AI Impacts survey critics flag vague wording, tiny effective sample and MIRI ties

DrTechlash · x · 2026-09-15

The new AI Impacts survey on AI extinction risk draws renewed methodological criticism: respondents assign probabilities to a bundled "extinction or similarly permanent and severe disempowerment" outcome that media report as "extinction"; organizers themselves admit respondents aren't forecasting experts and framing effects are significant; the effective sample is 8% of 19,874 emails, with three question variants answered by subsets of only 744, 392 and 353 people; and the survey sits within Berkeley's MIRI, whose co-founder Eliezer Yudkowsky advocates shutting AI down, with co-author David Krueger assigning 75% p(doom).

Related event: Major AI researcher survey puts extinction risk near 18%, sparking debate(14 posts)→

Original post →

More from AGI Musings

AGI Musings channel →