Elamur Research⌕
☰
home / papers / 2024 / AgentReview: Exploring Peer Review Dynamics with LLM Agents

AgentReview: Exploring Peer Review Dynamics with LLM Agents

Paper · 2024 · Yiqiao Jin, Qinlin Zhao, Yiyang Wang et al.

Summary

Studying peer review empirically is hard. Outcomes depend on many interacting factors, some important ones such as reviewer bias or motivation are hidden, and real review records are confidential. AgentReview addresses these problems by simulating conference reviewing entirely with LLM agents playing reviewers, authors and area chairs. The simulation follows a five-stage pipeline: initial reviews, author rebuttal, reviewer–AC discussion, meta-review, and an accept/reject decision at a fixed 32% acceptance rate. Researchers can vary one factor at a time:

  • Reviewers: commitment, intention and knowledgeability.
  • Area chairs: authoritarian, conformist or inclusive style.
  • Authors: whether their identity is visible to reviewers.

The authors ran this on 500+ ICLR submissions from four years. They found that a single biased reviewer can shift about 37% of decisions, and that rebuttals matter less than reviewer traits. They interpret these effects through sociological ideas such as conformity, groupthink and authority bias. Because the setup needs no real reviewer data, it offers a privacy-safe testbed for designing review mechanisms. However, the simulated outcomes were not validated directly against actual human reviews.

Key points

  • The resource contains more than 53,800 generated documents (reviews, rebuttals, revised reviews, meta-reviews and decisions), including over 10,000 reviews. Code is at github.com/Ahren09/AgentReview.
  • Involving one biased reviewer changes 37.1% of decisions. Letting ACs rely on their own judgment changes 30.6%.
  • After discussion, rating spread shrinks by 27.2% (standard deviation) as reviewers drift toward consensus. One under-committed reviewer lowers overall commitment by 18.7%.
  • Revealing author identity for just 10% of papers alters 27.7% of decisions. Biased reviewers lower ratings by 0.17 among themselves and also pull unbiased reviewers' ratings down by 0.25.
  • When numeric ratings are hidden from the AC, the set of accepted papers overlaps only weakly with the baseline (Jaccard 0.20). The authors caution against using LLMs as substitutes for human reviewers.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Carnegie Mellon UniversityPittsburgh · United States598 works · 191,226 citations🎓Georgia Institute of TechnologyAtlanta · United States152 works · 27,763 citations🎓University of Science and Technology of ChinaHefei · China145 works · 114,409 citations🎓University of California, Los AngelesLos Angeles · United States99 works · 56,333 citations🎓University of California, Santa BarbaraSanta Barbara · United States67 works · 10,550 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Arts and Humanities · 17,450 worksDecision Sciences · 249 worksStatistics, Probability and Uncertainty · 65 worksHistory and Philosophy of Science · 62 works

Cite

Cite as Yiqiao Jin, Qinlin Zhao, Yiyang Wang, Hao Chen, Kaijie Zhu, Yijia Xiao and Jindong Wang. 2024. AgentReview: Exploring Peer Review Dynamics with LLM Agents. arXiv:2406.12708.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2406.12708 · arXiv 2406.12708

@misc{jin2024agentreview,
  title = {{AgentReview: Exploring Peer Review Dynamics with LLM Agents}},
  author = {Yiqiao Jin, Qinlin Zhao, Yiyang Wang, Hao Chen, Kaijie Zhu, Yijia Xiao and Jindong Wang},
  year = {2024},
  doi = {10.48550/arxiv.2406.12708},
  eprint = {2406.12708},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2406.12708}
}

