A group of 25 prominent mathematicians, all recipients of the Fields Medal, have expressed growing unease about how AI research labs, including OpenAI, are handling breakthroughs in mathematical problem-solving. They argue that the rush to produce AI-generated proofs risks undermining the traditional processes of verification, attribution, and scholarly communication.
Recently, NYU professor Tristan Buckmaster accused OpenAI of pressuring him not to credit a collaborator affiliated with Anthropic for solving a significant math problem. Buckmaster also questioned whether OpenAI used insights from their work with Codex to develop its own proof during an intensive inference session.
In response to criticism from researchers at CalTech, OpenAI withdrew its sponsorship of a mathematics event at the institution. This incident underscores the tensions between AI labs and the academic community.
The mathematicians emphasize that while AI's ability to solve complex problems could benefit humanity, these solutions must be clearly understood and integrated by the math community. They warn that hastily announced AI proofs often lack proper documentation, isolation of new methods, and appropriate citations, raising concerns about plagiarism and attribution.
There is also apprehension that AI labs might incorporate researchers' work into their models without consent, potentially discouraging open collaboration. The financial resources available to AI labs enable them to rapidly produce proofs, possibly outpacing traditional researchers and incentivizing secrecy.
This open letter builds on the earlier Leiden Declaration, which addressed the implications of AI-generated proofs and recommended actions for mathematicians, institutions, and policymakers.
The mathematicians highlight that the value of mathematics extends beyond proofs to the intellectual framework that fosters new ideas and education. They caution that the challenges faced by the mathematical community reflect broader issues that many scientific and creative fields will encounter as AI transforms workflows.
Ultimately, they stress the importance of maintaining focus on the original goals of scholarly work amid the rise of AI-driven methods.