AI 2027 Authors Pull Their Superhuman AI Researcher Forecast Into 2027
Unfortunately, it seems like the intelligence explosion will be happening in 2027 after all. Most of this post analyzes the new METR report (though other sources of evidence are also important). If they’re right, that implies an AI-driven acceleration that is 5 months ahead of the AI 2027 scenario. I am scared, sad, and angry.
Context
If these forecasters are right, AI that does AI research better than people could arrive within about a year, far sooner than most plans assume.
Daniel Kokotajlo is a former OpenAI researcher who leads the AI Futures Project, the group behind the AI 2027 scenario. AI 2027 described how AI systems that automate AI research could set off a rapid intelligence explosion. METR is an independent group that evaluates what frontier AI models can do.
Deep dive
Daniel Kokotajlo posted on Nov 10 that the intelligence explosion seems like it will happen in 2027 after all. He said most of a new AI Futures Project post analyzes a new METR report, and that if METR is right, the implied AI driven acceleration is five months ahead of the AI 2027 scenario. He added that he is scared, sad and angry.
Andrew Curran summarized the update on X: the AI Futures Project says the world is five months ahead of AI 2027, and the group now believes a Superhuman AI Researcher will arrive sometime in 2027.
Coauthor Eli Lifland wrote that METR estimates a nearly 2x acceleration in Anthropic's progress due to AI. He said his previous median for a Superhuman AI Researcher was about 2030 and is now 2027, and that he is very worried humanity may no longer be in control of its future within 3 to 12 months.
Buck Shlegeris of Redwood Research posted separately, linking the same update, that he puts about a one in three chance on AI companies building superintelligence within the next year, and that he thinks such systems would probably take over if that happens.
The update also drew pushback. Some replies on X said much of the main evidence behind it is redacted, and others questioned how the acceleration shows up in METR's chart. These are forecasts and personal estimates, not measured outcomes, and the METR findings have not been independently reviewed.
What to expect next
Watch for METR to publish more detail on how it measured the speedup, and for Anthropic, OpenAI and Google to respond to the claim that AI is accelerating their own research. Other forecasters may revise their timelines in the coming weeks.
The key open question is whether the measured acceleration holds up under outside review, since much of the evidence is not public. Expect safety groups and policy makers to cite this update in debates about how fast to act on frontier AI.