OpenAI says its artificial intelligence systems have produced a solution to one of the Millennium Prize Problems, the small set of famously unsolved questions that has defined the frontier of modern mathematics for a quarter century. The claim, reported Tuesday by The New York Times, would be an extraordinary milestone for machine reasoning â if it survives the scrutiny of the mathematical community.
That scrutiny is the part that matters, and it has not happened yet.
What the Millennium Problems are
In 2000, the Clay Mathematics Institute named seven problems it considered the most important open questions in mathematics and attached a $1 million prize to each. The list includes the Riemann hypothesis, the P versus NP problem, the NavierâStokes existence and smoothness problem, the Hodge conjecture, the Birch and Swinnerton-Dyer conjecture, the YangâMills existence and mass gap problem, and the Poincaré conjecture.
Only one has been resolved. The Russian mathematician Grigori Perelman proved the Poincaré conjecture in the early 2000s, then declined both the prize money and the Fields Medal. The other six have resisted generations of effort by the world’s best mathematicians, and several are considered decades away from resolution by conventional means.
That is precisely why a claim of this kind arrives with such weight â and such skepticism.
Why verification is the hard part
In mathematics, announcing a proof and having a proof are different things. Major claimed solutions to famous problems appear regularly, and most collapse under expert review. Even legitimate breakthroughs can take years to confirm: Andrew Wiles’s proof of Fermat’s Last Theorem contained a gap that took roughly a year to repair, and Perelman’s work required teams of mathematicians several years to fully verify.
A proof produced or assisted by an AI system raises additional questions. Is the argument written in a form human referees can follow and check? Has it been formalized in a proof assistant such as Lean, which can mechanically verify each logical step? How much of the work was generated by the model versus guided, corrected, and assembled by human mathematicians? And can the result be reproduced?
The Clay Institute’s own rules also impose a slow timetable: a solution must be published in a peer-reviewed journal and accepted by the broader mathematical community for a sustained period before a prize is awarded. Nobody is collecting a check this week.
The broader AI stakes
The announcement lands amid an intensifying race among AI labs to demonstrate genuine reasoning rather than pattern matching. Advanced mathematics has become the preferred proving ground, in part because results are objectively checkable and in part because it is difficult to argue that a system merely memorized an answer that no human has ever written down.
AI systems have already made real contributions to mathematics in recent years, solving competition-level problems, assisting with formal verification, and helping researchers find constructions and counterexamples. A full solution to a Millennium Problem would be a step change from that â evidence that machines can operate at the level of original, deep mathematical discovery.
It would also raise uncomfortable questions for the field about authorship, credit, and what mathematical understanding means when the argument comes from a system that cannot explain itself in ordinary terms.
For now, the appropriate response is the one mathematicians have applied to every such claim: show the proof, and let the community check it. The answer will not come from a press release. It will come from the referees. Read More
Leave a Reply