25 Fields Medallists sign declaration against AI math race

Twenty-five Fields Medallists signed a declaration on September 11 accusing AI labs of racing through famous math problems for publicity, not conceptual understanding. At stake is who gets credit when a lab claims to have solved a Millennium Prize problem faster than mathematicians can check the work. The signers run from Pierre Deligne, a 1978 winner, to this year's winner, Yu Deng, and include Scholze, Viazovska, and Kontsevich.

Tao's declaration argues labs get rewarded for benchmark scores and speed. Mathematics runs on verification, attribution, and a human chain of who checked whose work. The declaration says rushed announcements with no writeup raise "severe attribution and plagiarism questions." It warns that rapid AI problem-solving "could destroy fertile ground instead of breathing life into new ideas."

TechCrunch quotes the open letter's fuller line. Solutions get "announced in a rush, leaving no time for a proper writeup, the isolation of new methods and ideas, and citing relevant previous work." The story reached 1,003 points on Hacker News.

The same day, the Clay Mathematics Institute issued its first statement on a Navier-Stokes claim, saying the problem "has apparently been settled." It names no lab and no author. On credit it says only that its process "is deliberately unhurried." The body that decides a Millennium Prize winner acknowledged the result and declined to referee who gets credit for it.

A day earlier, Tao posted a crowdsourced list of resources on AI and mathematics. It groups advocacy statements, reports, and mathematician essays, including an MIT committee report on AI in academic training, leaving Tao's own writing off it.

Software doesn't have a prize committee, but it faces the same attribution problem. When an agent hands back a 400-line module that's correct but eroded, the question is the same. Who checked it, and what can the next person rely on? Clay's move is to slow the credit process down until the work can be interrogated. A pull request could use the same patience.