AI Is Not Going to Kill My Love of Math

Summary

This first-person essay considers what AI-assisted proofs could mean for mathematics and for people who pursue the subject for discovery, not only for usable results. It opens with recent claims that a company and several individuals resolved the Navier-Stokes problem with AI assistance, alongside a reported counterexample and other rapid short-term developments such as a counterexample to the Jacobian conjecture. The author argues that these results show AI-assisted proof can produce knowledge quickly, while emphasizing that its long-term value for research remains unclear. Drawing on Terence Tao’s warning about premature AI proofs, the essay suggests that solving a difficult problem too early could deprive researchers of the subproblems, concepts, and connections discovered along the way, as happened during work surrounding Fermat’s Last Theorem. The author also accepts that skipping exploratory work may be reasonable in applied fields when the goal is a formal result, and notes that language models can sometimes connect fields or offer new perspectives. A central limitation, in the author’s view, is that current AI-generated proofs often appear terse, dense, and difficult for humans to interpret or find beautiful, although future systems might improve at producing satisfying explanations. The essay worries that an apparent end to unsolved problems, or a frontier pushed beyond human reach, could discourage young people from studying mathematics. It ultimately remains uncertain about the outcome, but argues that learning and understanding mathematics may retain value even when computers provide difficult solutions, much as human-created strategies remain meaningful in computer-dominated chess.