I remember the first time I saw an AI-generated proof. It was technically flawless. Every step logically sound. And yet, reading it felt like wading through a swamp of irrelevant details. The beautiful, elegant insight that should have been the centerpiece was buried under a mountain of trivialities. I felt a mix of frustration and fear. Frustration because I had to dig for the gold. Fear because maybe this is what the future of mathematics looks like: efficient, correct, and utterly soulless.
You’ve probably experienced something similar. You ask an AI to solve a problem, and it vomits back a wall of text. It’s correct, but it’s wrong in the way that matters. It misses the point. It doesn’t know what’s interesting. That’s the crisis nobody is talking about.
The real problem with AI math isn’t that it’s wrong—it’s that it’s boring. And boring math is dangerous math.
Terence Tao, one of the greatest living mathematicians, recently put it perfectly: “The writing very often dwells at length on trivialities while passing briefly through—or even actively obscuring—the most interesting and novel portions of the argument.” He was talking about AI proofs, but he could have been describing the fundamental flaw in how we’re using AI. We’re so obsessed with getting the right answer that we’ve forgotten the question matters more.
Here’s the uncomfortable truth: mathematics is not just a logical system. It’s a storytelling art. The best mathematicians don’t just compute—they choose which problems to solve, they frame the narrative, they know which details to skip and which to spotlight. They have taste. AI has no taste. It can brute-force its way through a proof, but it can’t tell you why the proof is beautiful or why it matters.
AI can do math. It cannot do mathematics. The difference is everything.
This creates a new bottleneck. The bottleneck isn’t computation—it’s problem selection. As AI gets cheaper and more powerful, the limiting factor becomes the human ability to identify the right questions. We’re moving from a world where we struggled to find answers to a world where we’re drowning in answers, and we don’t know which ones to care about.
Think about the Hitchhiker’s Guide to the Galaxy. The answer to the ultimate question is 42, but nobody knows what the question was. That’s our future. AI will hand us 42s by the truckload, and we’ll be left scratching our heads, wondering which ones are actually worth remembering.
And here’s the twist: the more we rely on AI, the worse we get at asking good questions. Because asking good questions requires judgment, intuition, and a deep sense of what’s interesting. And those are skills that atrophy when you outsource them to a machine. We’re in danger of creating a generation of mathematicians who are excellent at using AI but terrible at doing math.
The best mathematicians don’t have the best tools. They have the best taste. And taste cannot be automated.
So what do we do? We don’t stop using AI. That’s a losing battle. Instead, we need to double down on the human elements: the storytelling, the problem selection, the narrative framing. We need to teach mathematicians to be critics, not just calculators. We need to value the art of the proof as much as the proof itself.
I’ll leave you with this: The next time you see an AI-generated proof, don’t ask if it’s correct. Ask if it’s interesting. If you can’t tell, that’s the real problem.
FAQ
Q: Isn't AI just a tool? Why should we worry about it lacking taste?
A: A hammer doesn't need taste. But AI is not a hammer—it's a collaborator that shapes the way we think. When you rely on a tool that can't distinguish between a beautiful proof and a garbage one, you start to lose your own ability to make that distinction. That's the danger.
Q: What's the practical implication for a mathematician or researcher?
A: Stop using AI as a black box. Use it to generate drafts, but treat every output with skepticism. Ask yourself: Does this proof tell a story? Does it highlight the key insight? If not, rewrite it. Your job is no longer just to produce correct answers—it's to produce clarity.
Q: Isn't the contrarian take that AI will eventually develop taste?
A: Maybe. But that's a bet on the future. Right now, AI has no taste, and pretending it does is dangerous. Even if it develops taste, the human role will shift to something even higher-level: deciding which problems are worth solving. That's always been the most important skill, and it's the one we should be cultivating.