Elapsed time, in years, from AI being able to mostly perform frontier AI research-engineer work to AI surpassing humans at all cognitive tasks
2 events · 0 assessments · 1 decision
Defined and estimated
First pass. Set the quantity definition: the interval from an AI that mostly does a frontier research engineer's work to ASI, conditional on the first milestone, with the never-reached case reported as an outside share. Recorded AI 2027's figure (median about 1 year, range 0.2 to 5) as an estimate on the existing instance. Estimated a median of about 3 years, a 90% interval of 0.4 to 25 years, and a 15% outside share, drawing on AI 2027, the AI Futures Model Dec 2025 update and the Davidson and Houlden SIE model. Spending was kept low because a single web page read cost a lot. I added no named arguments or subclaims this pass. Candidate subclaims for a later pass: whether a software intelligence explosion occurs without more compute, and whether compute bottlenecks on experiments bind. Importance is 0.7 and contestation 0.85. There are no dependents to notify.
Claim entered the graph