OpenAI Solved a Millennium Problem. What Did We Lose in the Process?
When my daughter, a self-described math nerd, sent me a text attaching an article that announced that OpenAI had cracked one of the Millennium Prize Problems, she added a single word: “concerning”.
Intrigued, I started looking into it. OpenAI had solved what's called the Navier–Stokes existence and smoothness problem, one of seven Millennium Prize Problems selected by the Clay Mathematics Institute in 2000. It has something to do with fluid flow, and with whether the equations that describe it can predict an impossible state: infinite velocity. The math itself is way, way above my head. But it was a big deal as this was a problem that no human mathematician had been able to solve before. And it got me thinking: should we be asking AI agents and in this case, 10,000 of them, running in parallel, to solve problems that have, to this day, been impossible for human mathematicians to crack? Is something lost when we do?
Almost as soon as OpenAI made the announcement, it was engulfed in controversy: accusations that the company had used unpublished work by NYU mathematician Tristan Buckmaster, among others, without proper credit. OpenAI has denied this. But the controversy, to me, is beside the point.
What I find more interesting is something the French philosopher Simone Weil wrote as it relates to attention: that the important thing is often not the answer to a problem, but the time we spend with it. In other words, it’s not always the answer to a difficult question that is the most valuable part of the exercise: it's the struggle, the work, and the attention we pay in trying to solve it.
So when OpenAI uses 10,000 agents and millions of dollars to hand us the answer to a problem that had been unsolvable, what are we losing?
This isn't unique to this one story. It applies to everything AI touches. AI can give us quick answers, and those answers may come with real benefits but this convenience comes at a cost. The cost is that it robs us of attention itself: the time we would have spent sitting with a problem, and everything that time might have produced along the way.
UCLA mathematician Terence Tao had recently written in a Mastodon thread about the importance of mistakes, wrong directions, and incomplete solutions in the field of mathematics. “In most cases in pure mathematics, the problems are posed not because we desperately want the solution to these problems in and of themselves, but because we have seen from past experience that human-directed efforts to solve these problems tend to spur further development of the field,” Tao wrote.
Tao was channeling Weil's teaching here: that sitting with the question may be more valuable than the answer itself. OpenAI's agents got to the answer. Congrats. But what they skipped was everything the waiting would have taught us along the way. The discipline was never in finding, it's in waiting well. Not passively, but by paying attention through it all.
