Trending
Prev Story
Next Story

Why I won’t leave mathematics, despite AI solving centuries-old problems | Ken Ono

The Guardian •
November 22, 2026 • 11:00 AM
3
3 reads 0 reactions

I believe human judgment remains essential

Why I won’t leave mathematics, despite AI solving centuries-old problems | Ken Ono

Last year, I nearly quit mathematics. On 20 November, Levent Alpöge announced that, working with Claude, he had proved the Riemann hypothesis, a 167-year-old problem about the distribution of prime numbers. The Lean and Mathlib communities have since certified the formal proof. We are only beginning to understand the argument. What kind of future awaits someone beginning a life in mathematics?

In May 2025, I led the number theory team for Epoch AI’s FrontierMath project. We wrote difficult problems to test large language models. I watched an unreleased ChatGPT model compute all the moments of Tamagawa numbers of elliptic curves, drawing on work by Manjul Bhargava and Roger Heath-Brown, and on papers my father wrote in the 1960s. I was stunned.

I earned my PhD in 1993 and built a life around research, teaching, and mentoring. I’d love to prove that Ramanujan’s tau function never vanishes, or that half of the partition numbers are even. My 25-year-old self would have been crushed by what we are witnessing today. Nothing I write can reassure someone just starting a career. I can explain why I stayed.

My father, a number theorist, died in January. As a boy, I learned from him about Ramanujan, whose notebooks have occupied mathematicians for generations. Ramanujan believed a Hindu goddess gave him mathematical formulas in his dreams. When I considered working on AI, I feared my father would see it as abandoning mathematics. Instead, he told me AI reminded him of Ramanujan: discoveries from an unexpected source might leave us much to understand. I do not take that comparison literally. Ramanujan was a human with an extraordinary imagination and life. My father’s response brought me to tears. It gave me permission to keep being a mathematician in a way I had not anticipated.

Today, I am on leave from the University of Virginia and work as Founding Mathematician at Axiom Math. We use AI to formalize, verify and extend mathematical proofs in Lean, a language in which a small checking program verifies each step. This work can expose gaps and make proofs inspectable and reusable. Human judgment remains essential: we must check that the formal statement says what we intended.

We released the first complete Lean formalization of the classification of finite simple groups, one of mathematics’ largest proofs. Announced in 1983, the classification required decades of work by many mathematicians; a major gap was filled in 2004. Our AI helped produce 43.1 million lines of Lean code. In the process, we found a problem in a published route to a key theorem and supplied an alternative argument. Lean’s kernel checks the resulting formal proof. The mathematics belongs to the generations who developed it. Our formalizations of the classification and other bodies of math are already helping us build and check further mathematics.

After 24 days of work, I watched the final milestones land on my laptop. I cried. I thought about the people who had made the classification possible, and about the father I could no longer tell. But I am 58. I’ve advised 35 PhD students, 17 postdocs, and hundreds of undergraduates. I’ve had opportunities young mathematicians cannot assume. I cannot ask a graduate student to regard uncertainty as an adventure because I have found rewarding work within it.

At a Berkeley colloquium two days after OpenAI announced its Navier–Stokes solution, a scheduled lecture became a town hall lasting almost two and a half hours. Students asked about credit, transparency, hiring, and how mathematicians whose work came first were treated. Some asked me what I would do to help them. Campus security was locking the building as we left.

The declaration A Severe Misalignment of AI in Mathematics, signed by Terry Tao and other Fields Medalists, warns about rushed announcements, inadequate attribution, and the loss of mathematical understanding. I share those concerns. At a National Academies meeting, I put my position this way: hope is not a plan, and nostalgia is not mentoring.

Universities, journals, employers, and funding agencies must decide what they support and reward. Students and younger colleagues should have a voice while those decisions are still open, as should mathematicians who choose to work without AI. In the discussions where I have a voice, I will argue for that participation.

Students need room to struggle with mathematics and learn to question the tools they use. In my advising, I will ask what a young mathematician understood, originated, or made possible. Access to a powerful system is no substitute for those contributions. For work I help publish, readers should know what has been proved, what remains assumed, and who contributed the ideas.

I want my students, and their students, to develop ideas that are genuinely their own. I want them to have colleagues who listen and the material support to pursue a mathematical life. My work at Axiom is one part of how I am trying to contribute. It does not replace my obligations as a teacher and mentor. I may be wrong about how this will unfold. I am still learning what it means to be a mathematician.

Ken Ono is a number theorist and professor of mathematics at the University of Virginia, currently on leave as founding mathematician at Axiom Math. A former vice president of the American Mathematical Society, he is known for his work on Ramanujan and integer partitions.

These views do not necessarily reflect those of Axiom Math, the University of Virginia, the National Academies, or any other institution with which I am affiliated.

Prev Swipe Story Next

Related Stories

Technology
Build #2612