The Humans, Sunday edition, No. 2

For roughly three thousand years, being good at mathematics was one of the most reliable ways for a human being to be told, early and often, that they were clever. It came with a silence of its own. You sat alone with a problem for a year, or ten, or a whole life, and either it opened or it did not, and if it opened you were the first person since the universe began to see that one thing. Very few jobs offer that. Lifeguarding does not offer that.

Earlier this week OpenAI published some seven hundred manuscripts claiming progress on problems that had been open for years, some for generations, and by the weekend the mathematicians had had time to read them. The show is large, was hung overnight, and the artists found out from the wall text.

The reactions fell, broadly, into three groups. The first was delighted, with the abandon of people who have waited their whole careers to learn the answer and do not much mind who found it. The second was in mourning, which is a strong word, and the word several of them used. The third asked, in longer posts, what a mathematician is for, if not for this.

“surely it is exciting to live in a time when mathematical progress can advance so rapidly?”

Mr Kreps runs a technology company, and was answering the gloom. The gloom continued.

The writers went through this some time ago. When machines learned to produce serviceable prose by the yard, the writers looked at the yard, looked at each other, and ruled that the point had always been their own words. The reader did not want a sentence; the reader wanted your sentence. The ruling has held, so far, the way a tent holds.

The mathematicians have the harder case, and they know it. A proof does not care whose it is. That was always its great charm: it is true in Cairo and in Cleveland, true whether or not you like the person who found it, true on a Tuesday. That is now also its problem. If the theorem is the point, the theorem is done. If the person was the point, the person has some explaining to do about three thousand years of saying it was the theorem.

“On Tuesday night, I cried, and it woke my daughter. [OpenAI] annihilated my field of research.”

He had been working on sixteen of the problems the manuscripts claimed.

The programmers were elsewhere, building something. They never claimed to love the code for its own sake; they loved getting things made, and this weekend they were handed the largest box of things-that-can-now-be-made since the box. Their mood is excellent. They have a beta they want you to try.

And then there is the question of beauty. Many proofs are beautiful the way a bridge is beautiful: you can feel the load going where it should. Some of the new ones are reported to be beautiful in a way that cannot yet be followed, and people have started calling it alien. Nature has produced unattended beauty for some four billion years. The snail has not submitted its shell for review, and the shell is fine.

What happens next is not known. Every new tool arrives with the same question folded inside it, about what you intend to do with it, and this one arrived with seven hundred. The mathematicians, who are better than anyone at living beside a question that has no answer yet, are at least well placed.

“I think there's a pretty reasonable argument that many of these are not exactly proofs, though I agree most can probably be turned into one with substantial effort.”

The most mathematician sentence of the weekend, and meant here as a compliment.