For nearly 80 years, one of mathematician Paul Erdős's geometric puzzles sat unsolved, a deceptively simple question that defeated generations of mathematicians. Now an AI system from OpenAI has done what human experts couldn't—and in the process, proved that Erdős himself was likely wrong about the answer.
The problem is called the planar unit distance problem, and here's what makes it elegant: imagine scattering dots on a piece of paper. What's the maximum number of pairs of dots that can be exactly the same distance apart?
It sounds almost trivial. But that's the genius of Erdős's problems—they're easy to state, brutally hard to solve, and often reveal deep mathematical structures hiding beneath simple surfaces.
Erdős conjectured in 1946 that as you add more dots, the maximum number of equal-distance pairs grows only slightly faster than the number of dots themselves. It was an elegant hypothesis: even as the system gets more complex, there's a kind of speed limit to how interconnected these equal distances can become.
For eight decades, mathematicians tried to prove Erdős right. They couldn't. And now we know why: he appears to have been incorrect.
The AI didn't solve this by brute force pattern-matching—the usual criticism when AI tackles mathematical problems. Instead, it did something genuinely surprising: it reached into different branches of mathematics that involve complex number systems and geometric approaches that hadn't been obviously connected to the problem before.
The result? Hundreds of pages of calculations showing that far more equal distances are possible than Erdős believed. No universal speed limit exists in the way he theorized.
Now, before you get too excited about AI replacing mathematicians: human experts still had to verify the proof. The AI generated the insight and the computational heavy lifting, but mathematicians reviewed the work, reformulated findings, and confirmed the reasoning was sound. It's collaboration, not replacement.
What makes this genuinely significant isn't just that AI solved a hard problem. It's that the solution required creative mathematical insight—choosing to apply tools from branches of math that weren't obviously relevant. That's not pattern recognition. That's closer to what we'd call intuition, or at least a computational analog of it.
There's also something poetic about the result. Erdős was one of the most prolific mathematicians in history, famous for posing problems that revealed hidden depths. That an AI proved him wrong on one of his own conjectures feels like a passing of some torch—not from human to machine, but from one era of mathematical exploration to another where both work together.
The limitation here? We still don't fully understand why the AI chose the approaches it did. The proof is valid, but the reasoning process remains somewhat opaque. It's like receiving a correct answer from a brilliant colleague who can't quite explain their thought process. Useful, but unsettling.
Still, this is one of the clearest examples yet of AI moving beyond pattern-matching into genuine mathematical discovery. And if an 80-year-old Erdős conjecture can fall, what other long-standing problems might be next?
