MIT retracts high-profile paper claiming AI supercharges discovery

CAMBRIDGE, MA., May 16 — MIT has asked arXiv and the Quarterly Journal of Economics to withdraw “Artificial Intelligence, Scientific Discovery, and Product Innovation,” a 2024 preprint that claimed an AI assistant in a materials-science lab dramatically boosted the pace of new-compound discoveries and related patent filings.

The study, written by second-year economics Ph.D. student Aidan Toner-Rodgers, drew early praise from Nobel laureate Daron Acemoglu and fellow economist David Autor, who last year said he was “floored” by the results.

A January complaint triggered a confidential review by MIT’s Committee on Discipline. In a statement released 16 May, the Institute said it now has “no confidence in the provenance, reliability or validity of the data and in the veracity of the research,” and confirmed that the author “is no longer at MIT.”

Because arXiv normally requires authors to retract their own manuscripts—and Toner-Rodgers has not—MIT petitioned the repository directly, warning that hosting the paper “may violate arXiv’s Code of Conduct.” It simultaneously asked QJE editors to pull the submission from review.

Acemoglu and Autor, now distancing themselves from a study they once spotlighted, said public clarification was essential because the unreviewed paper was already shaping policy debates on AI’s effect on science. The episode amplifies scrutiny of headline-grabbing AI claims and underscores research institutions’ growing insistence on transparent data and reproducible methods as artificial intelligence spreads through the laboratory.

*Sources:

*https://economics.mit.edu/news/assuring-accurate-research-record

*https://www.wsj.com/tech/ai/mit-says-it-no-longer-stands-behind-students-ai-research-paper-11434092