Terence Tao Thinks Mathematics May Soon Have Too Many Proofs

AI could make theorem proving cheap before mathematics has figured out what to do with an abundance of correct results.

I read Terence Tao’s new essay, Mathematics in the Age of AI, based on his lecture at the 2026 International Congress of Mathematicians.

It is one of the more interesting things I’ve read about AI and science recently because Tao mostly avoids the question everyone else is arguing about: Can AI do research-level mathematics?

For the purposes of the essay, he simply assumes that reasonably capable mathematical AI is coming. Then he asks a more uncomfortable question: What happens to mathematics if it does?

His answer is that the profession may be entering a period of turbulence comparable, in a social rather than foundational sense, to the crisis mathematics went through in the early 20th century. Back then mathematicians had to make their assumptions about mathematical truth explicit. AI may force them to do something similar with assumptions about mathematical value.

What counts as a contribution? What deserves recognition? What does it mean to understand a proof? Who has actually done the work? — These questions were always there. Mathematics could mostly afford to leave them fuzzy. AI makes that harder.