AI in Mathematics: AI Works for Humans, and on Them

In May 2026, OpenAI published a counterexample to the "Unit Distance Conjecture", refuting a conjecture from geometric graph theory that had been open since 1946. This followed other AI-assisted solutions to long-standing problems, marking a turning point. Mathematicians like Abhishek Saha now use AI models to handle routine work, describing their role as a conductor rather than the entire orchestra. Some, like Trefor Bazett, see the potential for a new golden age, referencing a Carnegie Mellon paper that solved a Ramsey theory problem using SAT solvers, LLM-generated code, and formal proof verification. Timothy Gowers, who predicted this era in 2000, has mixed feelings, having experienced AI solving his own research problems. He fears the destruction of mathematical culture if humans stop building expertise. The Leiden Declaration on AI and Mathematics, signed by over 3000 mathematicians, seeks to set guidelines. Epoch AI reported progress but noted that the hardest benchmark categories remain unsolved. Terence Tao compares the situation to the foundational crisis of the early 20th century, seeing it as a chance to strengthen mathematics, while Kirwin Hampshire's essay "The Dark Night of Mathematics" offers a more pessimistic view, suggesting that consolations are self-deception.