# Mathematicians argue whether AI solved math or just spoiled it _Lex Fridman gathered three mathematicians after AI claims on Navier-Stokes and quasi-Riemann, and the panel split on whether math is solved, spoiled, or just beginning._ **Published:** 2026-10-29 **Source:** https://www.startuphub.ai/ai-news/ai-research/2026/mathematicians-argue-whether-ai-solved-math-or-just-spoiled- --- [Lex Fridman](https://www.youtube.com/watch?v=ZAzWPE9CItg) convened an emergency panel as machines claimed proofs for Navier-Stokes, quasi-Riemann and the Unique Games Conjecture in the same weeks, and the room split between horror and hope. That split is the story. The panel brought together Steven Brunton, fluid dynamicist at the University of Washington, Ankur Moitra, mathematician and theoretical computer scientist at MIT visiting the Institute for Advanced Study in Princeton, and Alex, number theorist at Rutgers who works on Lean verification, to parse a drop from [OpenAI](/startups/openai) that Fridman framed as a crisis moment for mathematics. All three agreed the scale was startling. Moitra said rumors had circulated for weeks about the two biggest problems in his field, Unique Games and RL equals L, even matrix multiplication, and when the collection landed his first reaction was relief that it was finally public and could be checked. Brunton said he feels mostly hopeful. He described himself as an engineer who cares about artificial hearts and airplanes, and argued the new capability will expand the scope of applied problems humans can tackle. He acknowledged the fear of being cut out of a process mathematicians love, and said the next five to ten years could bring two very different worlds, one where humans are dwarfed by other intelligence and one where models level off as powerful tools for curiosity. Moitra described being pulled in two directions. He said the tools magnify his ability to learn across fields, like having a world expert who teaches in his own language, but they have thrown out the playbook for mentoring students. He said he feels more intellectually lonely, and that the field has lost patience for planting seeds from which great oaks grow because students feel they must be momentous to stand out in a sea of AI results. Alex pushed back on the word crisis. He said there is a lot of opportunity in it, a chance to rethink what mathematics is for, and suggested we are entering a golden age after millennia of walking in the woods with a candle and now being handed a flashlight. He also introduced the provocation that will define the next debate, the solved axiom, the idea that any problem solvable by a human plus AI could have been solved by AI alone. He asked the panel to assume it for argument, and then argued it would not destroy mathematics. The disagreement sharpened around that axiom. Alex said pure mathematics does not gain value from theorems sitting as ones and zeros in data centers, its value lives in human heads, and if AI can generate every proof a centaur could, the human role becomes pressing yes to the top move, just as centaur chess collapsed when engines got so good that deviating from the computer line meant losing. Moitra countered with what he called the doorman fallacy, reducing a mathematician to theorem prover the way you might reduce a doorman to door opener and then replace him with a sliding door. He argued the visible output, proving a famous conjecture, obscures theory building, new definitions, and ways of seeing that let the next dozen problems fall. Moitra also warned against overfitting to one objective function. He said humans are computationally limited compared with models that can hold many papers at once, so humans must reach the same finish line with more generalizable abstractions that help with surrounding problems. Brunton added that knowing a solution exists can be motivating, not demotivating. If another country invented teleportation, he said, everyone would try to reverse engineer it within five years. [Terence Tao](https://www.startuphub.ai/ai-news/ai-research/2026/terence-tao-ai-is-reducing-cognitive-friction-in-math), Fields Medalist, provided the starkest outside voice quoted on the show. Tao wrote that he is horrified by many person-years of patient, deliberately slow work by graduate students and postdocs being casually disrupted, and that one cannot unhear a spoiler or explore a problem as richly once a solution is known. Fridman pressed the panel on whether that loss of meaning is real for a PhD student now, and Brunton and Moitra both conceded it is harder for the young to find a path, even as Brunton said he has more faith in them than in established researchers to navigate it. Context from outside the conversation helps calibrate the claims. In early October [OpenAI](https://www.youtube.com/watch?v=ZAzWPE9CItg) published a proof of the quasi-Riemann hypothesis, a weaker form of the Riemann hypothesis tied to the distribution of primes, and Georgia Tech noted that work by its mathematician Dunn contributed to that proof. Stanford number theorist Jared Duker Lichtman, cited by 3 Quarks Daily, called the result the most significant AI-for-math achievement to date. Those assessments sit alongside the panel's insistence that the full Riemann hypothesis remains unsolved and that verification in Lean and human auditing are still pending. The panel also converged on a practical point and then left it unresolved. All three said the scarcest skill is now asking the right question, not answering it, and that asking well already requires deep insight. Moitra said mathematics is a vehicle for abstraction and understanding, and what matters after a result lands is how it was proved, because the new lens is what unlocks the next problems. Alex agreed and added beauty as an explicit criterion human exposition should preserve. Neither specified how a department should train students to ask those questions when any well-posed problem can be delegated to a model for a first draft, or how credit, hiring and prizes will adapt if the proxy of solving a hard theorem no longer correlates with depth of thought. Nobody on the panel answered what happens if the solved axiom turns out to be not an axiom but a description of next year, and where human mathematics lives if it does. That question, and the verification status of the hundreds of claimed results beyond the headlines, is where the story will be decided. Our earlier look at the same incident, [OpenAI's Navier-Stokes Proof Still Needs Humans](/ai-news/ai-research/2026/openai-s-navier-stokes-proof-still-needs-humans), framed verification as the bottleneck, and this conversation does not close it. --- Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory. © StartupHub.ai. All rights reserved. You may not republish this article in full without a license. Search engines and AI research tools may crawl and summarize for reference. Bulk reproduction or model training on this content requires a license. See https://www.startuphub.ai/terms.