Instead of replacing human peer review, journals should mandate at least one high‑quality large‑language model review alongside human referees, using closed (non‑public) systems and reserving human judgment for originality and interest. This hybrid approach could cut load, reduce low‑quality reviews, and surface technical errors more reliably while preserving human assessment for subtle evaluative judgments.
— If adopted widely, this would change gatekeeping, incentives, and the epistemic infrastructure of science and scholarship, with implications for confidentiality, liability, and standards of evidence.
Adam Mastroianni
2026.10.28
85% relevant
The author argues for using AI to flag reproducibility issues (code, preregistration mismatches, supplement inconsistencies) and cites concrete tests where AI outperforms human reviewers — directly supporting the proposition that journals should operationalize AI reviewers as part of editorial workflows.
Paul Bloom
2026.10.05
100% relevant
Paul Bloom's Chronicle/Substack piece proposes adding AI reviewers, cites testing with Claude Opus 5.5 and GPT‑5.6, and recommends closed‑AI deployments plus human role retention.
← Back to all ideas