AWS’s Sameer Vuyyuru on how GenAI can help telcos democratize AI

Telcos are becoming shrewder in their use of generative AI (GenAI), as they shift focus to “where they are seeing ROI, and a dropping of areas where they're not,” according to Sameer Vuyyuru, Head of WW Business Development for Telecom, AWS.

Improvement to network operations is one area where telcos see opportunity for GenAI, according to Vuyyuru.

“Network operations has been this really mystifying, esoteric area,” he says, citing increasing operational costs and a lack of a zero-touch operation as key obstacles. “It's an area ripe for transformation,” he adds.

What GenAI will help telcos tackle is “a multi-vendor, multi data-domain, with multiple operational tools and independent relational databases for different network components, all of which contribute to ineffective root cause analysis and inaccurate network topology,” explains Vuyyuru.

He believes predictive AI will continue to play a large role in managing network operations, but points to several reasons why communications service providers (CSPs) use of GenAI will increase, starting with its ease of use.

“We can't look at Gen AI as a standalone. It is about democratizing AI; it's about productivity improvements,” Vuyyuru says. “There is still a huge predictive AI component to network operations. But before, AI was limited to a set of highly trained IT players with very specific knowledge and skills, such as the ability to query SQL databases.

“Now, literally anyone with a working knowledge of IT and … an understanding of the network operations of the telco can query in natural language and find the answers they need … and initiate a root cause analysis or a remedial act.”

Telcos typically start using GenAI for network “service issues, where generative AI is amazing,” allowing the network “to be enriched with anomaly detection”, explains Vuyyuru. This can then lead to a virtuous circle of AI deployment.

“Once through the network root cause analysis, then … you go into healing actions, then service design and service modifications and service improvements. And that all flows into the network configuration [and] that end orchestration, which then again flows into services,” says Vuyyuru.

A choice of models

Another trend impacting how easily telcos can use GenAI to solve specific problems is the proliferation of language models.

“Two years ago, there were 30 models … now there are over 200,000 of all different sizes,” Vuyyuru points out. “There are models for industry, models for use case, models that are cheap, models that are expensive.”

This gives operators the opportunity to solve telco-specific business issues using industry-specific terms, local languages and their own internal business jargon, all of which make deployment faster and more accurate.

In addition, small language models can be “deployed on consumer equipment like a car or in a house, so you actually have any an AI agent, at your disposal,” explains Vuyyuru.

Rethinking data

Unlike predictive AI which draws on structured data, GenAI works with unstructured data. This makes it possible to tap into multiple data stores, rather than rely on the data sets for particular applications that only a few people master, according to Vuyyuru.

However, the use of unstructured datasets also means that telcos must reconsider how to store and access data. Indeed, many telcos are already investing in creating a scalable and more accessible data infrastructure.

“GenAI is an additional catalyst to move data infrastructure to a resilient, flexible infrastructure. And in most cases, that's about the cloud,” says Vuyyuru. “You're taking away this whole requirement to structure unstructured data, because GenAI can do that, naturally.”

Among the trends Vuyyuru is seeing is the emergence of ‘data puddles’ as opposed to data lakes, which enable telcos to “act on data where it is, and whatever format it's in, as long as you can query the data.”

Partnering with software vendors

At the same time, independent software vendors are playing an important role in helping telcos solve problems with AI and data.

“We are starting to see a lot of AI adoption through domain specific models, primarily through ISVs [independent software vendors],” says Vuyyuru. “ISVs are creating language models, and making sure that gets deployed to that particular use case.”

Snowflake is one example of an ISV that AWS has partnered with, collaborating to build a data foundation for AI.

AWS offers its partners AI infrastructure, powered by its own custom silicon and Nvdia GPUs. The combination of Amazon’s infrastructure layer, Sagemaker and Bedrock, “can use any model out there, and basically govern it, put the controls in place to make sure the performance is graduated properly,” according to Vuyyuru.

Vuyyuru expects to see ISVs specialize in areas such as lifecycle management or serviceability. “ISVs have domain expertise,” he explains. “They are absolutely necessary to … enable democratization of AI within telcos.”

By “simplifying these … domains into a single view, you allow developers in the telco, customer service administrators … with access to these tools … to improve efficiency by 10%, 20%, 60%,” says Vuyyuru.