Gary King's study: AI peer reviews beat paid human experts on every measure
ValerioCapraro · x · 2026-10-23
A new paper by Gary King and Steven Worthington finds AI-generated article reviews outperform reviews from well-compensated human experts — judged unusually high quality by editors and authors — on every measure the authors could construct. AI reviews scored higher on all seven assessed dimensions, and three former journal editors preferred the AI reviews in 90% of cases.
The study compares AI reviews against 134 reports from carefully selected, paid human experts. Commentator Valerio Capraro flags huge upside (faster, higher-quality feedback; fully AI-run journals) but warns a few AI gatekeepers could narrow which ideas survive, especially at the scientific frontier where AI hallucination risk is highest — potentially embedding errors at the core of human knowledge.
More from AGI Musings
- UnHerd essay: AI doomerism's proposed controls would be dystopian surveillance — DrTechlash · 2026-10-23
- Benjamin Bratton interview: rethinking human intelligence and agency — bratton · 2026-10-23
- A Claude-generated table of AI doom factors sparks debate on intuition vs knowledge — alexisgallagher · 2026-10-23
- Shield AI co-founder: military AI's biggest bottleneck is that the data doesn't exist — BrettKrieger12 · 2026-10-23
- Critic Slams Doom-Porn Excitement Around Worst-Case AI Scenarios — astralmatrix · 2026-10-23
- GTC rapid-fire: robot execs give four different verbs for how we'll 'program' robots in 5 years — lukas_m_ziegler · 2026-10-23