Leading AI expert delays timeline for AI's possible harm to humanity

Daniel Kokotajlo has revised his AI 2027 scenario, moving expectations for autonomous coding from 2027 to the early 2030s and citing 2034 as a possible horizon for superintelligence. Other experts say timelines are being pushed out and that the term 'AGI' is becoming less precise.

Daniel Kokotajlo, a former OpenAI employee, has updated the AI 2027 scenario he published last year. The original piece projected that AI would achieve fully autonomous coding by 2027 and that an intelligence explosion could follow. In the update, Kokotajlo and his co-authors now place autonomous coding in the early 2030s and suggest 2034 as a possible horizon for superintelligence. The new forecast does not specify a date for any potential large-scale harm to humanity. Key points: - Kokotajlo's AI 2027 initially named 2027 as the year for "fully autonomous coding." - In the update, the authors move likely autonomous coding to the early 2030s and cite 2034 as a possible superintelligence horizon. - The revised forecast no longer includes a date for potential large-scale harms to people. - OpenAI's CEO, Sam Altman, has described an automated AI researcher as an internal goal for March 2028 but said the company may fail to meet that goal. Summary: The revision shifts expectations about when AI might autonomously perform coding and when a form of superintelligence might emerge, and it has prompted experts to reassess shorter timelines. Undetermined at this time.