AI Solves Erdős' 80-Year-Old Dot Puzzle, Sparking Human Breakthroughs By Sawin And DeepMind

Look, math lovers are going absolutely wild right now. In May 2024, the folks at OpenAI dropped a paper that completely blew the dust off an 80-year-old mystery. An internal, general-purpose AI model found a real counterexample to a famous 1946 conjecture by the legendary mathematician Paul Erdős. This is about the planar unit distance problem, which asks a simple question: if you have a bunch of points on a flat sheet of paper, what is the absolute maximum number of pairs that can sit exactly one unit of distance apart?

With this fresh computer proof in hand, human mathematicians refused to sit on their hands. Just days after the OpenAI paper hit the web, Columbia University mathematician Will Sawin jumped right into the action.

By tracing the exact same line of logical steps laid out by the AI, Sawin pushed the math even further and snatched an even better result.

This shows a beautiful tag-team match between silicon and carbon—and the collaboration is spreading fast.

Over at Google DeepMind, researchers were busy having their own field day with these classic math riddles. Using their very own AI models, they quietly knocked down nine other open problems left behind by the prolific Hungarian genius in a single week—representing an unprecedented cleanup of historical academic challenges.

But under the hood, the most shocking part of this entire story is how such a tool did the heavy lifting. This was not some highly specialized, hyper-focused math calculator built only for geometry. Instead, OpenAI utilized that same general-purpose system—the kind of model normally used to write poetry or draft emails. It is like using a Swiss Army knife to perform open-heart surgery, and somehow succeeding.

How a Simple Dot Puzzle Defied Human Minds

To understand why this general-purpose triumph is so staggering, let us get down into the actual dirt of this problem because the setup is so simple a kid can grasp it. Imagine you have ten dots on a gigantic desk. You can slide them anywhere you want. If you place them in a straight line, you only get nine pairs that are exactly one inch apart. But if you arrange them in a clever grid or a sneaky circle, you can get way more pairs to hit that magic one-inch distance.

Paul Erdős wanted to find a master formula to predict the absolute limit of these pairs for any number of dots. It sounds easy, but the math gets incredibly messy, incredibly fast. This is why the smartest minds on earth got stuck in the mud for eighty years.

Why Top Mathematicians Are Completely Shook Right Now

Because this puzzle has defied human intelligence for nearly a century, the reaction in the halls of elite universities has been nothing short of a seismic jolt. Canadian mathematician Daniel Litt openly admitted this is the very first AI-generated math result that he actually finds interesting on its own merits. For a community that usually scoffs at AI mistakes, that is a massive shift in tone. They are realizing this is not a gimmick.

It is a genuine research partner that can see paths we completely missed.

The Hidden Math Engine Powering This Massive Breakthrough

To reveal those hidden pathways, the systems rely on a clever architectural design. Behind the closed doors of these tech labs, the secret sauce is how these systems hunt for needles in infinite haystacks. By avoiding random guessing, these models use a closed-loop system where the AI writes computer code to represent different point arrangements, and a hardcoded checker instantly scores the results.

This is similar to how Nature detailed Google DeepMind's FunSearch, which uses a pre-trained model paired with an automated evaluator.

By constantly rewarding the AI when it finds an arrangement with more unit distances, the system self-corrects and climbs its way to a record-breaking configuration.

It is a relentless, high-speed evolutionary game of math survival.

Where Do We Go From This Mind-Bending Spot

With pure mathematics yielding to this evolutionary approach, we are left to wonder: where does this wild train ride stop? In the coming years, we are going to see these AI tools move from simple point-and-line puzzles to hard-core physical design and secret codes. Think about designing brand new materials at the molecular level.

If we can use AI to arrange physical atoms in optimal ways just like those dots on paper, we can build hyper-efficient solar panels or unbreakable data protection.

I find this absolutely thrilling because it means we might soon find perfect packing structures for deep-space antennas.

According to research published in Science, using AI to navigate complex search spaces will completely redefine industrial chemistry by 2030. The era of human-only discovery is officially in the rearview mirror, and I, for one, am ready to step on the gas.