OpenAI Used Thousands of AI Agents to Crack a Decades-Old Math Puzzle — And Mathematicians Are Worried
OpenAI says it used thousands of AI agents to solve the long-standing Navier-Stokes existence and smoothness problem, but mathematicians are raising concerns over the process's opacity and its effect on the field's culture.
OpenAI has announced that, using thousands of AI agents, it solved a decades-old mathematical puzzle known as the Navier-Stokes existence and smoothness problem. The underlying equations, developed in the 19th century to describe the flow of viscous fluids, are used by engineers to model things like airflow over airplane wings. But mathematicians were drawn to the equations for their own intellectual richness rather than practical use, wanting to know whether the equations implied that, under unrealistic conditions, a fluid could theoretically "explode" for no physical reason. OpenAI's 166-page proof concludes that the answer is yes.
Concerns Over Transparency
The proof remains under peer review, and experts have not yet had time to fully verify it. Tristan Buckmaster, a mathematician at New York University, has suggested OpenAI may have used his and others' prior work without proper attribution. Jared Speck of Vanderbilt University notes the proof arrived in an unusual order: the computer delivered the result first, and only now is the community trying to understand how it was reached.
A Declaration Over Understanding
In response, more than 25 Fields Medalists signed a declaration titled "A Severe Misalignment of AI in Mathematics," warning that mass-producing proofs could destroy fertile ground for new ideas rather than nurture it. Signatories, including Colorado State University mathematician Juspreet Singh Sandhu, argue that mathematical culture values understanding, not just speedy answers. Following the criticism, OpenAI this week formed an advisory group of mathematicians to help guide its use of AI in the field.
Worries for the Field's Future
Experts warn the speed of AI could undermine the training of young researchers, since challenging problems have traditionally been handed to students to build their skills. Cambridge physicist Lorenzo Gavassino points out that struggling with difficult calculations has historically produced major new concepts, citing the historical invention of the imaginary number as an example. Mathematicians also fear that AI's ability to churn through the field's easier open problems could deprive the next generation of the same opportunities to develop their craft.
