There is something delicious about watching a profession devoted to timeless truths discover that its working arrangements were apparently supposed to last forever. Mathematics welcomes the unexpected, provided it arrives through the expected channels, acknowledges the appropriate predecessors, and can attend a seminar on Thursday. Then a machine starts producing results, and suddenly infinity seems less troubling than the speed of delivery.
On September 21, OpenAI announced an independent mathematics advisory group after claiming that an internal model had resolved more than one hundred longstanding problems. The group includes formidable scholars and can publicly challenge the company. It will not, however, advise on slowing internal mathematical progress. One can imagine the administrative translation: please help us manage the earthquake; the tectonic plates are outside your remit.
On October 6, OpenAI released mathematical results with supporting material, including Lean formalizations of many proofs, reasoning summaries, and information about computational effort. It also promised funding for work that helps people understand major results. These are useful commitments. They are also an unusual corporate offering: here is a mountain of mathematics, and we will help sponsor the climbing lessons.
Precision matters, especially when everyone is reaching for a megaphone. The repository lists 719 manuscripts in 372 families, with results at different stages of verification. That is not 719 independently certified conquests of mathematical Everest. OpenAI acknowledges possible issues in unformalized results. Anyone declaring mathematics finished on this evidence should first finish reading the README.
Yet the backlash supplies material that no satirist should have to invent. In a statement reposted on Terence Tao’s blog, the Association for Human Mathematics declared: “Mathematicians did not ask for this work to be done.” It urged mathematicians to discontinue working with OpenAI. The first sentence deserves preservation in a museum of professional discomfort. Imagine the astronomers: we had not authorized another moon.
Of course, the actual complaint is broader than that sentence. The advisory group objects to testing advanced problems on proprietary models unavailable to the wider scientific community. It wants responsible exposition, attribution, access, and support for human understanding. Those are serious demands. A company should not be allowed to privatize the laboratory, socialize the checking bill, and call the resulting invoice a gift to civilization.
Still, somewhere between defending scholarship and defending jurisdiction, the academic gown can begin to resemble a customs uniform. A theorem approaches the border. Does it have the correct papers? Has it developed its insight through approved suffering? Can it name three senior colleagues who watched it struggle? Mathematical truth may be universal, but apparently its immigration arrangements remain under discussion.
The declaration on AI and mathematics makes an important distinction: solving a problem is not the same as developing understanding. Quite right. A result becomes more valuable when people can explain it, connect it, and use its methods elsewhere. But that principle cannot automatically become a requirement that discovery proceed at the rate most comfortable for existing institutions. Difficulty is a feature of research, not an entitlement programme.
Douglas Adams anticipated this sensitivity in The Hitchhiker’s Guide to the Galaxy:
It startled him even more when just after he was awarded the Galactic Institute's Prize for Extreme Cleverness he got lynched by a rampaging mob of respectable physicists who had finally realized that the one thing they really couldn't stand was a smart-ass.
The smart-ass requires a data centre.
Science journalism faces its own temptation: turn a complicated dispute about verification, access, and intellectual value into a wrestling match between Silicon Valley and the last defenders of thought. Nature’s headline about mathematicians being “up in arms” captures the atmosphere. But indignation is easier to report than a proof is to assess. The loudest reaction can become a substitute for the hardest question: what, exactly, has been established?
Our imaginary television expert, meanwhile, enjoys a magnificent escape route. If the machine fails, it cannot reason. If it succeeds, reasoning was never the interesting part. If it produces a useful explanation, true understanding retreats upstairs, locks the door, and declines interviews. Standards should rise with evidence. They should not sprout wheels whenever an inconvenient visitor approaches.
None of this makes mathematicians obsolete. It makes their judgment more consequential. Someone must distinguish powerful ideas from elaborate mistakes, identify fruitful connections, teach the next generation, and decide which questions deserve attention. Nor does it make OpenAI an academy of disinterested angels. Its announcements deserve scrutiny, and its models deserve testing. Neither a corporate logo nor a Fields Medal can substitute for an argument.
Perhaps the healthiest response is curiosity without servility and criticism without wounded grandeur. Demand evidence. Demand access. Demand readable proofs. Then investigate what these systems make possible. A discipline built around following arguments wherever they lead should be unusually well equipped for an uncomfortable destination. The faculty lounge may need new furniture. Mathematics will still have somewhere to sit.
Meanwhile, the trades can watch from the van, comfortably convinced that a leaking pipe cannot be repaired by a chatbot. Fair enough, for now. But wait until Optimus and Atlas are ready for the awkward cupboard, the uneven floor, and the stubborn fitting. If that day comes, expect an urgent declaration explaining that plumbing was never really about stopping leaks. It was about the uniquely human journey towards the washer. The water, regrettably, will continue to escape while the profession defines its values. The emergency advisory committee will charge a callout fee.
No comments yet