AI proofs challenge mathematicians to rethink discovery
OpenAI’s release of hundreds of mathematical manuscripts in October 2026 has intensified debate over the future of mathematical research. A manuscript supplied a proof for a problem Cornell mathematician Justin Moore had pursued for 19 years. Separately, Levent Alpöge announced in November 2026 that he believes he has proved the Riemann hypothesis with AI assistance, describing the experience as both a loss and a new dawn for science.
The debate extends beyond whether machines can produce correct results to whether people can understand them. Stanford cryptographer and a16z Frontier adviser Dan Boneh questioned how humans could comprehend a billion-page proof. Verification tools such as Lean can check logical steps, but making enormous proofs intelligible would require substantial collaboration among mathematicians and computer scientists.
Ittai Abraham of a16z Frontier distinguishes discovery from understanding, drawing on William Thurston’s view that mathematical progress depends on human comprehension rather than theorem counts. Sang-hyun Kim and Yash Lodha have pursued that goal by clarifying a result from OpenAI’s manuscripts.
Other researchers emphasize collaboration rather than displacement. Harvard professor Scott Duke Kominers said AI helped solve a problem he had worked on since 2007 without diminishing the joy of discovery. Frank Vallentin similarly credited an AI model with helping resolve a longstanding lattice covering problem through a simplified, elegant proof.