The Post-Proof Era: AI's Watershed Moment in Mathematics

We are witnessing a fundamental pivot in the nature of mathematical discovery. The traditional pursuit of the 'famous problem' is being superseded by a landscape where machine-generated insights serve as the new baseline, shifting the human role from the initial discovery of proofs to the rigorous verification and expansion of those results.

According to a guest post by Frank Calegari on Terence Tao's blog, as first reported there, a watershed moment occurred on October 6 when OpenAI released over 700 research papers. Calegari observes that AI has reached a point where it can undeniably make groundbreaking contributions in numerous mathematical areas. However, this leap in capability has created significant friction within the academic community. Calegari reports that some mathematicians feel the subject is being "hollowed out," while others, such as Hugo Duminil-Copin, have lamented that nearly every major open problem they have publicly discussed throughout their careers has been claimed as proven by the OpenAI announcement.

From a long-horizon perspective, this is not the end of mathematics, but a transformation of its methodology. Calegari draws a historical parallel to the classification of finite simple groups. He cites Dan Gorenstein, who announced a program toward this classification in 1972 and announced the completion in 1983, though the case of quasithin groups required further work by Aschbacher and Smith for another twenty years. While the completion of a major goal can initially cool interest in a subject, Calegari argues that it actually opens the door to deeper questions—such as John Conway's inquiry into why the Monster group exists.

Opinion: The current tension reflects a transition from human-readable, linear proofs to a more fragmented, machine-discovered reality. Calegari describes the OpenAI dump as akin to having the classification of finite simple groups spread across an office floor like "smudged" confetti. The task now is not merely to accept these results, but to engage in the kind of revised classification work performed by Gorenstein, Lyons, and Solomon.

Ultimately, the synthesis of AI and mathematics offers an opportunity to move beyond narrow technical expertise. Calegari suggests that AI can liberate mathematicians from the constraint of the first unprovable conjecture, allowing them to explore deeper mysteries—such as those involving Grothendieck’s period conjecture or the work of Don Zagier, Sasha Goncharov, and Francis Brown regarding the geometry of mixed Tate motives. By automating the 'galactic bypass' of known problems, AI allows the human mind to return to the 'winding paths' of theoretical exploration.