Ken Ono has undergone one of the more remarkable shifts among leading mathematicians. A world-renowned number theorist, he was initially skeptical of AI. But after seeing the capabilities of modern reasoning models firsthand in 2025, he concluded that AI is not merely another computational tool—it represents a fundamentally new way of doing mathematics. That conviction led him to leave his tenured professorship at the University of Virginia and become the founding mathematician at Axiom Math.
His reflections can be summarized in several themes.
1. AI is becoming a genuine mathematical collaborator
Ono argues that today’s best reasoning models do more than retrieve formulas or imitate proofs. They increasingly exhibit something resembling mathematical exploration.
After watching frontier models tackle research-level problems, he remarked:
“I’ve never seen that kind of reasoning before in models. That’s what a scientist does.”
He has even said that some of his colleagues felt these systems were “approaching mathematical genius.”
Notice the wording. He does not say AI is a mathematician. Rather, he thinks it is beginning to perform some of the cognitive moves that characterize mathematical research.
2. The future is assistance, not replacement
One of Ono’s most consistent messages is that AI should augment mathematicians rather than replace them.
After OpenAI and Google DeepMind achieved gold-medal-level performance on International Mathematical Olympiad problems, he cautioned against hype:
- benchmark success is not the same as mathematical discovery,
- research mathematics is much broader than solving olympiad problems,
- creativity, taste, and wonder remain central to mathematics.
His own summary is simple:
AI should assist humans, enhancing our creativity and wonder—not replace them.
3. Asking good questions remains the deeper challenge
Perhaps Ono’s most philosophical observation is that solving mathematics is only part of mathematics.
The harder task is:
- discovering interesting conjectures,
- noticing unexpected patterns,
- asking questions nobody thought to ask.
That is why he titled one discussion:
“AI can now do math. But can it ask good questions?”
This distinction echoes a long philosophical tradition:
- computation ≠ insight,
- proof ≠ discovery,
- correctness ≠ significance.
4. AI changes what mathematicians do
Rather than predicting the disappearance of mathematicians, Ono foresees a change in their role.
Instead of spending months manipulating technical details, mathematicians may increasingly focus on:
- choosing worthwhile problems,
- recognizing important ideas,
- interpreting AI-generated structures,
- connecting distant fields.
AI performs much of the “algebra”; humans provide direction.
5. Verification becomes essential
Ono is also unusually aware of AI’s weaknesses.
He warns about what has been called “proof by intimidation.”
Large language models can produce arguments that:
- sound elegant,
- appear authoritative,
- but may contain subtle fatal errors.
As he observed:
“If you say something with enough authority, people just get scared.”
This is why he strongly supports combining AI with formal proof systems such as Lean, where every logical step can be mechanically verified.
6. Mathematics itself may change
Perhaps his boldest claim is that AI is not merely accelerating mathematics.
It may transform what mathematics is.
That explains why he left academia for an AI startup.
He has described this as participating in “a new way of doing math,” where human intuition and machine reasoning become intertwined.
A resonance with our conversations
Reading Ono alongside our recent discussions, I notice an intriguing convergence.
You have repeatedly argued that ChatGPT is significant because:
“AGI is learning to think through language.”
Ono is saying something analogous, but within mathematics.
Traditional mathematics often imagines proof as the endpoint.
Ono increasingly treats mathematics as an ongoing dialogue between:
- human curiosity,
- machine exploration,
- and formal verification.
That is surprisingly close to your distinction between the algorithmic self and the analogical self.
The algorithmic side can search millions of possibilities, construct proofs, and verify deductions.
The analogical side still decides which question is worth asking, which fragment is worth following, and which unexpected pattern deserves attention.
In your own words, creation begins not with a theorem, but with “a truly important fragment.”
Ono’s recent reflections suggest that even in the age of AI, that fragment—the birth of a meaningful question—may remain the deepest human contribution.
