverse and cautious about introducing these tools to make sure that they will in fact work in the live network.

Guy Daniels, TelecomTV (11:26):
Sure, absolutely. And every vertical, ourselves included are wrestling with how to integrate this and to get it working successfully. Paul, lemme come back to you. I think you'd like to come back with some additional comments here.

Paul Miller, Wind River (11:40):
Yeah, I think Beth raised a really important point, and that is really the difference between machine learning and ai. Really,

(11:47):
Machine learning has been around for a little while and in use in telco networks, we look at things like predictive outage or event correlation or root cause analysis, things that can be run at the edge of networks to determine faults. Anomaly detection for example, is a great example. These are not ai. These are machine learning algorithms, processing data, trained on certain actions that are very helpful to the service provider. The AI use that we're seeing at least initially is more in the core of the network where you're perhaps training a large language model on an API set and using that generative capability to generate API calls and dynamically manage infrastructure or to do more simplistic things like template generation and configuration management. That's happening more at the central part of the network where learning can happen for an AI system. Learning isn't really going to happen at the far edge. That's going to be more AI inference and the execution of AI algorithms

(12:40):
at the edge of network. So it's kind of interesting to see the difference between machine learning and its use cases in ai. And AI and machine learning is something that confuses a lot of people.

Guy Daniels, TelecomTV (12:51):
Yeah. Thank you Paul. I really hope we can explore this further during the live show with some of our viewer questions when they come in for the q and a. I'm sure there's going to be quite a lot of interest there in exploring this issue further. Right. Thanks everybody. Well, let's move on to another area then network operations. How are CSPs using genai to automate network operations or are they as well as improving operational efficiency and network reliability? Beth, let's come and start with you. You've mentioned a lot of use cases so far in areas where work is ongoing or just starting. What about this particular area?

Beth Cohen, Verizon (13:33):
So network operations has been an area that we have in fact deployed gen ai and I think that's true for other telcos as well, and that's because we're able to apply successful geno models to the operational aspects of the organization. You mentioned call centers and collaborative, those types of things. So that is certainly a component and the chat bots, but also I think being able to use gen AI to look for pattern recognition and fault and fault correlation, that's a very useful way of applying the gen AI principles to operational models and to make operations more efficient. So I'd say that's the area that we have had the most success. We've also had success with Paul mentioned with templates and automating a lot of our deployment areas, although I would argue that again, it's more machine learning than gen ai, but you need both. You need both tools. So I agree that this area is the most promising and I would say the furthest along in terms of being deployed within the telco networks today.

Guy Daniels, TelecomTV (15:14):
Great encouraging news there. Thank you. Beth and Paul, what are you seeing in the network operations area?

Paul Miller, Wind River (15:22):
I do agree with Beth that I think this is the fastest growing area and they're finding a lot of use cases where this can really help operational ease of use of systems. We have a demonstration on telecom TV that we went through. We took a large language model and trained it on an API set as I mentioned a moment ago. And that allows you to take regular human questions into the network. For example, do I have any Kubernetes certificates expiring? Do I have any problems in this particular geographical area? Can you tell me what alarms are in these remote systems? And you can literally ask it like that. Just asking a natural language question of the ai. Many of your viewers may have played with chat GPT, it's exactly like doing that except you're talking to the live network. And that's the real power of generative AI is that you could enable either lower skilled workers or eliminate huge amounts of scripting and manual processes and debugging things by just talking to the system that is then interfering for you and making the API requests and taking the information back from the network and then presenting it to you.

(16:27):
And it's a tremendous labor savings and a tremendous increase in ease of use for such a complex network. So I do think those operational use cases are being identified and they'll be the earliest places of deployment for ai.

Guy Daniels, TelecomTV (16:41):
Great. Thanks Paul. And Paul, can I come back to you in that one? Is this saving time or is it guaranteeing better results?

