AI’s Problem Solving Is a ‘Profound Disruption’ for the Math Community

Artificial intelligence has profoundly shaken up tech, education, entertainment and many aspects of everyday life. Now it’s come for math.

According to companies like OpenAI, the large language models they’re developing internally are solving an unprecedented number of longstanding mathematics problems with the goal to “push the frontier of human knowledge.“

But one thing these AI models apparently can’t do: read the room with the world’s most brilliant mathematical minds.

This week, Open AI released more than 700 math “preprints,” or preliminary research papers, on GitHub. The move further angered an academic math community already so up in arms that last month, 28 Fields Medal winners put up a website called Math and AI with an open letter accusing AI companies of having goals that are “severely misaligned” with those of mathematicians.

The Fields Medal is only awarded every four years (usually to three or four mathematicians) and is widely considered the Nobel Prize of math. One of the winners who signed the website declaration, Harvard professor Curtis McMullen, told CNET in an email that AI companies are not coordinating with math experts or using AI models in ways that could practically benefit society.

“The rollout of mathematical work by frontier AI companies has been carried out with disregard for the profound disruption to our community and its academic ecosystem,” McMullen said. “Its goal appears to be to increase the valuation of these companies at any cost.”

‘We can tell almost nothing’

In the months leading up to the Math and AI website and OpenAI’s repository dump, steam was already building over the outsized impact artificial intelligence would have on the study of mathematics. Given that these models were largely trained on written language and lacked a grasp of ground truth, some initially had such poor math skills that they could frequently misplace decimals or simply hallucinate incorrect answers to math problems.

But, as with other areas of AI advancement, the models have evolved at startling speed, going from so-called math dunces to wizards, even as some suggested that they still had a long way to go before displacing actual math experts.

Those who’ve devoted their lives to math study, however, aren’t just worried about their job security. They have fundamental issues with how AI models and companies approach math, including a lack of transparency in how they’re achieving these results and a lack of understanding about why these math problems matter in the first place. That’s something OpenAI has not addressed since the announcement, which, according to Scientific American, is piling on to a field “already in shock” just a month after the company solved the Navier-Stokes equation, a math problem so tough it had a $1 million bounty attached.

A spokesperson for OpenAI did not immediately return a request for comment.

In his email to CNET, McMullen acknowledged that the potential for AI as a tool for math research is “undeniable.” But so far, he said, research has not been carried out transparently or in cooperation with scientists. “In particular, the models used to achieve these results should be shared
with the scientists who understand and formulated the conjectures or questions under investigation,” McMullen said.

OpenAI has said that it has formed an advisory group hosted at the Institute for Advanced Study, but has not detailed how that will work or whether it will share its most advanced models, calling it only “a first step.”

Not everyone is upset about AI’s disruption of math society; for instance, Levent Alpöge, a mathematician working for another leading AI company, Anthropic, congratulated the team at OpenAI, posting on X, “There are some sad stories related to their users getting scooped … we should put that aside for today though. It’s obviously the most significant moment in mathematical history.”

But the GitHub drop has left many in math upset and without enough information on how many of this new batch of problems were solved. Gary Marcus, a cognitive scientist and professor emeritus at New York University, wrote on Substack that the publication “would never pass peer review. We don’t know what the procedure was … From the initial report we can tell almost nothing.”

The mass solving of problems, especially in the absence of context, misses the deeper goal of this study, McMullen said. “Mathematics is not a game of chess,” he said. “Its primary aim is to develop human understanding of abstract structures. Indeed, it is the breadth and versatility of this kind of understanding that underpins technological progress.”

Ironically, that progress includes the AI models that have ingested vast stores of human knowledge. It remains to be seen whether their math proofs will advance human understanding or just create artificial confusion.

Omar Gallaga has covered technology, digital culture and other topics for outlets including CNET, NPR, Wired, Texas Monthly, MSNBC, Consumer Reports, The Washington Post, the Los Angeles Times, The Atlantic and the Austin American-Statesman, where he was a longtime tech reporter, editor and podcaster. He lives in the Texas Hill Country.

See full bio