OpenAI creates independent math advisory group as AI solves 100-plus problems

OpenAI said Monday it is forming an outside group to weigh in on its mathematics research, a response to mounting criticism from mathematicians about how the company handles its results. The group, called the Advisory Group on Mathematics and Artificial Intelligence, will be hosted at the Institute for Advanced Study in Princeton, New Jersey.

In the same announcement, OpenAI said the internal model behind its disputed solution to the Navier-Stokes Millennium Prize problem has since resolved more than 100 other open problems spanning most areas of mathematics. OpenAI did not name the problems or publish proofs alongside the claim.

The new group arrives after a rough few weeks for OpenAI’s standing among mathematicians. Earlier this month, 25 Fields Medal winners, mathematics’ top honor, signed an open letter warning that AI labs are racing to claim credit for solved problems in ways that undercut the work of human mathematicians. Only one of the nine mathematicians OpenAI named to its new group, the Institute for Advanced Study’s Camillo De Lellis, also signed that letter.

What the group can and can’t do

OpenAI describes the group’s role as advisory: members will help judge how significant new results are and coordinate when and how they get released. They will not be paid, and OpenAI says they are free to voice disagreement publicly and to manage their own membership, which gives them some distance from the company that convened them.

But the group has no power over how fast OpenAI publishes. The company’s own announcement states plainly that “the group will not be responsible for advising us on how to pace our internal progress on mathematics.” The Institute for Advanced Study echoed that limit in its own release, saying it does not have decision-making power at any AI company and that responsibility for a company’s decisions rests with that company.

That leaves the central complaint from the Fields Medalists’ letter largely unaddressed. Their objection was not to AI solving problems but to the pace and manner of the announcements, which they said crowd out the deliberate, peer-reviewed process mathematics normally relies on. A group that can advise but not slow anything down does not change that pace.

OpenAI has not said which of the 100-plus problems the model solved, who will verify them, or on what timeline. The company also has not said whether the advisory group will review the Navier-Stokes claim itself, which drew scrutiny for being published without the usual peer review process. Whether outside mathematicians get to check that work before OpenAI publicizes its next batch of results is the question the new group’s early actions will answer.