The AI Review Lottery: Widespread AI-Assisted Peer Reviews Boost Paper Scores and Acceptance Rates

2024

Open publication workspace · Sign in to read the full PDF.

AI-generated summary

1) This study investigates the prevalence and causal impact of AI-assisted peer reviews on submission scores and acceptance rates at the 2024 International Conference on Learning Representations (ICLR).
2)
- Estimates that at least 15.8% of reviews at ICLR 2024 were written with AI assistance using an LLM detector.
- Demonstrates through pairwise comparisons that AI-assisted reviews systematically assign higher scores than human reviews to the same submissions.
- Shows via matched-sample analysis that receiving an AI-assisted review increases a paper's overall acceptance rate, particularly for borderline submissions.
3) Tags: peer review, artificial intelligence, large language models, conference acceptance, causal inference

Check the original publication for accuracy and context.