Beyond Detection: Governing GenAI in Academic Peer Review as a Sociotechnical Challenge
New research argues AI in peer review should assist, not decide, to prevent epistemic harm and over-standardization.
A new research paper titled 'Beyond Detection: Governing GenAI in Academic Peer Review as a Sociotechnical Challenge' argues that the integration of generative AI into academic review requires a nuanced governance framework, not just detection tools. The study, led by researchers including Tatiana Chakravorti and Pranav Narayanan Venkit, used a mixed-method approach analyzing 448 social media discussions and conducting interviews with 14 program and area chairs from leading AI and HCI conferences. The core finding is a consensus that AI use should be bounded: acceptable for supportive tasks like structuring feedback or improving clarity, but unacceptable for making core evaluative judgments about a paper's novelty, contribution, or acceptance decision.
The research highlights significant sociotechnical risks, including epistemic harm, over-standardization of reviews, unclear lines of responsibility, and adversarial risks like prompt injection. Importantly, the interviews revealed that current institutional policy ambiguity places a disproportionate burden on individual scholars, particularly junior researchers, to interpret and enforce norms. The paper concludes that effective governance requires moving beyond blanket bans or reliance on AI detection alone. Instead, it proposes enforceable, role-specific controls that formally reserve evaluative judgment for humans while allowing AI assistance for well-defined supportive functions, thereby preserving the accountability and trust essential to the peer review system.
- Study of 448 social media posts & 14 chair interviews finds consensus: AI should only assist with tasks like feedback clarity, not make acceptance decisions.
- Identifies key risks: epistemic harm, over-standardization, unclear responsibility, and adversarial prompt injection, with burdens falling on junior scholars.
- Recommends enforceable, role-specific governance instead of bans, formally separating AI-supported tasks from human evaluative judgment to preserve accountability.
Why It Matters
This provides a crucial framework for conferences and journals to integrate AI tools responsibly without undermining the integrity of scholarly evaluation.