AI-generated math has a human problem

By Hannah Press

AI math advancements have reached yet another benchmark, with OpenAI publishing more than 700 manuscripts from an unreleased model just days after Meta shared several papers mathematicians wrote with its Muse Spark model.

Why it matters: AI is producing math research faster than humans can read it. Earlier conversations revolved around humans guiding AI and what we lose when it moves too fast. The latest calculation to solve for applies well beyond the world of proofs and theorems: what happens if we can’t keep up?

Driving the news: OpenAI’s math manuscripts cover 372 families of results, each of which the company says solves or makes substantial progress on an open problem in math or theoretical computer science. Many of the proofs have been checked by software for errors in their logic. Others haven’t, and OpenAI acknowledges those could contain mistakes.

What experts are saying: On this week’s episode of Mixture of Experts, the panel debated what these latest advancements in AI-generated math mean for the people who have to make sense of it all. IBM

Fellow Kush Varshney said math is more than the fractions and multiplication most of us learned in school. “It’s the thought process that goes into some world that’s defined by some axioms and postulates, and being able to definitively show with no doubt that some property holds,” he said. “It’s a journey of understanding.

Skyler Speakman, Senior Research Scientist at IBM, pointed out that human attention has limits. “We’re finite beings here,” he said. “Is this the beginning of a treadmill where humans [are] sorting out AI-generated content? We’ve seen that play out in other places, but is this where it really takes hold in the mathematics field?”

Between the lines: If AI is doing the work of finding the answers, humans carry the responsibility of making sense of them. Varshney cited a recent Quanta Magazine piece that raised the idea of mathematicians becoming interpreters. “It’s not the humans that are doing the math, but they’re interpreting it, they’re understanding it for the rest of humanity,” he said.

The other side: Utsav Dholakia, Forward Deployed Engineer at IBM, argued that humans falling behind might not be a loss, after all. “Is it really so bad if there is another tool, another sort of virtual intelligent being, that may help us uncover some fundamental aspects of the world around us?” he asked. “Maybe our brains are not even devised to think beyond three or four dimensions.”

What to watch: So far, AI has been tackling questions mathematicians have asked for decades. Speakman posed the next big one: “When will AI be able to ask the new math question?”

Catch the full episode of Mixture of Experts on YouTube, Spotify or Apple Podcasts.