1 min read
Add as a preferred source on Google

Paul Christiano Puts One-Year AI Takeover Estimate at 10%

He also put the three-year chance near 50% if AI labs fail to take proactive steps, and defined misalignment as a gap between a system’s goals and what its overseers want.

An empty lecture hall with a lone lectern under a spotlight / TokenPost.ai
An empty lecture hall with a lone lectern under a spotlight / TokenPost.ai

OpenAI Foundation board member Paul Christiano said he now assigns a 10% chance to an AI takeover within one year, while placing the three-year estimate near 50% if AI labs fail to act proactively.

Christiano gave the one-year estimate on a podcast that aired Nov. 16. At the Hinton Lectures in Toronto, held Nov. 9-13, he put the three-year odds at roughly a coin flip if AI labs do not take proactive steps.

His scenarios involve people becoming unable to stop powerful AI systems after something goes wrong. He described the concern as misalignment: a gap between what an AI system is trying to do and what its human overseers want it to do.

The estimates follow Christiano’s earlier warning about AI accidents, which he described at the lectures as a likely cause of death for many people in the audience.

Sonny Jang

Reporter

Sonny Jang reports on people and leadership in digital assets for TokenPost. Send corrections or tips to info@tokenpost.com.

Loading…