Four-Year Study: Experts Including Superforecasters Repeatedly Underestimated AI Progress
Afinetheorem · x · 2026-09-24
ResearchFRI shares interim findings from the most comprehensive series of studies on expert AI forecasts run over four years. Experts—top economists, computer scientists, biologists—dramatically underestimated AI capability progress every year, and superforecasters underestimated it even more. Track records on diffusion measures were mixed: AI revenue forecasts were major underestimates while electricity-share forecasts are on track. Notable overestimates include mid-2025 models' ability to help amateurs with biorisk lab tasks and the rollout speed of self-driving cars. It's too soon to judge long-run forecaster performance.
More from AGI Musings
- Teach students to use AI feedback relentlessly — never let AI write for you — paulnovosad · 2026-11-11
- Beyond anthropocentrism: making AGI a participant in life's unfolding, not its squasher — danfaggella · 2026-11-11
- Nearly 30% of US STEM PhD dissertations now have AI-generated abstracts, NBER study finds — joshgans · 2026-11-11
- Research finds AI writing feedback as good as human, but students resist it — paulnovosad · 2026-11-11
- Economist Paul Novosad: writing classes shouldn't encourage AI use — paulnovosad · 2026-11-11
- French netizen proposes datacenters pay local residents compensation — clemnt · 2026-11-11