Written by Jeffrey Dustin
VANCOUVER (Reuters) – Ilya Sutskever, former chief scientist at OpenAI, one of the biggest names in artificial intelligence, had a prediction on Friday: Logical capabilities will make technology less predictable.
After accepting the Test of Time award for his 2014 paper with Oriol Viñals and Google's Kwok Lu, Sutskever said there was a big change on the horizon for artificial intelligence.
He said the idea his team discovered a decade ago, that expanding data to “pre-train” AI systems would send them to new heights, was beginning to reach its limits. More data and computing power led to the creation of ChatGPT, which OpenAI launched in 2022, which received worldwide acclaim.
“But pre-training as we know it will undoubtedly end,” Sutskever announced to thousands of attendees at the NeurIPS conference in Vancouver. “While computing is growing, data is not growing, because we only have one Internet,” he said.
Sutskever offered some ways to push the boundaries despite this dilemma. The technology itself could generate new data, or AI models could evaluate multiple answers before settling on the best response for the user, to improve accuracy, he said. Other scientists have set their sights on real-world data.
But his speech culminated in a prediction of the future of super-intelligent machines, which he said “clearly” awaits, a point with which some disagree. Sutskever this year co-founded Safe Superintelligence Inc in the wake of his role in the short-lived ouster of Sam Altman from OpenAI, something he said within days he regretted.
He said that AI agents who have been working for a long time will bear fruit in the future era, and will have a deeper understanding and be self-aware. He said that artificial intelligence will solve problems just like humans do.
There is a catch.
“The more reason there is, the more unpredictable it becomes,” he said.
Thinking about millions of options can make any outcome unclear. For example, AlphaGo, a system created by Alphabet's (NASDAQ:) subsidiary DeepMind, surprised experts of the highly complex board game with its mysterious 37th move, on its way to defeating Lee Sedol in a match in 2016.
Likewise, Sutskever said, “Chess AI, which is really good, is less predictable than the best human chess players.”
AI as we know it will be “radically different,” he said.
