There is a significant amount of activity across industry standardization bodies to look at what is needed and how to implement AI, particularly in operational network environments, through groups ranging from the AI-RAN Alliance, the O-RAN Alliance, the Telecom Infra Project (TIP), 3GPP and others.
So what functions is AI likely to support in telecom networks? Panelists on a recent RCR Wireless News webinar outlined some possibilities.
Service assurance. AI capabilities are being explored for more intelligent service assurance, noted Chris Murphy, EMEA CTO for Viavi Solutions: Making sure that the network is operational and that any impairments or outages are dealt with quickly. That means AI being able to recognize that something is amiss, figure out what the problem is and what the response should be to fix it, including when that requires, say, a human being dispatched to the piece or a piece of infrastructure being replaced—and mitigating service impacts while that fix is being made, Murphy said.
Optimization of security. “When you go into a big telco today, they can have hundreds and thousands of security overlapping security policies,” said Stephen Douglas, head of market strategy for Spirent Communications—and in many cases, he added, “everybody’s just scared to touch them.” AI is being tasked with helping to clean up and optimize overlapping security policies, also also for things like co-pilots to help with dynamic configuration of security policies, he added. Additionally, there is testing going on now to demonstrate AI’s ability to identify threat traffic, and how AI could be used to generate attacks to test cyber defenses.
A more responsive network management. “If you have the right AIops tools, you will be able to mitigate issues faster, close loops on issues faster, have predictions on your network. So not just after the fact, but also provide predictions to your network, add co-pilots to your network to assist engineers to troubleshoot and also allow better capacity planning and so on,” said Sagie Fanish, senior director of AI infrastructure for DriveNets.
Supporting flexibility. While there are discrete tasks like those just mentioned, Murphy said that there is also an overall degree of network flexibility that AI is expected to support. “The network needs to be flexible. The network needs to be able to respond to changing demands—not just the impairments and the outages, but, people are going to change how they use the network. New services will come and go, people will change how they move around and where they want to get the services. So being to do that using AI is going to make the service continuity complete, and the SLAs … able to be met,” he explained.
Testing and assuring AI in the telecom network
So how will the industry ensure that AI is able to successfully and reliably perform those roles? Testing, of course. But that looks quite different when it comes to something as fluid as AI, and advanced simulation and emulation—particularly, the use of digital twins—is expected to be key to putting AI through its paces in the telecom network.
“You’re moving from something which is a well-defined component, which has well-defined interfaces and you can say, is it conformant? Is it interoperable with other components from the same vendor or different vendors? Can you scale it up? Does it still perform when it’s loaded?” All of those questions now need to be answered about any given AI model put to use within telecom networks, said Murphy. “[AI] can change, and should change, as the stimulus on it changes. So being able to deal with those complexities in the lab, and as you deploy the network and in the field, is a critical piece of being able to be successful in this world.”
And, he added, testing AI isn’t just about ensuring reliable and repeatable day-to-day performance—testing also has to explore what AI does when it encounters the unexpected and out-of-the-ordinary, and whether it reacts appropriately. “You want to be able to make sure that you can handle the unexpected handle those corner cases, which don’t often arise— and may never arise,” Murphy said. “But there are going to be some corner cases at some point, which are unusual situations which you need to be able to deal with autonomously.
“Being able to create those scenarios as early as possible is key to building those solutions which are resilient and going to deliver those networks in the next generation,” he added.
The type of AI model being used matters as well, and that is an area which is changing rapidly, particularly when it comes to genAI—both in terms of which models will be used within telecom networks, and which models telecom networks will have to support use of.
“From the telecom point of view, the small language models are starting to come to the fore,” Douglas said, adding that such models can “be very demand specific, or very niche to specific tasks they want to do at the device level.”
He expects to see a mix of various types of AI, including Large Language Models (LLMs) and Small Language Models (SLMs), and those uses will have to be reflected in a distributed networking architecture.”It’s not just in the data center, it’s going to be distributed in our regional data centers, our edges, right onto the devices,” he said. ” I think we’re going to have to deal with that in terms of what that means in terms of traffic performance behavior, and how that all gets tested.”
“When you test infrastructure for AI, try to even simulate like a [mixture of experts or MoE] type of model like Grok or DeepSeek or multi-node inference, like in DeepSeek, to try to see if your network caters to those types of models—because it’ll only get more complex,” Fanish recommended. “It will get to the trillion-parameter models. … So your network needs to be ready.”
What do telecom companies need to understand about AI right now? Fanish said first of all, “They need to understand that the revolution is here.”
He added: “AI is happening. You need to adjust your infrastructure for it. The telco companies specifically have the right … operational power to leverage this AI boom and build their virtualized cloud environments, build their bare-metal environments, play the cloud game now, play the neo-cloud game. But that requires adjusting the infrastructure, leveraging your real estate … smartly, and understand that AI has its own implications and demands. So just try to adjust to it. It’s not the typical services that we know and love from the telco world.”