References

65 references, 14 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Joshua D Angrist. 2014. The perils of peer effects. Labour Economics, 30:98–108. find
  2. Anthropic. 2024. Introducing the next generation of claude. find
  3. Yushi Bai, Jiahao Ying, Yixin Cao, Xin Lv, Yuze He, Xiaozhi Wang, Jifan Yu, Kaisheng Zeng, Yijia Xiao, Haozhe Lyu, et al. 2023. Benchmarking foundation models with language-model-as-an-examiner. arXiv:2306.04181.
  4. Otomar J Bartos and Paul Wehr. 2002. Using conflict theory. Cambridge University Press. find
  5. Chi-Min Chan, Weize Chen, Yusheng Su, Jianxuan Yu, Wei Xue, Shanghang Zhang, Jie Fu, and Zhiyuan Liu. 2023. Chateval: Towards better llm-based evaluators through multi-agent debate. In ICLR.
  6. Canyu Chen, Baixiang Huang, Zekun Li, Zhaorun Chen, Shiyang Lai, Xiongxiao Xu, Jia-Chen Gu, Jindong Gu, Huaxiu Yao, Chaowei Xiao, et al. 2024a. Can editing llms inject harm? In ICML 2024 Next Generation of AI Safety Workshop. find
  7. Yuyan Chen, Yueze Li, Songzhou Yan, Sijia Liu, Jiaqing Liang, and Yanghua Xiao. 2024b. Do large language models have problem-solving capability under incomplete information scenarios? In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics. find
  8. Yuyan Chen, Songzhou Yan, Panjun Liu, and Yanghua Xiao. 2024c. Dr.academy: A benchmark for evaluating questioning capability in education for large language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics. find
  9. Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, et al. 2024. The llama 3 herd of models. arXiv:2407.21783.
  10. Yanai Elazar, Jiayao Zhang, David Wadden, Bo Zhang, and Noah A Smith. 2024. Estimating the causal effect of early arxiving on paper acceptance. In CLeaR, pages 913–933. PMLR. find
  11. Charles W Fox, Jennifer Meyer, and Emilie Aimé. 2023. Double-blind peer review affects reviewer ratings and editor decisions at an ecology journal. Functional Ecology, 37(5):1144–1157. find
  12. Junjie Huang, Win-bin Huang, Yi Bu, Qi Cao, Huawei Shen, and Xueqi Cheng. 2023. What makes a successful rebuttal in computer science conferences?: A perspective on social interaction. Journal of Informetrics, 17(3):101427. find
  13. Irving L Janis. 2008. Groupthink. IEEE Engineering Management Review, 36(1):36. find
  14. Albert Q Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, et al. 2023. Mistral 7b. arXiv:2310.06825. find
  15. Bowen Jiang, Zhijun Zhuang, Shreyas S Shivakumar, Dan Roth, and Camillo J Taylor. 2024. Multi-agent vqa: Exploring multi-agent foundation models in zero-shot visual question answering. In The IEEE/CVF Conference on Computer Vision and Pattern Recognition 2024 Workshop on What is Next in Multimodal F… find
  16. Jelena Jovanovic and Ebrahim Bagheri. 2023. Reviewer assignment problem: A scoping review. arXiv:2305.07887. — [arXiv:2305.07887](https://arxiv.org/abs/2305.07887) find
  17. Robert B Cialdini and Noah J Goldstein. 2004. Social influence: Compliance and conformity. Annu. Rev. Psychol., 55:591–621. find
  18. Zemian Ke, Haocheng Duan, and Sean Qian. 2024. Interpretable mixture of experts for time series prediction under recurrent and non-recurrent conditions. arXiv:2409.03282. — [arXiv:2409.03282](https://arxiv.org/abs/2409.03282) find
  19. Matteo Cinelli, Gianmarco De Francisci Morales, Alessandro Galeazzi, Walter Quattrociocchi, and Michele Starnini. 2021. The echo chamber effect on social media. PNAS, 118(9):e2023301118. find
  20. Kayvan Kousha and Mike Thelwall. 2024. Artificial intelligence to support publishing and peer review: A summary and review. Learned Publishing, 37(1):4– 12. find
  21. Corinna Cortes and Neil D Lawrence. 2021. Inconsistency in conference peer review: revisiting the 2014 neurips experiment. arXiv:2109.09774. — [arXiv:2109.09774](https://arxiv.org/abs/2109.09774) find
  22. Ji-Ung Lee, Haritz Puerto, Betty van Aken, Yuki Arase, Jessica Zosa Forde, Leon Derczynski, Andreas Rücklé, Iryna Gurevych, Roy Schwartz, Emma Strubell, et al. 2023. Surveying (dis) parities and concerns of compute hungry nlp research. arXiv:2306.16900. find
