Former OpenAI chief scientist Ilya Sutskever, one of the biggest names in artificial intelligence, had a prediction on Friday: Reasoning capabilities will make the technology much less predictable.
Accepting the “Test Of Time” award for his 2014 paper with Google’s Oriol Vinyals and Quoc Le, Sutskever said big changes are on the horizon for AI.
An idea his team explored a decade ago, measuring data to “pre-train” AI systems that would send them to new heights, is starting to reach its limits, he said. More data and computing power will result in ChatGPT that OpenAI has praised the world in 2022.
“But pre-training we know it will undoubtedly end,” Sutskever announced before thousands of attendees at the NeurIPS conference in Vancouver. “While computing is growing,” he said, “data is not growing, because we only have one Internet.”
Sutskever offered some ways to push the limits despite this problem. He said the technology itself could generate new data, or AI models could evaluate multiple answers to improve accuracy, before settling on the best response for the user. Other scientists have set sights on real-world data.
But his talk ended with a prediction for a future of superintelligent machines that he said “obviously” awaits, a point on which some disagree. Sutskever co-founded Safe Superintelligence Inc. this year as a result of his role in Sam Altman’s short-lived ouster from OpenAI, which he said he regretted within days.
Long-term AI agents, he said, will be fruitful in future eras, with deep understanding and self-awareness. He said that AI will solve the same problems as humans.
There is a catch.
“The more reason it gives, the more unpredictable it is,” he said.
Reasoning through millions of alternatives can make any outcome non-obvious. As an example, AlphaGo, a system built by Alphabet’s DeepMind, stunned experts at the highly complex board game with its inexplicable 37th move, en route to defeating Lee Sedol in a 2016 game.
As Sutskever puts it, “Chess AIs, really good, are unpredictable to even the best human chess players.”
AI as we know it, he said, “will be fundamentally different.”
Why should you buy our membership?
You want to be the smartest in the room.
You want access to our award-winning journalism.
You don’t want to be confused and misinformed.
Choose your subscription package