AT&T is using a generative AI system to help control network operations, a move that the carrier said is a step toward greater network autonomy.

Raj Savoor, VP of network analytics and automation at AT&T, explained in a blog post that the AT&T Geo Modeler system is active network wide. It uses synthetic data and a network foundation model to simulate and predict network coverage “under dynamically changing network and environmental conditions.”

“Using ray tracing, the Geo Modeler simulates radio transmissions in complex geospatial environments and integrates with multiple internal systems to help automate network decisions and change how the network is configured to improve overall performance and help manage connectivity nationwide, even during extreme events or as the seasons change,” Savoor wrote.

If one of those “events” were to impact a specific cell tower, the Geo Modeler platform can determine how nearby towers can be optimized and used to compensate for coverage gaps.

“It can also recommend everyday changes to enhance network performance and get the best out of our resources, especially when needing to adapt to changes in the environment or when planning network coverage to support a major public event,” Savoor added. “These adjustments are fully automated and can be as simple as changing antenna angles.”

Savoor did note that the network feedback data is “closely monitored by employees to ensure the AI provides accurate recommendations.”

AI for network autonomy

The Geo Modeler is being touted as a “step toward network autonomy,” something that AT&T is increasingly using AI to boost.

AT&T CTO Jeremy Legg recently told attendees at an investor conference that the carrier was seeing significant progress in integrating AI into its operations, noting “we’ve got over 600 models in production across our infrastructure.”

However, the executive said the carrier’s approach toward feeding these models is different from the traditional “Reddit” model used by a lot of companies to train their applications.

“[Reddit doesn’t] have any of our data, they don't have our network data, they don't have our customer data,” Legg said. “So as we've begun to ingest that data into these LLMs [large-language models], we are tuning it for us, and the insights and the data science that we're using are very specific to the kinds of problems that we have.”

One of those is network operations, and “how do we deal with RF (radio frequency) and spectral efficiency.”

“How do we use that in order to remediate issues in the network itself faster,” Legg said. “An example is we've loaded every trouble ticket in the company into our generative and agentic AI frameworks. It now recommends what the fixes are and can write the code to do the fix, and should we just go ahead and automate that. We're still human-in-the-loop there, but these are major, major changes.”

Analysts have pointed to the benefits AI can bring to the deployment and management of cloud-based open RAN networks, which are more complex orchestration challenges due to the disaggregated multivendor ecosystem. The use of AI could help close performance gaps for open RAN architectures compared with legacy RAN models.

AT&T has a long history of looking toward AI to improve network performance. One of its network executives spoke of that opportunity at an industry event nearly 10 years ago.

Brian Daly, director of core network and government regulatory for AT&T, told attendees at the 5G North America conference in late 2016 that the company is looking at AI as a way to operate its network more efficiently by using it to make decisions that currently might require human interaction today. “We see AI combined with SDN and NFV as a way to provide us with efficiencies that may not exist today,” Daly said at that time.