For 80 years, mathematicians wrestled with a simple-sounding puzzle: how many pairs of points can lie exactly one unit apart on a plane? The question came from Paul Erdős, one of the most prolific minds in math history. He guessed the best way to arrange points was a grid, like dots on graph paper. That idea shaped decades of research.

Last month, an AI from OpenAI shattered that long-held belief. Without human help, it found a new way to place points that beats all grid-like patterns. This isn’t just a small improvement. It disproves Erdős’ famous conjecture and changes how we think about the problem.

The Puzzle and the Breakthrough

The unit distance problem asks: given n points on a flat surface, how many pairs can be exactly one unit apart? At first, a square grid seems like the obvious answer. Points spaced evenly create many pairs at that distance. Erdős thought no other layout could do much better.

But the AI took a different path. It used algebraic number theory, a branch of math usually far from geometry. Instead of working with simple grids, the AI created complex grids based on algebraic integers. These grids live in higher-dimensional spaces and then get projected back to two dimensions. This approach packs more unit distances into the same number of points.

This leap isn’t just a clever trick. The AI leveraged advanced concepts like the Golod-Shafarevich criterion, a tool from 1960s number theory. Human mathematicians hadn’t applied these ideas here before. The AI connected dots between distant fields and found a pattern humans missed.

Mathematicians React and What It Means

Top mathematicians quickly verified the AI’s work. Fields Medalist Tim Gowers praised it as a milestone. He said if a human had done this, the paper would be accepted immediately by top journals. Princeton’s Will Sawin, who helped refine the proof, called it a major advance and explained how the AI’s use of growing algebraic number fields was key.

This breakthrough is a first in many ways. It’s likely the first major open math problem solved mainly by an AI. The proof is informal but rigorous enough to be polished into a traditional format. Humans cleaned up and extended the AI’s output to make it publishable.

What’s striking is how the AI handled the problem. It combined ideas from multiple math areas, showing a broad understanding beyond any single specialist. The AI explored many possible approaches, something humans can’t do efficiently. It also found conceptual leaps that weren’t obvious before.

Still, human insight remains vital. AI currently lacks the deep creativity to pick the most interesting questions or see the biggest picture. But this event shows AI can now push past long-standing limits and open new research directions.

Looking Ahead: The Future of Math and AI

This result hints at a new era for mathematics. AI can sift through huge amounts of knowledge and test countless ideas without tiring. It can grind through tedious proof strategies and spot overlooked connections.

But what happens in ten years? If AI keeps improving at math, human mathematicians might shift to guiding and interpreting AI outputs. Entire fields may evolve as machines handle more routine or exploratory work.

The breakthrough also raises questions about trust and understanding. Some AI-generated proofs may grow complex and hard to verify by humans. We may need new tools to audit and explain AI math results.

Beyond math, this could ripple into physics, cryptography, and computer science. Math is the foundation of many sciences. If AI can invent new structures and proofs, it might accelerate discoveries in many areas.

For now, the AI’s solution to Erdős’s unit distance problem is a landmark. It shows machines can not only solve tough puzzles but can change how math itself is done. The future of discovery could be a partnership between human intuition and AI’s relentless logic.