X discussion: a new paper from Harvard's Gary King says AI article reviews beat top human peer reviews "by every measure we could construct", and Ethan Mollick urges academia to adapt fast

Background

Peer review is central to academic publishing, but reviewers are hard to find, reviews are slow, and quality varies. At 00:27 UTC on 2026-10-23, Harvard social scientist and statistician Gary King (@kinggary) posted a new paper with Steven Worthington, "AI Article Reviews Dominate Human Peer Reviews at Their Best: Evidence and Policy Choices". It compares AI reviews directly with the best human reviews, which is why it spread quickly among academics and AI people.

What the paper says

According to the abstract on the paper page, the comparison is not against ordinary reviews but against reviews by well-compensated human experts that journal editors and authors judged to be unusually good, and AI-generated reviews came out better by every measure the authors could construct. The second half discusses journal policy: possible gains in publication speed and quality and fewer adversarial interactions, alongside downsides the authors call "concerning".

Benchmarks

This is not a model benchmark. Wharton professor Ethan Mollick (@emollick) wrote in his post that the study covers 124 Harvard social science papers, from statistical analyses to formal models to qualitative work, and used Claude Opus 4.8. Many reposts on X also cite figures such as "former journal editors picked the AI review 87% of the time" and "5.5% vs 15.9% wrong claims"; we have not checked each one against the paper, so cite the paper itself. Engagement when we pulled it: Mollick's post had about 1,675 likes, 97K views and 249 replies; King's post had about 121K views and 371 bookmarks.

What it means for developers and researchers

The main split in the discussion: supporters see AI as a fast first-pass reviewer that speeds up science, while skeptics worry about AI acting as a gatekeeper, about weak big-picture judgment, and note nobody has checked whether the papers themselves got better. Read the paper page and King's post to judge for yourself, and if you use a model to review your own drafts, still check every factual comment by hand.