OpenAI creates mathematicians’ panel after disputed AI results

OpenAI has established an independent panel of mathematicians to advise it and other artificial intelligence companies on how mathematical research should be evaluated, communicated and released. The move follows a series of high-profile AI-generated results that have damaged the company’s standing among researchers, according to reporting by The Verge.

Called the Advisory Group on Mathematics and Artificial Intelligence, or AGMAI, the nine-member body is hosted by the Institute for Advanced Study in Princeton, New Jersey. The institution is one of the most prominent centers for mathematical research and was previously associated with figures including Albert Einstein, John von Neumann and J. Robert Oppenheimer.

The panel includes senior academics from universities such as Stanford, Harvard, Oxford and Cambridge. Its members include several Fields Medal winners and recipients of MacArthur fellowships. That concentration of distinction has given the group immediate credibility, but has also prompted concern that it may not reflect the experiences of mathematicians outside the most elite institutions.

A panel with broad powers but few settled details

OpenAI says AGMAI will be able to offer advice without being asked, comment publicly on the company’s effect on mathematics and publish its recommendations. The panel is also expected to receive early access to OpenAI research. Its members will not be paid by the company, and the group can alter its own membership.

The proposed work covers several areas: judging the importance of emerging results, coordinating their publication, considering professional and academic standards, and examining how AI systems might assist mathematical research and education. OpenAI has said the group’s usefulness depends on its members being willing to challenge the company rather than simply endorse its plans.

Yet the arrangement remains loosely defined. The Verge spoke with mathematicians who wanted to know how much authority AGMAI will possess, how its members were chosen and whether OpenAI will follow its recommendations. The panel’s website provides little information about its formation. It says OpenAI initially approached some mathematicians about creating an external board, after which discussions led to the creation of a separate independent organization and the recruitment of additional members.

Martin Hairer, a member of the group and a mathematics professor at Imperial College London and EPFL, addressed those concerns in a blog post. He said not every member had been contacted by OpenAI and maintained that the resulting group is independent of the company and other leading AI laboratories. Apart from ordinary confidentiality obligations, he said, members have not accepted restrictions on their public statements. The Institute for Advanced Study is providing technical support, while the mathematicians receive no compensation from OpenAI.

Hairer said AI has already changed mathematical research and created difficult questions about credit, priority and whether humans understand ideas produced with machine assistance. He argued that refusing to engage with AI companies would not solve those problems. At the same time, he acknowledged that the nine mathematicians cannot be treated as a representative sample of the entire profession.

Questions about representation

The panel’s membership has become one of its central vulnerabilities. Mathematics is divided into highly specialized fields, and a small group may not have the expertise needed to assess results across the discipline. Researchers also differ sharply in how AI is affecting their work, careers and institutions.

Francesco Fournier-Facio, an incoming professor at Heriot-Watt University, told The Verge that involving more mathematicians could improve the situation, but questioned whether AGMAI reflected the people whose work is being disrupted by OpenAI’s pursuit of major mathematical achievements. He described the group as resembling an “ivory tower.”

Simon Machado, a researcher at ETH Zurich, likewise questioned whether highly decorated academics could speak for mathematicians facing more ordinary professional pressures. He said the panel’s creation was “shrouded in mystery” and expressed doubt that its members were the right representatives for broader political and social questions about AI.

Trust will determine whether AGMAI can do more than provide a respected name to OpenAI’s communications strategy. Hairer warned that AI companies could present the panel’s comments in ways that benefit their public-relations efforts. He nevertheless argued that the company’s willingness to engage should be met with an effort to improve its conduct, particularly after criticism of how it has announced and explained recent results.

A large backlog of disputed claims

The panel’s first major assignment is already defined. OpenAI has said an unreleased model that solved the Navier-Stokes Millennium Prize problem also resolved more than 100 longstanding open problems across multiple areas of mathematics. The company has previously indicated that it is approaching another Millennium Prize problem.

AGMAI is therefore being asked to help manage what OpenAI describes as a large group of significant results awaiting release. The challenge is not only whether the claims are correct, but also how they are verified, credited, explained and distributed to researchers.

Hairer said the group is still deciding what advice it will offer. Members are collecting views through discussions with colleagues, a public feedback form and conversations on Proofs and Prompts, a blog focused on mathematicians’ response to AI. The panel’s success will depend on whether that consultation expands beyond its prestigious founding membership—and whether OpenAI accepts scrutiny when the advice is inconvenient.