25 Fields Medalists Warn AI Is Making Mathematics Shallower — and Say Other Fields Are Next

Twenty-five Fields Medal winners have signed a joint statement warning that artificial intelligence is degrading mathematics by rewarding solved problems over genuine comprehension — and they argue the same pressure is closing in on every other line of intellectual work. The signatories, among the most decorated mathematicians alive, say the aims of the AI industry and the aims of their discipline are 'severely misaligned.' The statement was reported by The Decoder.

The core of their complaint is about incentives. Machine learning systems can now churn out proofs, answers and worked solutions at a volume no human department could match, and the surrounding industry treats that output as progress. The medalists counter that mathematics advances through understanding — knowing why a result holds, where it connects, and what it opens up — not through stockpiling finished answers. A flood of correct-looking solutions, they suggest, can leave the field less capable than it was before.

That argument carries weight because of who is making it. The Fields Medal is awarded every four years to a handful of mathematicians under 40, and its recipients typically go on to shape the direction of major subfields. Twenty-five of them signing one document is not a routine complaint about new software; it is a coordinated signal from the top of the profession that the tools now being sold as research accelerators may be eroding the thing they are meant to assist.

The statement also widens the frame beyond pure math. The signatories describe the problem as a symptom of a broader threat to intellectual work, one that reaches any field where output can be measured and generated faster than it can be understood. That list is long: theoretical physics, parts of biology, law, economics, and any research area where publication counts and benchmark scores stand in for insight. If the diagnosis holds, the damage will not stay inside university math departments.

What remains unsettled is what the mathematicians want done about it. The statement does not propose a specific fix — no funding condition, no publishing standard, no benchmark reform — which leaves open how institutions should respond when AI-generated results keep arriving faster than anyone can verify them. The test ahead is whether hiring committees, journals and grant agencies treat understanding as the metric that matters, or keep rewarding the volume that AI makes cheap.

The warning lands as AI labs increasingly pitch their systems as partners in scientific discovery, with formal math proof-solving held up as a flagship capability. If the people who know the field best are saying those benchmarks measure the wrong thing, the burden shifts to the labs to show otherwise. Watch whether any of the twenty-five attach their names to concrete policy proposals, and whether the next round of AI math claims addresses the distinction they are drawing.