With the advancement of AI capabilities, AI reviewers are beginning to be deployed in scientific peer review, yet their capability and credibility remain in question: many scientists simply view them as probabilistic systems without the expertise to evaluate research, while other researchers are more optimistic about their readiness without concrete evidence. Understanding what AI reviewers do well, where they fall short, and what challenges remain is essential. However, existing evaluations of AI reviewers have focused on whether their verdicts match human verdicts (e.g., score alignment, acceptance prediction), which is insufficient to characterize their capabilities and limits. In this paper, we close this gap through a large-scale expert annotation study, in which 45 domain scientists in Physical, Biological, and Health Sciences spent 469 hours rating 2,960 individual criticisms (each targeting one specific aspect of a paper) from human-written and AI-generated reviews of 82 Nature-family papers on correctness, significance, and sufficiency of evidence. On a composite of all three dimensions, a reviewing agent powered by GPT-5.2 scores above each paper's top-rated human reviewer (60.0% vs. 48.2%, p = 0.009), while all three AI reviewers (including Gemini 3.0 Pro and Claude Opus 4.5) exceed the lowest-rated human across every dimension. AI reviewers' accurate criticisms are also more often rated significant and well-evidenced, and surface a distinct 26% of issues no human raises. However, AI reviewers overlap far more than humans do (21% vs. 3% for cross-reviewer pairs), and exhibit 16 recurring weaknesses humans do not share, such as limited subfield knowledge, lack of long context management over multiple files, and overly critical stance on minor issues. Overall, our results position current AI reviewers as complements to, not substitutes for, human reviewers.
On the limits and opportunities of AI reviewers: Reviewing the reviews of Nature-family papers with 45 expert scientists
AI reviewers demonstrate superior performance in identifying correct criticisms compared to human reviewers, yet exhibit limitations in subfield knowledge and context management that distinguish them from human peers.
- Year
- 2026
- Venue
- arXiv 2026
- Stars
- 17
- Authors
- 58
- Hosting
- Abstract onlyARXIV-DEFAULT
Cite
Notes
Only stored in your browser.
Attribution
- Abstract & full text
- arxiv.org/abs/2605.20668ARXIV-DEFAULT
- TL;DR
- Semantic Scholar
Abstract
Authors
58Graham NeubigXiang YueJinheon BaekJiHye ParkAkari AsaiPranjal AggarwalDongkeun YoonIan WuSeungone KimEdward ChoiSean WelleckKiril GashteovskiCarolin LawrenceRuoqi LiuAlice OhJuyoung SukViktor ZaverkinSpase PetkoskiDaniel R. SchriderIlija DukovskiFrancesco SantiniBiljana MitreskaYong JeongKyeongha KwonYoung Min SimDragana ManasovaArthur PortoBiljana MojsoskaMakoto TakamotoMarko ShuntovHyunjoo Jenny LeeNiyazi Ulas DinçYehhyun JoSunkyu HanChungwoo LeeHuishan LiEsther H. R. TsaiErgun SimsekKhushboo ShafiYeonseung ChungAleksandar ShulevskiHenrik ChristiansenYoosang SonElly KnightAmanda MontoyaJeongyoun AhnChristian LangkammerHeera MoonChangwon YoonNikola StikovMooseok JangJunhan KimYeon Sik JungWoo Youn KimJae Kyoung KimIshraq Md AnjumHyun Uk KimDrew Bridges