Paul Miller, Wind River (16:49):
Well, in a classic technology answer, it's doing both, right? So imagine being able to fire a question at an AI like that and ask it a question that you've not constructed a script for. Go find something for me in the network. It can then automatically produce that answer and go get the live data and return it to you that saved you hours or days of writing scripts and testing them and automating them in the network and plugging them into backend systems because you're just asking the AI question, it's taking care of that for you. So that represents a tremendous labor savings, which then results in lower opex to run the network.

Guy Daniels, TelecomTV (17:25):
Thanks, Paul. I think we've got, Beth, do you want to come in on what Paul was saying there?

Beth Cohen, Verizon (17:31):
Yeah, sure. So I agree it's both, although I would say it's probably more on increasing the operating efficiency and less so on time saving, although we do certainly see time saving as well. But I think what's more important, particularly for the telcos is the performance enhancement IE, the higher uptime we're seeing and the quicker response to faults and fault addressing faults and correcting them. So I'd say more on the performance end of things rather than on the time-saving end of things.

Guy Daniels, TelecomTV (18:17):
Okay, great. Thank you very much. Thanks everyone. Well, we've covered a number of areas here and I'm sure we will have more viewer questions about additional areas and use cases in a live Q and a show, which we can look at. But another question from me now, GNAI has huge and wide potential, but how does a telco avoid getting carried away with all the hype we heard earlier about where it is in the hype curve and also avoid spreading its efforts and research too thinly because no one really wants to end up with abandoned proof of concepts and research projects. Martin, can I come to you? How do telcos avoid this trapper getting carried away and wasting time and effort?

Martin Halstead, HPE (19:04):
Yeah, so I think a lot of that is down to some of the telco engagements that we've been engaged with so far. It's kind of interesting that the deployments are looking at from a platform perspective. So you would go out and talk to various platform vendors. Our organization, Hewlett Packard Enterprise is one of those amongst others, plus hyperscalers, et cetera. And you would have an organization dedicated to selecting a platform. You would also then have typically the skill sets on in the telco to go and develop the applications that would sit on that. So maybe going to system integrators to get those built as a separate activity. So the danger with all of this is it's very hard to quantify a total cost of ownership and especially when at the same time you have your existing networking vendors that are inside the network that are generating AI features on their own.

(20:15):
So there's no sort of centralized approach of coordination across all of the different operations teams to be able to come up with a coherent strategy. That means that developments that are promising in one domain can be replicated or passed on to other parts of the business. So I think some issues, teething issues that hopefully over time will get addressed, especially as telcos would need to look at AI as being transformative to their business. If it's going to be transformative, then you would need more of a centralized approach as to how you would deploy it and make sure that the vendor community instead of just building features into existing products, would also align to that end strategy as well. To me, that's the only way that you can realize TCO, get around things like how do you access the data across all of those different domains within the telco and address some of the security concerns as well. And as Beth said at the beginning of this,

(21:27):
A major factor is once you have an application functioning and producing results, making sure that the business trusts it. Because if they don't, then again there is a danger that all of that development work just through human lack of trust, it could be wasted.

Guy Daniels, TelecomTV (21:52):
Thanks very much. Martin. Can I ask a follow up, Martin, is a lot of this to do specifically with telcos or is it generally with enterprises and verticals? They're still figuring out teams, reporting structures and coordination internally?

Martin Halstead, HPE (22:07):
Yeah, I don't think this is just a telco issue at all. I mean obviously arguably telcos have additional burdens in terms of regulatory requirements, the need for example data not to leave the network, et cetera. Not every enterprise has the constraints that a telco does, but I think that there's a lot of parallels in various other industry verticals as well that would mirror this problem for the telcos that the telcos have too.

Guy Daniels, TelecomTV (22:38):
Thanks very much. Martin. Beth, youd like to come in on this as well?

