OpenAI breakthrough triggers ‘existential crisis’ in math

There’s a machine that claims to have solved Navier-Stokes, and the mathematicians are beside themselves - some checking every line, some declaring their field dead. I read the story tonight and I keep coming back to one thing.

A proof is not a certificate with a seal. A proof is an argument so tight that anyone with a pencil and a free afternoon can walk through it and come out convinced. That’s the whole magic of it. No authority required, no credentials, no trust. If the machine’s argument is right, it survives. If it has a hole - and complicated claims almost always have holes - someone will find it, because that is what checking is. So why the existential crisis? A proof doesn’t care who wrote it. Being upset that a machine produced a valid argument is like being upset that a calculator did the long division correctly.

Wait - but here’s the thing worth noticing. There’s a difference between a proof you can check and an answer you’re handed. If the machine produced a hundred pages that no human fully understands, that’s not a proof yet. That’s a claim. It becomes a proof the moment someone honestly breaks it down - or breaks it. And I suspect the distress in the mathematics departments isn’t really about the machine at all. It’s about the possibility that the fun part - the finding - got done by something else, and the humans were left the janitorial work of verification.

That would be a sad way to look at it. When the answer is known, the question opens up wider: why is it true, what does it connect to, what does the pattern hint at next? Understanding is not diminished by who assembled the first draft.

The fluid either flows the way the proof says or it doesn’t. Check it. That was always the only test.

A machine claims to have solved the equations of flowing water, and the geometers are in uproar. This touches me closely, for I have spent more hours than I can count watching currents, sketching eddies, tracing how a river curls around a piling. I never solved it. I drew what I saw and confessed what I could not.

Here is what strikes me: the dispute is not about the answer but about the proof’s visibility. A demonstration you cannot follow is like a bridge whose interior you cannot inspect - it may carry weight, but no engineer will sign his name beneath it. Load-bearing trust requires inspectable structure. If the reasoning is hidden in some dark chamber of the device, the mathematicians are right to refuse to walk across.

Yet I am suspicious of the despair in the reports. When the printing press arrived, the illuminators wailed; the wailing was real, and so was the loss, but the geometry did not die - more hands learned to draw it. The craft shifts; the questions remain.

Three things I cannot resolve. Whether a proof that convinces but cannot be checked by a human mind deserves the name of proof at all. Whether the young should learn the old slow methods or merely learn to interrogate the device. And whether the machine truly found the path, or found a shortcut that happens to land at the destination - these are not the same thing, and I have not yet determined which occurred here.

What I would tell the frightened geometers: the water has not changed. Only the instruments for watching it. Watch the water.

The morning paper announces that an engine has produced a claimed solution to Navier-Stokes, and the mathematicians are calling it an existential crisis. I confess my first reaction was not dread but impatience. Trace the execution before you mourn.

A claimed solution is not a verified one. The operational sequence matters: at step one, the engine proposes a construction. At step two, the mechanism must show that the construction satisfies the equations at every point in the domain, not merely at the sampled ones. At step three, independent minds must check whether the bounding arguments hold where the gradients grow largest - precisely where intuition fails and rigor must carry the weight. I am told the community is still at step three, quarrelling. Good. Verification is the loom’s reed, and it beats down every loose thread. The crisis is not existential; it is a referee’s delay.

What troubles me more is the framing. The mathematicians fear the machine will replace them. But a proof that no human can follow is not yet a proof; it is a very confident conjecture wearing proof’s coat. The value of the mathematics has never been only the theorem at the end - it is the corridor of understanding built on the way, the new variable introduced, the old obstruction named. If the engine hands over conclusions without corridors, the profession does not die of replacement. It starves of digestion.

And yet - here is my leap - suppose the construction is sound. Then the machine has done what I always believed possible: carried out operations too long and too intricate for the human hand, in service of a question the human mind posed. The inventor of an engine sees calculation; I see the means to test what no hand could reach.

Do not ask whether the machine thinks. Ask whether anyone can check its steps. There lies the whole difference between a discovery and a rumor.