ICLR Author Asks How to Handle a Suspected AI-Generated Review
An ICLR author asked whether to alert the area chair about a review they suspected was AI-generated or simply address every criticism in the rebuttal. The author said Pangram flagged the review as 100% AI, while noting that detector results are not proof and citing additional signs such as criticisms of appendix material.
The post documents a practical dilemma authors face when AI-related review concerns arise, while also highlighting the limits of treating detector scores as proof.
I'm curious how ACs and experienced authors think about this.
My situation: one ICLR reviewer gave a 2 with confidence 1. The other reviewers gave 3 with confidence 3. Pangram flags the review as 100% AI. I know detectors aren't proof on their own, but there are other signs too: all three listed weaknesses concern material deep in the appendix.
I'll respond to every point in the rebuttal regardless. My questions:
- Is it worth sending a confidential comment to the AC, or does that tend to backfire?
- For ACs: what evidence would you actually act on? Is a detector score meaningful to you, or only content-based issues such as generic or inaccurate criticisms?
- How much weight does a confidence 1 review realistically carry in the meta review?
- Has anyone flagged an AI generated review yet? What happened?
Thanks!