AI Is Solving Unsolved Math Problems, and Mathematicians Are Split on What Comes Next

OpenAI’s counterexample to the Unit Distance Conjecture, an open problem in geometric graph theory since 1946, has become a focal point for the field’s AI moment. The Decoder reports that human researchers adapted the core proof technique one week later to disprove another major conjecture.

The broader pattern is no longer limited to isolated demos: AI models are finding counterexamples, spotting patterns, and helping turn arguments into machine-checkable proofs. Epoch AI’s FrontierMath: Open Problems and OpenAI’s Astra model are presented as signs that frontier systems are being aimed directly at unsolved mathematics.

Some mathematicians frame AI as a major workflow upgrade. Abhishek Saha of Queen Mary University of London wrote that frontier AI models are already “at least as good as a solid and indefatigable PhD student” in his area, and said GPT-5.5 Pro helped compress routine work that would previously have taken weeks.

Timothy Gowers, however, describes a more unsettled reaction after GPT 5.6 Pro solved two problems on the first attempt that he had spent considerable time working on. His central concern is the “possible destruction of mathematical culture” if fewer people build the deep expertise needed to understand a rapidly expanding body of AI-assisted results.

The current wave of progress does not mean AI can solve all of mathematics. The article notes that AI appears stronger in areas such as graph theory, while other areas remain resistant.

Epoch AI’s benchmark shows that in the two hardest categories, “Major Advance” and “Breakthrough,” AI has not yet solved a single problem. The six remaining Millennium Prize Problems, each carrying a $1 million prize from the Clay Mathematics Institute, also remain unsolved by AI and humans, and OpenAI’s Astra could not solve them either.

Terence Tao’s view, as summarized in the article, is cautiously optimistic: if AI accelerates theorem proving, mathematicians still need to check, explain, contextualize, and teach the results. He warns that proof scarcity could become proof overload, with results arriving faster than the community can review and absorb them.

Kirwin Hampshire offers a darker interpretation, questioning what remains spiritually and creatively meaningful if AI can complete mathematical work at scale. Together, these views point to the same practical takeaway: AI may change mathematicians’ role from primarily producing proofs to deciding which results matter, how they should be trusted, and how they fit into broader theory.