What's the Point of Learning Math If AI Can Do It?

"Why do I have to do this if AI can do it?"

Every high school and college student I talk to uses AI. What started as help with math homework has turned into one-prompt solutions and the results are astounding.

To put things into perspective, consider the International Mathematical Olympiad (IMO), the hardest math competition for high school students. In 2026, a score of 29 out of 42 was enough for a gold medal, and only 7 of the 666 contestants earned a perfect score. That same year, two AI systems earned officially graded perfect scores of 42 out of 42, and several more matched them in independent tests. That's six new, unseen problems, among the hardest anywhere, all solved by a machine. AI can now write code, prove theorems, and predict the structures of proteins and in a few more months, who knows what other field it will absorb and autocomplete.

So should we tell our students they're cooked? That their homework, college assignments, and future problems are all just a one-shot prompt away?

I don't think so, and the reason comes down to what math actually is.

Math is more than the answer

Math isn't just about finding solutions. Much of it is about navigating uncertainty: trying an approach, hitting a dead end, backing up, and trying again. A different route, a different process, falling down and getting back up until meaningful progress is made. While the correct answer marks where a problem ends, it can't show you all the paths that led there or the ones that didn't.

Those dead ends are actually where the most interesting questions naturally emerge. Get stuck on a problem and you start asking: Why didn't this work? What if I change one thing? Am I missing a pattern here? Fermat's Last Theorem is a good example. It stumped mathematicians for over 350 years, and when one attempted proof hit a dead end in 1847, the effort to understand what went wrong helped spark the discovery of algebraic number theory. Even if AI can answer the question you give it, knowing which questions are worth asking comes from walking those paths yourself.

Intuition is developed, not downloaded

Ask yourself: how many ways can you tie a knot? If you can tie your shoes, that's one way. If you can tie a bow around a present, that's another. Hanging a hammock from a tree, another. Different knots for different problems. Now ask AI to tie your hammock to a tree. The knot may hold perfectly well. But when it's time to adjust the tension or pack up, will you know how? You might end up taking a pair of scissors to the rope, not because the knot was bad, but because you never understood how it was tied.

AI solutions are autocompleted knots. You have to tie many kinds of knots, many different ways, before you can feel how much tension one will hold or how to loosen it. That kind of judgment requires understanding and intuition, which can only be developed over time through practice.

The role of practice cannot be overstated. Real-world problems don't come with answer keys; nobody hands you a clean setup, checks your work afterward, and updates as the world naturally changes. Math is one of the rare places where problems have well defined answers, where the stakes are low, but the challenge can be adjusted to stimulate anyone.

Mathematician Terence Tao put it well in a podcast interview, describing math as "a good playground to think where the cost of failure is cheap." When a civil engineer makes a mistake, a bridge can collapse. When a surgeon makes one, a life can be lost. The specific tools, Tao says, matter less than the ability to face a problem full of unknowns: "How do you even start?... how to not get paralyzed when you're faced with anything complex." Practicing math is where students get to build that skill at the lowest possible price.

Math is a training ground for the mind to navigate uncertainty. The reward of learning comes not from the final solution, but from the winding journey toward it, even when the path takes you astray. Every problem you work through yourself is a knot you've learned to tie: you know how it holds, how to adjust it, and how to undo it when it's time to try something new. And maybe you discover a new knot along the way. AI can tie a knot for you, but learning to tie your own is the point.