We can now produce in two hours what used to take us a week. We are getting used to that. But something that happened in math this week pointed out the actual problem we need to solve for.
On September 6, OpenAI announced that its agents had solved the Navier-Stokes problem. It is one of the seven Millennium Prize problems, the list the Clay Mathematics Institute put out in 2000 with a million dollars attached to each one. Mathematicians had been stuck on this one for decades. The AI solved it in about four days.
My degree is in mathematics, and it has always been a passion of mine. So this topic stood out within the AI news and piqued my interest in a different way. It made me ask myself:
My first reaction was that this is phenomenal, and I still think it is. A problem some of the best mathematicians alive spent whole careers on was closed in four days. And the thought right after that is the obvious one: if it can do this, what else can it do now? There is a much longer list of problems in engineering and medicine that nobody has solved yet. If four days is what it takes now, can we point it to all of these and expect a solution?
Then I read Terence Tao on this, and it made me look at the whole thing differently. He says: “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.”
Spending time and effort on these problems is how we advance. Decades of mathematicians failing at Navier-Stokes gave us new methods, whole areas of mathematics that did not exist before and a lot of very good mathematicians who got that good by working on this one thing for years and not solving it.
So AI solves it now, but is this where the value truly is?
Not in this one. Nobody outside the AI lab can see what the model actually did, and that is what Tao objected to.
But I don’t think this is the main problem. You can already ask a model to take you through its reasoning, and I assume we will get to a point where we can sit with these systems and go through the whole path: what it tried, where it got stuck, why it dropped one idea and kept another. So let us say we can.
We still have a problem.
It takes someone who has been through the process, through the trenches, to sit with what the AI produced, discuss it, pull it apart and have a chance of getting back what we lost by skipping the process.
But what about the generations after this one, the ones who have had AI helping them the whole way? Will they have what they need to build the same skill? I am not sure, but there is a chance they may not.
Someone on your team builds a campaign plan in two hours that used to take a week. The thinking happened somewhere in there: what was compared, what was thrown out, why an audience was dropped. None of it is visible to the team, to whoever reviews it or, honestly, even to the person who made it.
Writing it all down does not fix that, because you still need someone who can read it and find where it goes wrong. That person got good by doing it the slow way for years, and the slow way is the first thing we cut.
So what does this mean for our jobs? What is going away is the value of being the one who designs the output. AI does that now, and being excellent at it is not worth what it used to be.
Then where is the value now? I see three things.
If you can answer those, it is because you spent years doing the work by hand. Nobody gave you that. You got it the slow way, because when you started there was no fast way.
I do not think the answer is making people do things the slow way on purpose. That is the grumpy old lady talking, wanting her old ways back. The old way of building this skill only worked because there was no other way to get the job done, and that is over.
So we have to build it some other way, and I do not think anyone has worked out how yet. Maybe it comes from arguing with the AI instead of just using it. Maybe it comes from being handed work you did not do and being made to take it apart, as early and as often as we were once made to produce it. Maybe the next generation builds something I cannot picture, the way we do things the generation before us could not have imagined.
This is what I think is more valuable for us to work on. AI will take care of more and more of these big solutions. What falls on us is to solve for the path that goes missing.