AI, Erdős, and the Great Dot Mischief
Acronyms and terms:
AI: Artificial Intelligence, computer systems designed to perform tasks that normally require human-like reasoning, pattern recognition, or decision-making.
OpenAI: An artificial intelligence research and product company that announced the recent mathematical result discussed here.
Erdős problem: A mathematical problem posed or popularized by Paul Erdős, the famously prolific Hungarian mathematician who left behind a small mountain range of difficult questions.
Unit distance problem: A geometry problem asking how many pairs of points can be exactly one fixed distance apart.
Discrete geometry: The study of geometric objects arranged in separate countable pieces, such as points, lines, grids, and patterns.
Combinatorial geometry: The part of mathematics where counting and geometry shake hands, argue, and occasionally break furniture.
Algebraic number theory: A deep area of mathematics that studies numbers using abstract structures; in this story, it unexpectedly helped solve a problem about dots on a plane.
AI did not wake up one morning, drink tea from a chipped bhar, and solve the meaning of existence. Let us not give it a silk shawl yet. What it did was stranger, smaller, and in some ways more interesting: it found a new way through an old Paul Erdős problem about dots on a flat surface.
Dots.
Not cancer.
Not consciousness.
Not why the ceiling fan makes a noise only at 2:17 in the morning when a single unemployed middle-aged man in the Calcutta boondocks is trying to sleep and failing with professional consistency.
Dots.
That is the first lovely absurdity.
Take a piece of paper. Put some dots on it. Now count how many pairs of dots are exactly one unit apart. One inch, one centimeter, one whatever-you-like, as long as the distance is fixed. If two dots are exactly that distance apart, they are a successful pair. Like two passengers in a bus who have somehow found arm space without causing a constitutional crisis.
Now the question becomes: if you have many dots, what is the maximum number of such exact-distance pairs you can arrange?
This is the kind of question that looks harmless enough to be printed on a school worksheet. Then it quietly eats eighty years of mathematical life.
Paul Erdős asked it in 1946. Erdős was not a normal mathematician in the office-and-cabinet sense. He was a wandering Hungarian problem machine, a man who traveled from university to university with very little luggage and an alarming ability to make other people’s brains catch fire. He would appear, ask a devastatingly simple question, and leave behind a room full of professors staring at the blackboard like men who had just been cheated by a very small child.
The unit distance problem was one of those questions.
For decades, the best intuition was this: the most efficient dot arrangement probably looks something like a square grid. Think graph paper. Think window grill. Think those ruled school notebooks where geometry lived before coaching centers turned childhood into a logistics industry.
A square grid gives you many equal distances. Each dot has neighbors. Left, right, up, down. It is tidy. It is believable. It has the respectable smell of mathematics after a bath.
Erdős suspected that the number of exact unit-distance pairs could not grow too much faster than the number of dots. In loose language, if the number of dots was , the number of special pairs was expected to be only a little more than in the long run. Not wildly more. Not a festival crowd. More like a disciplined queue, which in our part of the world is already a fantasy novel.
Then AI walked in and made the queue misbehave.
OpenAI announced in May 2026 that one of its reasoning models had found an infinite family of point arrangements beating the long-standing Erdős expectation. Not just one clever drawing. Not a lucky diagram on the back of a tea packet. A whole repeatable method.
And the method did not come from drawing better graph paper.
It came from algebraic number theory.
This is where normal people reasonably begin looking for the exit.
Do not.
Here is the idea without the ceremonial fog.
You think the problem is about arranging dots on a flat page. That is the visible game. But sometimes mathematics is like an old North Calcutta house. The front room is simple: a chair, a table, a calendar from a sweet shop. Then somebody opens a side door and suddenly there is a courtyard, a staircase, three locked rooms, a retired uncle, and a cat with property rights.
The visible room was dots.
The hidden house was number theory.
The AI did not merely try more dot patterns faster. That would be impressive but not shocking. Computers are good at trying many things, just as mosquitoes are good at finding the one part of your ankle not covered by cloth.
The shocking bit is that the model found a route through a different mathematical neighborhood. It used deep number-theoretic structures to manufacture arrangements of points that create more exact unit distances than the old grid-based intuition suggested.
This is not like using a bigger hammer.
This is like trying to fix a leaky roof and discovering the real solution involves the railway timetable, the moon’s gravity, and your neighbor’s goat. You may object on aesthetic grounds, but if the leak stops, you must at least inspect the goat.