Beth Cohen, Verizon (22:41):
I do. So I think one area that can help, and I think it's just starting, is in the open source community because as Martin mentioned, telco data is subject to regulatory privacy requirements, quite strict. I should add. And I know that there's, at least I'm aware of at least a number of open source projects including the Anuket project, which is working on developing a anonymized data model, which will allow telcos to be able to test these algorithms. And getting back to my earlier comment about making sure that it works in the live production and trusting the results,

(23:37):
we need a set of anonymized data that we can test these algorithms on to make sure that they're working and the dataset obviously has to be big and we have to be careful about how we handle it.

Guy Daniels, TelecomTV (23:53):
Yeah, sure. And it's got to be your dataset as well, hasn't it? And your data, be careful the data you're using. Anybody else want to come in on this question about how best to do this, how to avoid wasting resources and time and effort? Paul, let's come to you.

Paul Miller, Wind River (24:13):
I think that the vendors in the telco network are going to be heavily involved in this. Obviously we tend to provide a lot of the solutions that the carriers look for, and generally they're tied to very specific problems that we're solving or that the carrier's looking for us to solve. I think that's going to be one of the things that keeps us focused and prevents us from losing time is that, or wasting and abandoning projects is these have to be very focused on an actual customer need and solving in that customer problem. An example there where we're starting to see AI happen in the 5G ran is around the SMO and the real-time, Rick, the ability to dynamically control the radio signal, perform beam forming and energy savings. These are very tangible things of great value to the service provider. And so the tools that are necessary to solve those problems, including AI, will be leveraged by the vendors to produce those types of products.

(25:03):
So I think we see the natural thing happening where the service providers pull from the industry based on their need and then the vendors will produce products focused on that.

(25:15):
We do have to be careful doing it. We have seen generative AI is and LLMs are pretty new technology. We've seen indications, for example, of hallucinations from the AI where it provides nonsensical answers. And so a lot of testing and a lot of rigorous qualification is going to have to be established before these really used at high scale within a service provider network. So they're truly trustable.

(25:38):
in the way that we want to use them.

Guy Daniels, TelecomTV (25:41):
Yeah, a lot of work there. Thank you very much Paul and Antonio, let me come across to you. This could well be our last question of this panel. So what are your thoughts about how telcos can avoid getting carried away and avoid doing unnecessary work?

Antonio Guzmán, Telefónica (25:58):
Sir, which is the practice that we have performing in Telefonica because our strategy is to launch companywide initiative terms. Are all the teams involved in the definition, not only the operation but also the digital marketing products and all the estimates inside of the company are represented in different working groups. We understand that to fight this kind of hype that the GenAI may come, I think it's important to have an internal initiative. This is complementing what my colleagues has said. It's not only a matter to define a proper strategy or how to preserve how to define a partnership strategy, but also internally

(26:39):
you have to set up not only the selection of which kind of use cases may be relevant for the company, but you also have to perform re-skilling programs to set up all the teams and try to figure out which is the best option and with which is the best usage of the technology. Because otherwise the usage and the accessibility to this kind of technology is out of control internally of the company.

(27:06):
So you have to perform a single program that fulfill all the requirements and the overall company has to be aware of which can be the risk and which can be the benefit and to perform a single proposal intent of the users of the tools.

(27:20):
Yeah.

Guy Daniels, TelecomTV (27:21):
Great. Thanks very much Antonio. A lot of these points are ones that we heard a few years ago with Cloud native wasn't with that transformation program. It looks like a lot of the same issues are kind of coming up again with the AI initiatives as well. Well, we must leave it there for this particular show. I know we will continue this debate during our live q and A show later. But for now, thank you all so much for taking part in our discussion. And if you are watching this on day two of our AI Native Telco summit, then please do send us your questions and we'll answer them in our live q and a show, which starts as usual at 4:00 PM UK time. The full schedule of programs and speakers can be found on the telecom TV website, which is where you will also find the q and a form and our poll question For now though, thank you for watching and goodbye.