OpenAI’s Release of Hundreds of Math Results Draws Wonder and Worry

OpenAI has published more than 370 mathematical results produced with its most advanced artificial intelligence models, a release that has impressed many mathematicians and unsettled others who worry about how the work will be checked and who gets to use the tools that made it.

The findings, released Tuesday, span areas including algebra, mathematical logic and theoretical computer science, the Guardian reported. NBC News described a community split between delight and alarm: one mathematician said colleagues should be celebrating in the streets, while another compared the experience to being hit by a tsunami.

The release follows what the Guardian described as OpenAI’s announcement last month that its models had produced a solution to the Navier-Stokes problem, one of the Millennium Prize problems that carry a $1 million reward. That claim, like the new batch, will take time for outside experts to evaluate in full.

The sharpest response so far has come from the Institute for Advanced Study in Princeton, New Jersey, whose mathematics advisers, according to the Guardian, said in a statement that they do not endorse testing the field’s hardest problems on private models that the wider community cannot use. AI systems can now produce arguments that the person who prompted them may be unable to understand, verify or take responsibility for, the statement said, and human understanding of mathematics must remain central to responsible research.

The institute’s advisers also asked AI companies, the newspaper reported, to give mathematicians around the world equitable access to their models. Relying on proprietary in-house systems, they warned, risks a two-tier system in which corporate labs race ahead while the rest of the discipline is left behind.

Some researchers raised a narrower concern about credit. Tristan Buckmaster, a New York University mathematician who has worked on the Navier-Stokes problem, told the New York Times that mathematicians who feed partial work into AI models may be supplying steps that help the systems reach their results, according to the Guardian’s account of that interview.

OpenAI’s position is that the results show what its systems can now do and that it wants to work with the field rather than around it. The company said, according to the Guardian, that it would partner with the Institute for Advanced Study to give mathematicians a voice in how the work proceeds, though it did not say it would stop pointing its models at major open problems.

For a discipline that prizes proofs a person can follow line by line, the questions are practical ones: who checks hundreds of machine-generated results, how quickly, and with what access to the systems that produced them. The answers may shape how mathematics is done for years.