Nvidia exec thinks industry is ‘very close’ to autonomous AI

  • Nvidia’s telecom exec told Fierce we’re quite close to a future in which AI managers rule the roost
  • But enabling AI to manage AI will require changes to the way the tech is trained and deployed
  • The exec also argued 6G can’t just be an evolution of 5G

There’s a future not too far on the horizon where artificial intelligence (AI) becomes its own boss. We’ve told you before that autonomous networks are likely to be rolled out in the coming years. But Nvidia’s (NASDAQ: NVDA) VP of Telecom Ronnie Vasishta told Fierce there are a few things that have to happen first.

According to Vasishta, we’re actually “very close” to a world in which AI manages AI. “It’s more of a way of thinking about AI moreso than the technology barriers to making that happen” that’s standing in the way.

Vasishta explained that autonomous AI will require two different layers of the technology: a foundational generative AI layer and a management AI layer. But that in turn will require changes to the way AI is trained and deployed, he said. It will also require AI to be interoperable.

“The things that need to happen if I take it to, say, a network level…it’s how to enable autonomous generative AI to interact with another generative AI. That today hasn’t really started to happen yet,” he said. “Most of the generative AI training and deployments have been driven by the user, and the user has tended to be an enterprise or a person. At some point, AIs are going to be users.”

And on the deployment side of the equation, “that will require the management AI to be an overlay.”

AT&T recently highlighted concerns about latent bias cropping up in a world of autonomous AI and potentially causing catastrophic problems without a human in the loop to spot wrong decisions. The operator also shared the ways in which it is actively working to uncover such bias as it vets hundreds of AI applications.

Vasishta agreed AI guardrails important, but argued implementing such safeguards – even for autonomous AI – is relatively simple and can actually reduce risk.

“Whether it’s foundational layer AI or management layer AI, it’s the same. You have to ensure that those foundational guardrails are put in. But you can also then on your management layer have additional levels of guardrailing,” he said. “The management layer gives you the opportunity to have an additional security level as well – just because another AI is managing an AI doesn’t mean you’re adding more levels of concern, you’re actually probably eliminating levels of concern.”

Next gen networks

Vasishta also weighed in on how networks need to evolved to support an AI-infused future. He told Fierce 5G has yet to deliver on what was promised. And so, he said, 6G can’t just be an extension of that.

“Compute has to be an integral part of that network – AI compute – for it to really operate and deliver all the services that we expect, whether it be autonomous vehicles or it be generative AI delivered to your phone,” he concluded. “I think the conversation around 6G has been evolutionary and it needs to be revolutionary."