When AI Does the Proofs, What's Left for Mathematicians?

Large language models have gone from regurgitating textbook math to producing original results in just a few years. Last summer, systems from Google DeepMind and OpenAI earned gold-medal scores at the International Mathematical Olympiad; this year DeepMind’s Aletheia autonomously generated a publishable PhD-level result in arithmetic geometry, and an OpenAI system disproved a conjecture in combinatorial geometry that top mathematicians called journal-worthy. A parallel advance pairs LLMs with proof assistants like Lean, Isabelle, and Rocq, automating the laborious ‘formalization’ step of translating human proofs into machine-checkable code—work that Math, Inc.’s agent Gauss used to help formalize Maryna Viazovska’s Fields Medal–winning sphere-packing proof in days.

The technical milestones frame a more unsettling question for the profession: what is a mathematician for once machines can do the reasoning? For decades, computation accelerated proofs—starting with the contested computer-assisted four-color theorem 50 years ago—but humans still owned the creative core: forming conjectures from intuition, devising proof strategies, and verifying results. AI is now encroaching on all three.

The piece argues the stakes are less about productivity than meaning. Mathematicians describe their work as a slow, deliberate pursuit of understanding—the ‘beautiful’ moment when a hard problem suddenly clicks. If AI removes the struggle, it also removes the source of that satisfaction, forcing the field to reckon with motivation and purpose rather than just output.