A former OpenAI researcher says he's concerned AI poses a risk to humanity.
"I think maybe there's something like a 10-20% chance of AI takeover, [with] many [or] most humans dead, " Paul Christiano said during an appearance on the Bankless podcast. "I take it quite seriously."
Christiano, who runs the non-profit Alignment Research Center and previously ran the language model alignment team at OpenAI, was being questioned on the likelihood of a "full-out Eliezer Yudkowsky doom scenario" by hosts David Hoffman and Ryan Sean Adams. Yudkowsky has been described as an "AI doomer" and has been warning about the potential risks of the technology for more than 20 years.
Christiano says his views differ from Yudkowsky's when it comes to the speed that the technology will likely develop. "Eliezer is into this extremely fast transformation once you develop AI," he said. "I have a little bit less of an extreme view on that."
"I tend to imagine something more like a year's transition from AI systems that are a pretty big deal, to kind of accelerating change, followed by further acceleration, et cetera," he said. "I think once you have that view then, sort of, a lot of things may feel like AI problems because they happen very shortly after you build AI."
"Overall, maybe you're getting more up to a 50/50 chance of doom shortly after you have AI systems that are human level," he said.
Christiano joins a growing number of voices that have been unnerved by developments in AI. Recently a group of AI experts signed an open letter that called for a 6-month pause on advanced AI development.
Christiano did not immediately respond to Insider's request for comment made outside normal working hours.
