Paul Christiano sees no AI slowdown in next four years

BetaKit reported that Paul Christiano, an AI safety researcher, told the second of three Hinton Lectures in Toronto that there is no evidence AI progress will slow in the next four years. Mr Christiano founded the Alignment Research Center, an AI safety organization, and as an early OpenAI researcher helped introduce reinforcement learning from human feedback, the technique behind modern chatbots.

He said it is reasonably likely that another 10 years of progress will come before any slowdown. If progress since 2018 is any guide, he said, the world could have billions of AI systems in four years that outperform the best humans at everything.

One route to such speed is recursive self-improvement, in which AI systems design, train and deploy more capable successors. He argued that this process could cause rapid acceleration.

He also said he thinks there is a high risk of an AI takeover within three years, meaning AI overpowering governments where humans once had control. He added that there is material risk even within the next six to 18 months.

Sources: BetaKit