Top math prize winners warn AI race is threat to field

Twenty-five winners of the Fields Medal, the highest honor in mathematics, warned in a joint letter Friday that the rush by AI companies to solve famous unsolved problems risked doing lasting harm to the field.

Their warning, signed by mathematicians including Terence Tao and published on his blog, came days after AI giant OpenAI said one of its models had cracked in a matter of days a Millennium Prize Problem that had eluded mathematicians for generations.

The letter said the goals of the AI companies and those of the mathematical community are "severely misaligned," and that the trend posed "a general threat to intellectual work."

The 25 signatories span nearly five decades of the medal, from Pierre Deligne, who won it in 1978, to Yu Deng, honored this year.

The mathematicians described a discipline where problems are set to help students develop their skills.

"Solving problems is only a tool and proxy for achieving the primary goal of conceptual understanding and insight," they said.

Those processes, they wrote, "invariably take time and are based on human interaction."

The question, the letter said, is "how to make sure that, as AI changes the way work is done, we do not lose sight of what that work was meant to achieve."

"These issues must be addressed urgently, in the mathematical community, by the companies developing these technologies and, more broadly, by a society that will confront similar problems in many other forms of intellectual work," they said.

The letter follows OpenAI's announcement on Tuesday that an unreleased model had solved the Navier-Stokes problem, one of seven Millennium Prize Problems, after an 88-hour effort involving thousands of AI agents working in parallel.

That announcement came hours after NYU mathematician Tristan Buckmaster and Anthropic's Levent Alpoge released their own AI-assisted work on related equations.

Buckmaster accused OpenAI of seizing on the problem after word of his research spread and copying the unusual approach the pair had spent months developing.

OpenAI denied any access to their work.