What's the Future for Pure Math Research in the Age of AI?

Commenters debated whether AI will transform or merely assist pure-math research. Frieren argued AI mostly functions as a thematic miner of human knowledge - useful like Mathematica rather than a replacement - and emphasized that generative models are chatbots, not mathbots, often producing text with the “statistical texture” of papers but low probability of correctness. Others echoed that view: AI might carry out calculations or even proofs, but figures like ghusto and paulpauper said humans still choose why to do math and which problems matter, and AI won’t make publishing easier despite increasing paper volume. Several noted practical limits of current models and the continued need for expert human judgment.

Opinion split sharply over training data, novelty, and long-term social effects. Demibabs, smitty1e, and sedan_baklazhan worried that models either hallucinate beyond their training set or “eat the seed corn” if human-generated future data dries up; many doubted the idea of safe self-training. Lumost raised the opposite risk - that firms could run out of trainers or we might be near genuine superintelligence - while dfydx sketched a bootstrapping scenario where one small AI-generated nugget enables further advances. FeteCommuniste worried about preserving deep human understanding if AIs dominate frontier work, and patcon flagged a copy error in a claim about who decides problems. Overall, commenters split between tool-optimists and skeptics focused on data, correctness, and the survival of human expertise.