Mathematicians debate whether correct AI-generated proofs need human understanding

Rohan Paul reports on X that Harvard mathematicians are split over whether a correct AI-generated proof must be understood by humans to count as a real result.

The debate followed OpenAI's claim about a proof related to the Navier-Stokes equations. Paul says proofs both establish that a statement is true and teach methods mathematicians can reuse, and argues that AI proofs often do only the first. Paul does not name the mathematicians or quote their positions.

In a later post, Paul reports that Terence Tao says AI-driven automated proofs of hard problems, including Navier-Stokes, leave the field in disarray and harm its long-term development. Tao calls this approach irresponsible strip mining, meaning it reaches immediate goals at the cost of the field's future progress. Paul attributes the remarks to a New Scientist video interview.