Examples of how operators are using AI today include network anomaly detection and root cause analysis
Telecom companies have been using artificial intelligence (AI) and machine learning (ML) in their operations for years. However, the current telco environment sets the stage for further innovation around AIOps, or Artificial Intelligence for IT Operations. That’s because as 5G continues to evolve, it also continues to become more complex. Virtualization and disaggregation are happening in tandem with deployment of network workloads in hybrid cloud environments. As a result, configuring, provisioning and assuring networks through manual — or even the standard automation strategies that telcos have been relying on for years — is no longer possible. Panelists at the Telco Cloud and Edge Forum spoke to this transformation, addressing key questions like how do advancements in generative AI (GenAI) fit into the conversation, how are operators using AI in their operations and what challenges persist?
In addition to complexity, Chris Murphy, regional CTO for EMEA at VIAVI Solutions, told the audience that the availability of reliable data is also a challenge in modern networks. “We have more disaggregation and more interfaces that we can hope to get data out [of] and understand how the network is performing, root cause problems, and understand how we can resolve those,” he said, but added that the data must be collected, harmonized, cleaned and correlated.
“Different network layers, different parts of the network. It’s not always easy to bring the data together to come up with a coherent view of what’s going on so we can perform the advanced analytics and autonomous decisions that need to be made. But, in terms of the opportunities, I think the opportunity and the imperative really that we have to deliver is the intent-based, end-to-end automation, which is I think what we’re ultimately aiming for,” he continued.
Murphy shared that Viavi is starting to see consumer-level AI, like chatGPT, being merged into operational networks. “There’s a clear trajectory that the industry is moving towards,” he said, adding also that things like anomaly detection, root cause analysis, opening trouble tickets automatically are some specific examples of where AI is being used today.
For its part, AT&T is already using OpenAI’s chatGPT for an internal application called Ask AT&T. Released in June, the application helps coders and software developers become more productive and translates customer and employee documentation from English to other languages, and even simplifies that same documentation and make it easier to use. Future use cases for GenAI, according to AT&T, include upgrading legacy software code and environments; making its care representatives even more effective at supporting customers; and giving employees quick and simple answers to HR questions.
More broadly, though, AT&T is using AI as a sort of co-pilot, the carrier’s Network Chief Technology Officer Ajay Rajkumar told event attendees. AI currently helps the operator perform certain automated network optimizations, such as acting as an additional quality agent and providing recommendations for network parameter changes. The reason AT&T is taking the copilot approach to AI in its operations, rather than allowing it to fly solo, is because the carrier is still approaching AI — and GenAI, in particular — with caution.
“If there are biases or hallucinations — as is a problem with typically generative AI — And I’m drawing a distinction because it is generating new ideas from either what it has seen or what it has learned … those need to be really curtailed,” said Rajkumar. “You could just not … say that [a] chatGPT-like structure could be used in [an] operational network … The cost of small mistakes or one singular mistake can be huge. These are critical networks … Once there is a reliability, we may be able to move forward with the level of automation that we’re hoping we’d be able to get.”
Right now, though, he said using AI as a co-pilot to assist operations with root cause analysis at an early stage or in real time is a very real possibility. But, for GenAI to be truly ready for operational networks on a wider scale, Rajkumar argued that significant and specific foundational model training must be performed. “Not generic network data, but … very specific network data for a given operator or a circumstance,” he clarified.
A final consideration in this discussion is, of course, security, as there are always concerns around where the data is coming from and who can view said data. However, Murphy noted recent developments in the industry, such as secure enclaves two sets of data owned by different entities can be accessed by the AI system, but the data itself cannot be shared. “They can both benefit from bringing their own data, which has personal identifiable information and sensitive commercial information in it. Bring it together into an enclave and then ask it a question and see what the answer is, and it hides the data from each of the parties who’s showing it … maybe there’s a role for that sort of thing down the road,” he said.
So, perhaps there are still a few kinks to be worked out. However, both panelists were optimistic about AI’s future role in network management and optimization: “We’ll have beautiful end-to-end automation of autonomous networks,” Murphy predicted.
Click here for more on how telco AI supports network automation and 5G monetization.
