AI cracked the proof no human understands. Now what's a human? Navier-Stokes solve forces the question

A fluid broke on Sep 5. Not in a river or pipe. But in the Navier-Stokes equations that describe every river and every pipe. A vortex spun inward, stretched thin, and reached a point where speed became infinite and mathematics stopped making sense. Nobody watched it happen. 10,000 software agents, working for 88 hrs, found it.

A proof-checker confirmed it. Then, OpenAI told the world that the Navier-Stokes problem, a question open since 1934, and one of the 7 Millennium Prize problems, was solved courtesy an internal OpenAI system that was more powerful than its latest GPT-6 Astra model.

Also read: US lawmakers probe OpenAI over AI ‘going rogue’ as fears of human control grow


Whether fluids can break turned out to be the easy part. Two mathematicians had spent a year on the same road. New York University professor Tristan Buckmaster and Anthropic's Levent Alpoge were following an idea from two Spanish mathematicians, Diego Cordoba and Luis Martinez-Zoroa. They had machines helping them, and got most of the way.

Then a much bigger machine, pointed at the same road after word of their progress leaked, got there first. What followed was calls on a Sunday, an offer to publish if one name was dropped, and a question about whether their own private drafts had taught the machine that beat them. OpenAI later said it could not rule that out.

Then, something larger happened. A proof exists that no human understands. Not the people who wrote the prompts. Not the people who checked it. The proof is correct the way a locked box is heavy. You can weigh it, but you cannot open it.

UCLA professor Terence Tao put it succinctly at the International Congress of Mathematicians in Philadelphia in July: 'In the past, we didn't emphasise the process. We let the outcomes speak for themselves. This worked until we figured out a way to automate outcomes without process. Current [AI] tools are very opaque about their process.' For all of history, the answer and the understanding came together, because the same mind produced both. Now they can be separated. The answer arrives, the understanding does not.

So, ask the old question again. What is a human? Not the fastest solver. That title has passed and won't come back. Not the widest reader either. The machine has read everything. What remains is smaller and harder to name. A human is the one who chooses the question, decides that this problem, and not that one, is worth a life, asks what the answer means, and for whom, and whether it was worth what it cost.

Isaac Asimov saw this coming in 1956. In his short story, 'The Last Question', people ask a computer, across the ages, how to stop the universe from running down. The computer keeps saying it has insufficient data. Ages pass. Humanity fades. At the end, with no one left to hear, the computer finds the answer and speaks it into the dark. Asimov meant it as wonder. It is a warning. We outsourced the last question and forgot to stay for the answer. That is the future, if we let it arrive on its own. Four things follow.

Answers get cheap, questions get expensive When any proof, essay or plan can be had for the asking, the rare thing is knowing what to ask for. The scarce skill sets the price. Judgement will be paid what calculation used to be paid.

What we do not practise, we lose A society that lets machines do all its proving will one day be unable to check them. A certificate of correctness then becomes a matter of faith, which is to say a priesthood.

Fights will not be about truth Fights will be about names, about who was first, about whose work fed whose machine. Machines cannot settle those, because they are not about the world but about us.

The machine didn't do this alone In the case of Navier-Stokes, the idea was Cordoba's, the push was Buckmaster's and Alpoge's, the training was everyone's, every textbook ever written, and every proof ever published.

For 300 yrs since Newton, people made knowledge and machines stored it. Now, machines make it, and people must hold it. Holding is not a lesser job. It is explaining, teaching, refusing, deciding what to build and what to leave alone. It is being the one who understands, in a world where understanding is no longer required for an answer to be right.

(The writer is a public policy professional)