  23. Mike D’Arcy, Tom Hope, Larry Birnbaum, and Doug Downey. 2024. Marg: Multi-agent review generation for scientific papers. arXiv:2401.04259.
  24. Jiangshu Du, Yibo Wang, Wenting Zhao, Zhongfen Deng, Shuaiqi Liu, Renze Lou, Henry Peng Zou, Pranav Narayanan Venkit, Nan Zhang, Mukund Srinath, et al. 2024. Llms assist nlp researchers: Critique paper (meta-) reviewing. arXiv:2406.16253.
  25. Manling Li*, Shiyu Zhao*, Qineng Wang*, Kangrui Wang*, Yu Zhou*, Sanjana Srivastava, Cem Gokmen, Tony Lee, Li Erran Li, Ruohan Zhang, Weiyu Liu, Percy Liang, Li Fei-Fei, Jiayuan Mao, and Jiajun Wu. 2024. Embodied agent interface: Benchmarking llms for embodied decision making. In NeurIPS. find
  26. Miao Li, Jey Han Lau, and Eduard Hovy. 2024. Exploring multi-document information consolidation for scientific sentiment summarization. arXiv:2402.18005. — [arXiv:2402.18005](https://arxiv.org/abs/2402.18005) find
  27. Ruosen Li, Teerth Patel, and Xinya Du. 2023a. Prd: Peer rank and discussion improve large language model based evaluations. arXiv:2307.02762.
  28. Yuchen Li, Haoyi Xiong, Qingzhong Wang, Linghe Kong, Hao Liu, Haifang Li, Jiang Bian, Shuaiqiang Wang, Guihai Chen, Dejing Dou, et al. 2023b. Coltr: Semi-supervised learning to rank with co-training and over-parameterization for web search. TKDE, 35(12):12542–12555. find
  29. Weixin Liang, Zachary Izzo, Yaohui Zhang, Haley Lepp, Hancheng Cao, Xuandong Zhao, Lingjiao Chen, Haotian Ye, Sheng Liu, Zhi Huang, et al. 2024. Monitoring ai-modified content at scale: A case study on the impact of chatgpt on ai conference peer reviews. arXiv:2403.07183.
  30. Weixin Liang, Yuhui Zhang, Hancheng Cao, Binglu Wang, Daisy Ding, Xinyu Yang, Kailas Vodrahalli, Siyu He, Daniel Smith, Yian Yin, et al. 2023. Can large language models provide useful feedback on research papers? a large-scale empirical analysis. arXiv:2310.01783.
  31. Haoxin Liu, Zhiyuan Zhao, Jindong Wang, Harshavard-han Kamarthi, and B Aditya Prakash. 2024. Lst-prompt: Large language models as zero-shot time series forecasters by long-short-term prompting. In ACL. find
  32. Ryan Liu, Steven Jecmen, Vincent Conitzer, Fei Fang, and Nihar B Shah. 2023a. Testing for reviewer anchoring in peer review: A randomized controlled trial. arXiv:2307.05443. — [arXiv:2307.05443](https://arxiv.org/abs/2307.05443) find
  33. Ying Liu, Kaiqi Yang, Yue Liu, and Michael GB Drew. 2023b. The shackles of peer review: Unveiling the flaws in the ivory tower. arXiv:2310.05966. — [arXiv:2310.05966](https://arxiv.org/abs/2310.05966) find
  34. Yuxuan Lu and Yuqing Kong. 2024. Calibrating “cheap signals” in peer review without a prior. NeurIPS, 36. find
  35. Xu Ma, Yuqian Zhou, Huan Wang, Can Qin, Bin Sun, Chang Liu, and Yun Fu. 2022. Image as set of points. In ICLR. find
  36. Leslie D McIntosh and Cynthia Hudson Vitale. 2023. Safeguarding scientific integrity: Examining conflicts of interest in the peer review process. arXiv:2308.04297. — [arXiv:2308.04297](https://arxiv.org/abs/2308.04297) find
  37. OpenAI. 2023. Gpt-4 technical report. arXiv Preprint, arXiv:2303.08774.
  38. Joon Sung Park, Joseph O’Brien, Carrie Jun Cai, Meredith Ringel Morris, Percy Liang, and Michael S Bernstein. 2023. Generative agents: Interactive simulacra of human behavior. In UIST, pages 1–22. find
  39. Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. In EMNLP, pages 3982–3992.
  40. Martin Saveski, Steven Jecmen, Nihar Shah, and Johan Ugander. 2024. Counterfactual evaluation of peer-review assignment policies. NeurIPS, 36. find
  41. Significant-Gravitas. 2023. Autogpt. https://github.com/Significant-Gravitas/ AutoGPT. ↗
  42. Ivan Stelmakh, Nihar B Shah, Aarti Singh, and Hal Daumé III. 2021. Prior and prejudice: The novice reviewers’ bias against resubmissions in conference peer review. HCI, 5(CSCW1):1–17. find
  43. Dimity Stephen. 2024. Distinguishing articles in questionable and non-questionable journals using quantitative indicators associated with quality. arXiv:2405.06308. — [arXiv:2405.06308](https://arxiv.org/abs/2405.06308) find
  44. Mengyi Sun, Jainabou Barry Danfa, and Misha Teplitskiy. 2022. Does double-blind peer review reduce bias? evidence from a top computer science conference. Journal of the Association for Information Science and Technology, 73(6):811–819. find
