Terence Tao and 24 Fields Medalists Warn of AI Misalignment in Mathematics

Summary

Terence Tao says he is one of 25 initial signatories, all Fields Medalists, to a declaration warning that AI companies’ pursuit of mathematical problem-solving benchmarks is misaligned with the goals of mathematical research. The statement follows discussions among the signatories over the preceding week and invites additional signatures, although Tao notes that the urgency left little time for a consultative process. He writes that large language models have improved dramatically in recent months and can solve major outstanding problems in many areas of mathematics. In the signatories’ view, solving a problem is only a proxy for the deeper goal: developing conceptual understanding, insight, and methods that can be discussed, simplified, documented, and transmitted through the mathematical community. They warn that rapidly producing true-or-false results, and announcing solutions before proper writeups, can leave no time to isolate new ideas, acknowledge earlier work, or address attribution and plagiarism concerns. Tao also argues that AI-generated ideas still depend on mathematicians to develop and integrate them into the field’s intellectual canon; without that human transmission chain, results may not become lasting mathematics. The declaration places these concerns within a broader risk to scientific, creative, and other intellectual work, where systems may produce outputs without fostering the training, understanding, and question-formulation that work traditionally develops. Tao says AI could enhance mathematical study, but the outcome will depend on decisions by the humans controlling the technology, making action by mathematicians, AI companies, and society urgent.