Post by Alex Edmans

Not only are researchers using AI to churn out papers, but referees are using AI to review them. Worryingly, scores from AI detection software found it difficult to distinguish between: - Reports wholly written by AI, and then touched up by a human - Reports initiated by a human reading a paper, coming up with his/her own comments, jotting them down as notes, and asking AI to turn them into full prose Pangram scored both within the 30-50% range. Researchers can spend over a year responding to referee reports, generating tons of new regressions or model extensions that only ever go into an Online Appendix or a 50-page response document to reviewers. A valid critique of peer review is that authors waste substantial time (that could be redirected towards new research that advances knowledge) doing an analysis just because one person asks for it. Now, authors may end up doing it even if no person asks for it. By Lamar Pierce, Claudine Gartenberg, Alex Murray, and Sharique Hasan. https://lnkd.in/eGk3dpxQ (Over the next two days I will discuss potential solutions).