'In a few years, there will be nothing human mathematicians can do better than AI,' Fields Medal winner warns

A University of California, Berkeley professor who won the Fields Medal — often called the Nobel Prize of mathematics — has warned that if AI continues advancing at its current pace, human mathematicians will soon have no place left to stand.

Richard Borcherds, a UC Berkeley mathematics professor and 1998 Fields Medal laureate, said in an interview with the Daily Californian that "if AI continues to develop at its current rate, within a few years there will be nothing that human mathematicians can do better than AI." The New York Post also reported the remarks Sunday (local time).

The interview came shortly after OpenAI announced that its AI agent appeared to have solved one of mathematics' most celebrated unsolved problems — the existence and smoothness of solutions to the Navier-Stokes equations.

Borcherds said he was at a loss for what to tell students who dream of pursuing mathematics. "I have no idea what to advise," he said, adding that he had posed the very question to ChatGPT — only to find that even ChatGPT said it did not know what the future held.

He drew parallels to past technological disruptions. "There have been many technological breakthroughs in the past. The power loom put hand-loom weavers out of work; the automobile took jobs from horses," he said. "But this is the first time something like this has happened in mathematics."

He noted, however, that universities are slow to change, and predicted that mathematics departments would likely continue operating much as they do now for at least a few more years, regardless of AI's advances.

Borcherds said AI use in student homework has become widespread and that it has forced him to change how he grades. "If you grade on correctness, you give students a huge incentive to use AI instead of doing the work themselves, and honest students get penalized," he said.

He acknowledged the current approach was "not a great system" but said there was no better alternative for now.

He added that while AI-generated work used to be detectable by its characteristic mistakes, that is no longer the case. "Before, you could sometimes spot it from AI's typical errors, but AI has improved so much that it barely makes noticeable mistakes anymore," he said. Online exams in his courses have seen average scores rise sharply as a result, and he now bases grades primarily on in-person tests where electronic devices are not permitted.

UC Berkeley requires professors to specify AI-use rules in their syllabi. The university also considered AI detection software but concluded it was not reliable enough to adopt as a campus-wide standard.

Borcherds said he would be relieved if the worst impact AI had on society turned out to be nothing more than making homework assessment harder. "Compared to the broader changes AI will bring, that would be a small thing," he said.