Sean McManus, TelecomTV (00:05):
Hi, my name is Sean McManus from Telecom tv. I'm at the Open Networking and Edge Summit, and I'm joined now by Ranny Haiby from the Linux Foundation. Thank you for joining me.

Ranny Haiby, The Linux Foundation (00:14):
Yeah, thanks. Hi.

Sean McManus, TelecomTV (00:15):
We are hearing that Agentic AI is a really hot topic at the moment. Tell us a bit about what it means for telecom operators and how it can help them to run the network.

Ranny Haiby, The Linux Foundation (00:25):
Yeah, so it's a good question. I must preface by saying that we're all still trying to wrap our mind around it and trying to keep up with the base of hype in the AI world. But I think to me, when I hear agentic ai, the difference between other types of AI is AI that can actually take actions and do things automatically, and that brings us closer to what we like to call autonomous networks. So when we think about egen ai, we imagine things like actions that are currently being taken by humans, and these humans may already be using some sort of AI to kind of get recommendations and ask questions and get answers. But Egen AI takes it to the next level where the human is pretty much taking out of the loop and things that were traditionally done by humans, like changing configurations, restarting systems, tweaking the network.

(01:25):
Think about all the stuff that's traditionally was done at the knock, the network operation center. A lot of that can be replaced by AI with the aid of agents. So these agents can act as replace the humans and take the actions and make sure the network is run properly and even get the network to run better than the way it was run by humans because they have all the knowledge of the AI at their disposal. So we're hopeful that that would bring new opportunities to network and make the way networks are run and operated and maintain much more efficient because the AI can do things much more efficiently than a human.

Sean McManus, TelecomTV (02:14):
But what are the risks of using AI in the network?

Ranny Haiby, The Linux Foundation (02:18):
So with any kind of technology that consumes data, there are many risks related to how the data is flowing and what type of data is flowing where that's kind of one type of risk of controlling your data and making sure what you're sending to the AI is not sensitive. For example, if there are privately identifiable information that exists in your network and you want to use some AI algorithm, you need to make sure that the data you're sending, especially if it's an algorithm that runs on a public cloud, for example, you need to make sure that the data going out does not include this very sensitive information. So that's on the data side. The other aspect of risk or dangers in AI is how do you make sure that the response generated by the AI is really trustworthy? So there are mechanisms for validating the responses, and everybody's by now, I think, familiar with the concept of hallucinations.

(03:20):
So that's one thing you want to weed out without the hallucinations of these AI models. So we just announced a project called ous that is going to deal with many of these challenges, which are common, not just telecom, but I think we are launching it with specifically telecom use cases in mind. And Salus is going to be a set of tools and methodologies that are consumable by API that can be applied when you use AI to make sure A, your data is sent securely and not containing this sensitive information. And also that the model return response that is trustworthy and you can act upon. So these are kind of the two main focus areas for this new project, and we hope it's an open source project, so we hope it will be a community around it that will help grow it and add more functionality.

Sean McManus, TelecomTV (04:15):
Great. So that's bringing guardrails effectively to the network.

Ranny Haiby, The Linux Foundation (04:18):
Yeah, a lot of it is guardrails, but a few other things like validating models and scoring models and benchmarking models for their accuracy and things like that.

Sean McManus, TelecomTV (04:29):
Now, you've also announced here Essedum as another project. Tell us a bit about the problems for operators that aims to solve.

Ranny Haiby, The Linux Foundation (04:36):
Yeah, so seldom is aiming to address, I would say the common parts of building AI applications and solutions for telecoms. I think every organization that is building or developing an application usually deals with some common layers, like how to access the data, how to do some pre-processing, how to plug in the models. And this is not where the differentiation is. So everybody wants to differentiate, be ahead of the pack and innovate, but maybe 80, 90% of the work is just very common parts that are repeated over and over again. So ESOM is attempting to make that much more efficient. We had a generous code contribution from one of our member companies, Infosys, that did that internally and felt exactly that. Why should we do it over and over again where we can have a common thing that we can reuse in the form of an open source project. So that's the goal of Essedum, to provide these common layers that everybody needs when they develop AI applications or services for their network while still letting them innovate and differentiate on the higher layer, the application layer. So hopefully that will be an accelerator for adopting AI more easily for telcom applications.

Sean McManus, TelecomTV (06:05):
Tell us about the timescales for these two projects.

Ranny Haiby, The Linux Foundation (06:08):
So Salus is pretty much available today. I think this seed code is being transitioned as we speak to Linux Foundation hosted GitHub repo. So that should be available to anyone interested starting probably tomorrow For Essedum, we're taking a more cautious approach. We're trying to figure out exactly what we need to bring and what are exactly these building blocks of this framework that we need. So we have some potential contributions or code repos sitting at Infosys, but we're not sure which of them needs to come over. We also have started a technical steering committee from the project, which includes folks from Infosys, but also from Red Hat and hopefully other companies. And they're kind of bringing a broader perspective of what might work best for the community and what components we need. So we expect that the next following weeks or months will be dedicated to kind of architecting what needs to come there, and within two to three months, we'll probably have the first significant drop of code that people can actually start working with. But in the meantime, we welcome anyone to join this conversation that we're managing in the technical steering committee. It's open, everything that the Links Foundation does is in the open. So everybody is welcome to join the mailing list and express their opinion and contribute to that discussion of what needs to be the scope of the project. And again, the expected ETA for the actual code is about something like 10 weeks.

Sean McManus, TelecomTV (07:49):
That's brilliant. Thank you so much.

Ranny Haiby, The Linux Foundation (07:51):
Yeah, thank you.