  45. Gemini Team, Rohan Anil, Sebastian Borgeaud, Yonghui Wu, Jean-Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew M Dai, Anja Hauth, et al. 2023. Gemini: a family of highly capable multimodal models. arXiv:2312.11805.
  46. Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al. 2023. Llama 2: Open foundation and fine-tuned chat models. arXiv:2307.09288.
  47. John C Turner. 1991. Social influence. Thomson Brooks/Cole Publishing Co. find
  48. Alexander Ugarov. 2023. Peer prediction for peer review: designing a marketplace for ideas. arXiv:2303.16855. — [arXiv:2303.16855](https://arxiv.org/abs/2303.16855) find
  49. Richard E Nisbett and Timothy D Wilson. 1977. The halo effect: Evidence for unconscious alteration of judgments. Journal of personality and social psychology, 35(4):250. find
  50. Jeroen PH Verharen. 2023. Chatgpt identifies gender disparities in scientific peer review. Elife, 12:RP90230. find
  51. Mahsan Nourani, Chiradeep Roy, Jeremy E Block, Donald R Honeycutt, Tahrima Rahman, Eric Ragan, and Vibhav Gogate. 2021. Anchoring bias affects mental model formation and user reliance in explainable ai systems. In IUI, pages 340–350. find
  52. Qingyun Wu, Gagan Bansal, Jieyu Zhang, Yiran Wu, Shaokun Zhang, Erkang Zhu, Beibin Li, Li Jiang, Xiaoyun Zhang, and Chi Wang. 2023a. Autogen: Enabling next-gen llm applications via multi-agent conversation framework. arXiv:2308.08155.
  53. Yuxiang Wu, Zhengyao Jiang, Akbir Khan, Yao Fu, Laura Ruis, Edward Grefenstette, and Tim Rocktäschel. 2023b. Chatarena: Multi-agent language game environments for large language models. GitHub repository. find
  54. Chengxing Xie, Canyu Chen, Feiran Jia, Ziyu Ye, Kai Shu, Adel Bibi, Ziniu Hu, Philip Torr, Bernard Ghanem, and Guohao Li. 2024. Can large language model agents simulate human trust behaviors? In ICLR 2024 Workshop: How Far Are We From AGI. find
  55. Haoyi Xiong, Jiang Bian, Yuchen Li, Xuhong Li, Mengnan Du, Shuaiqiang Wang, Dawei Yin, and Sumi Helal. 2024. When search engine services meet large language models: Visions and challenges. IEEE Transactions on Services Computing. find
  56. Yixuan Xu, Steven Jecmen, Zimeng Song, and Fei Fang. 2024. A one-size-fits-all approach to improving randomness in paper assignment. NeurIPS, 36. find
  57. Yuan Yang, Siheng Xiong, Ali Payani, Ehsan Shareghi, and Faramarz Fekri. 2024. Can llms reason in the wild with programs? arXiv:2406.13764. find
  58. Da Yin, Faeze Brahman, Abhilasha Ravichander, Khyathi Chandu, Kai-Wei Chang, Yejin Choi, and Bill Yuchen Lin. 2023. Lumos: Learning Agents with Unified Data, Modular Design, and Open-Source LLMs. arXiv:2311.05657. find
  59. Guangyao Zhang, Furong Shang, Weixi Xie, Yuhan Guo, Chunlin Jiang, and Xianwen Wang. 2021. Do conspicuous manuscripts experience shorter time in the duration of peer review? arXiv:2112.09360. — [arXiv:2112.09360](https://arxiv.org/abs/2112.09360) find
  60. Jiayao Zhang, Hongming Zhang, Zhun Deng, and Dan Roth. 2022a. Investigating fairness disparities in peer review: A language model enhanced approach. arXiv:2211.06398. find
  61. Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q Weinberger, and Yoav Artzi. 2020. Bertscore: Evaluating text generation with bert. In ICLR. find
  62. Yichi Zhang, Fang-Yi Yu, Grant Schoenebeck, and David Kempe. 2022b. A system-level analysis of conference peer review. In EC, pages 1041–1080. find
  63. Yu Zhang, Xiusi Chen, Bowen Jin, Sheng Wang, Shuiwang Ji, Wei Wang, and Jiawei Han. 2024. A comprehensive survey of scientific large language models and their applications in scientific discovery. arXiv:2406.10833. find
  64. Qinlin Zhao, Jindong Wang, Yixuan Zhang, Yiqiao Jin, Kaijie Zhu, Hao Chen, and Xing Xie. 2024. Com-peteai: Understanding the competition behaviors in large language model-based agents. In ICML. find
  65. Kaijie Zhu, Jindong Wang, Qinlin Zhao, Ruochen Xu, and Xing Xie. 2024. Dynamic evaluation of large language models by meta probing agents. In ICML.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/jin-2024-agentreview-exploring-peer-review/