Mathematicians Feel Threatened by AI Yet Keep Using It
Summary
Mathematicians are increasingly relying on AI tools even as they fear the technology could undermine their field, obscure human contributions, and make some researchers obsolete. New York University professor Tristan Buckmaster accused OpenAI of using his work to move ahead on the Navier-Stokes existence and smoothness problem, which carries a $1 million bounty. OpenAI later said its investigation found that Buckmaster’s Codex prompts in the preceding two months could not have influenced the system, including through training, but the dispute exposed how difficult it is to establish who contributed to an AI-generated proof. OpenAI also amended an announcement about its Astra model after Andreas Thom said it overlooked his 2019 paper and related work when describing a mathematical problem as having made no recent progress. Thom says he does not know whether his ChatGPT interactions influenced the result, although he continues to use ChatGPT with updated privacy settings because it speeds up writing. The article says AI agents can produce results whose intermediate reasoning mathematicians do not fully understand, challenging the field’s traditions of peer review, attribution, and reproducibility. Townsend and other mathematicians describe a simultaneous sense of excitement about new capabilities and anxiety about their purpose, careers, and the future of younger researchers. Twenty-five Fields medalists signed an open letter calling AI companies and mathematicians severely misaligned, while more than 4,000 people signed the Leiden Declaration. More than 2,000 people connected to Caltech also urged organizers to suspend an AI mathematics hackathon sponsored by Anthropic and OpenAI, although OpenAI later withdrew. Researchers interviewed say efficiency gains may isolate mathematicians who avoid AI, but argue that laboratories and the mathematical community need rules for releasing results, crediting prior work, protecting research data, and guiding students through the transition.