MIT mathematician John Urschel: math is not solved by AI — the field is expanding faster than ever

Math is not solved — it's expanding

John Urschel, MIT assistant professor of mathematics and former NFL offensive lineman, pushes back on the idea that AI is closing in on the end of mathematical inquiry. The field isn't shrinking toward a final answer; it's growing faster than at any point in recorded history.

Urschel is direct on this: frontier AI models can now do what many skilled mathematicians have spent careers doing, but the effect has been to accelerate the frontier outward, not to exhaust it. More is being understood, more quickly, with more ground still ahead. His framing is that humanity has moved from climbing a mountain with ropes to having a helicopter — but there are still uncharted places that need exploring.

“The latest frontier models can do what a lot of good mathematicians have been doing for a lot of years. And the question now becomes, where do we focus our effort?... The amazing thing we're seeing is that we're pushing the frontier of what we understand and know mathematically faster than we ever have in history. And so we're understanding more, we're pushing boundaries further.”

What AI can and can't do in math

Current models are, in Urschel's assessment, very strong at solving well-posed problems that the mathematical community has been thinking about for years. The open question is whether they can do the generative work — identifying which new directions are worth pursuing, which problems matter, which undiscovered territory is worth entering. That, he says, remains unsettled. He declines to make a prediction.

The chess analogy gets a careful rejection. Chess reached a kind of equilibrium after AlphaZero; there are only so many optimal moves, and the field leveled off. Mathematics, Urschel argues, doesn't have that ceiling. The directions for new problems — especially in applied math — are far less bounded, and the field is more likely to explode outward than plateau.

Education as the real upside

Urschel is more optimistic about the education angle than about AI replacing research. He draws a direct parallel to his own experience growing up in Buffalo, where internet access to university lecture notes opened doors that geography and resources would otherwise have closed. AI, he suggests, could do for mathematical depth what the internet did for mathematical access — letting students reach more advanced concepts faster, and leveling the playing field in ways that textbooks and classroom hours alone never could.

Understanding still matters

Even as AI handles more of the computation, Urschel argues that understanding — what a proof or solution means for the broader mathematical ecosystem — becomes more important, not less. The emphasis on grasping ideas, not just producing them, runs through how MIT's mathematics faculty now approaches the field.

On whether mathematicians are still needed: yes, but in a changed role. Communicating mathematics, helping students build tools to solve problems across science and engineering, and working alongside AI systems are where human mathematicians are likely to concentrate effort.

The unresolved tension is generativity. AI can execute brilliantly on problems humans hand it. Whether it can identify the next important problem on its own is the question the field is still waiting to answer.