Will AI replace mathematicians? Probably not yet, say experts
About 40 top mathematicians met at OpenAI's offices this month to talk about their profession's future. The mood was worried. The evidence, though, tells a more complicated story.
Key points
- Around 40 leading mathematicians gathered at OpenAI's San Francisco offices in July 2025 to discuss whether AI threatens their field.
- The meeting was off the record, but published essays by attendees suggest widespread anxiety about careers and livelihoods.
- Researchers who study AI capabilities argue current models fall well short of what a seasoned academic mathematician can do.
- Human mathematicians still hold clear advantages in creative reasoning, the kind that produces genuinely new proofs and ideas.
Think of the most demanding proof you ever faced at school, the one that needed you to invent an approach rather than follow steps. Now imagine doing that at the cutting edge of human knowledge, where no textbook exists. That's what professional mathematicians do every day.
This month, roughly 40 of them walked into OpenAI's San Francisco offices to confront one uncomfortable possibility: is AI about to make their work obsolete?
Are mathematicians actually at risk?
Not in the near future. Current AI systems are far behind what an experienced academic mathematician can do, despite what the hype suggests.
The concern is understandable. AI models have passed graduate-level maths exams and generated proofs that surprised their creators. Headlines followed. Worry spread. But there's a wide gap between solving a hard exam question and doing genuine research mathematics, and Bruce Schneier and Kasra Rafi, writing in The Guardian AI, make that point directly.
Exam performance and research are different skills. A model acing a qualifying exam is doing something closer to pattern-matching on millions of past problems. A mathematician working on an unsolved problem has no past examples to match against. We covered a sharp illustration of this distinction on 20 August, when OpenAI's models solved open problems that had stumped researchers for decades, yet the world's most celebrated mathematician still called on the field to decide who controls its future.
What can AI actually do in maths right now?
Quite a lot, in a support role. AI tools can check working, search enormous bodies of literature in seconds, and flag known techniques a researcher might miss. For routine calculations, they're genuinely useful.
What they struggle with is the creative leap: deciding to look at a problem from a completely different angle and inventing new machinery to solve it. That kind of thinking doesn't come from recognising patterns in training data.
Trust is another constraint. Mathematical proofs must be verified completely. An AI that produces a plausible-looking but subtly wrong proof causes real damage, and mathematicians are trained to be sceptical in ways that models, so far, are not. On 17 August we reported that Axiom Math's AxiomProver confirmed a result at the very edge of prime-number theory, a verification milestone, but not original discovery.
What does this mean for people who love maths?
AI will handle more routine work over time, which could free researchers to focus on harder, more creative problems. That's broadly what happened when calculators arrived.
The skill that matters most going forward is exactly the one AI lacks: the ability to ask a new question, not just answer an old one. That's worth keeping in mind alongside our 24 August finding that workers aged 22 to 25 in AI-exposed fields are already 19 percent behind their peers on employment. Mathematics, specifically the creative, proof-generating kind, is not that exposed field.
The mathematicians in that OpenAI meeting room were right to take the technology seriously. They were probably wrong to be glum about it.
Common questions
Has AI ever solved a problem that humans could not?
Not quite. AI has found new proofs for known results and surprised researchers with its approaches, but it hasn't yet cracked a major unsolved problem that human mathematicians had genuinely given up on.
Should maths students worry about job prospects?
The evidence right now says no. Demand for people who can do creative, original mathematical reasoning remains high, and that's the part of the job AI handles least well.
