AI Article Reviews Dominate Human Peer Reviews at Their Best: Evidence and Policy Choices

Why it matters

Gary King and Steven Worthington, researchers affiliated with Harvard School of Public Health, report that AI-generated article reviews surpassed human expert reviews on every measure they could construct. The comparison was not against average peer reviews: the human reviews came from well-compensated experts and were judged unusually high quality by journal editors and authors. The finding arrives amid broader debate over how large language models should fit into scholarly publishing, where peer review is a bottleneck for speed and quality. The authors frame the result as evidence for journal policy choices rather than a simple endorsement of automation. They acknowledge concerning downsides while pointing to potential gains: faster, higher-quality publication, fewer adversarial interactions between reviewers and authors, and better experiences for editors, reviewers, authors, and readers. The study matters because peer review underpins scientific gatekeeping; if machine reviews outperform even elite human reviewers, journals face difficult decisions about which tasks to delegate, how to preserve accountability, and what role remains for human judgment.