Now comes the correction, because the internet has already learned to turn every achievement into either a coronation or a funeral.
AI did not solve all of mathematics.
AI did not make human mathematicians obsolete.
AI did not become Ramanujan with a graphics card.
What happened is precise and important: an AI model found a counterexample to a major Erdős conjecture about the unit distance problem. A counterexample means the old claim cannot be true as stated. One black swan is enough to ruin the sentence “all swans are white.” One better-than-expected dot construction is enough to ruin the belief that the grid-like lower bound was essentially the end of the road.
That is real.
But the complete shape of the whole unit distance problem is still not magically finished. There are still upper bounds, lower bounds, gaps, refinements, human explanations, and future fights. Mathematics does not end with one trumpet blast. It usually ends with six papers, two corrections, a seminar argument, and one elderly professor saying, “Yes, but what happens in the general case?”
This is why the human verification matters.
The AI produced the result. Then mathematicians examined it, digested it, cleaned it, and explained it. That last verb is not decoration. Explanation is where mathematics becomes public property rather than private lightning.
A machine can find a tunnel.
Humans still have to check whether the tunnel goes somewhere useful or merely deposits you under a tea stall in Behala with wet socks and philosophical injuries.
This is also why mathematics is a better testing ground for AI reasoning than many other areas. In politics, business, or family WhatsApp, almost any claim can survive if shouted with enough confidence. In mathematics, a proof must hold plank by plank. If one plank breaks, the whole bridge drops you into the river.
That is comforting.
Not completely comforting, because nothing is completely comforting at 51 when the future sits in the corner like an unpaid electricity bill. But somewhat comforting.
The important lesson is that AI’s power here was not just speed. It was shameless exploration.
Humans are clever, but we are also vain, tired, cautious, proud, hungry, anxious, status-aware mammals. A human mathematician may avoid a route because it looks ugly, unfashionable, too technical, or too likely to waste six months. AI does not care if the idea looks ugly. AI has no dinner invitation to lose. AI does not worry that someone at Coffee House will smirk and say, “Algebraic number theory for a dot problem? Very good, professor, next you will use fish curry to repair the tramlines.”
So it wandered where a person might hesitate.
That is not intelligence in the full human sense. It is something narrower but still potent. A tireless searcher with some reasoning ability and no social embarrassment is not a person. It is also not nothing. It is a new kind of mathematical animal, and we should neither worship it nor pretend it is a calculator with lipstick.
The Calcutta reader may ask: why should I care?
Fair question. You have prices rising, roads dug up, young people leaving, old people trapped in flats, and every second institution behaving like a badly folded umbrella. Why worry about dots on a plane?
Because this little dot story tells us something about the future of knowledge.
For a long time, many people assumed AI would be good at shallow tasks first: writing emails, summarizing documents, generating pictures of improbable men being attacked by even more improbable reptiles. Useful, sometimes ridiculous, often mildly cursed.
But mathematics is different. Real mathematics is not just fluent speech. It is disciplined truth. You cannot bluff your way through a theorem the way a politician bluffs through a press conference. A proof is either valid or it is not.
So when AI contributes to a serious open mathematical problem, even in one narrow case, the air changes.
Not because machines have become gods.
Because tools have become collaborators.
The telescope did not replace astronomers. It humiliated naked eyesight. The microscope did not replace biologists. It opened a tiny kingdom under the glass. AI may become something like that for abstract thought: an instrument that lets humans see parts of the idea-world we could not easily search alone.
But instruments also lie if misused.
A telescope can be badly aimed. A microscope can show contamination. An AI can hallucinate, overclaim, repackage old work, or produce something that sounds like thought but is really a brass band falling down stairs. That has happened before. It will happen again.
So the rule is simple: wonder, then verification.
Not wonder instead of verification.
This is where I become boring for one sentence and useful for the rest. In the modern world, the highest skill may not be knowing everything. Nobody knows everything. The highest skill may be knowing when to be impressed, when to be suspicious, and when to ask, “Who checked this?”
That question is unfashionable. It is also civilization.
In this case, mathematicians checked. They did not merely clap because the machine wore a shiny badge. They read the proof, translated it into human mathematical language, and connected it to known ideas. That is why the result deserves attention.
There is a second, quieter beauty here.
The problem began as something a child could understand: put dots on paper and count equal distances. The solution route went through mathematics so deep that most adults would rather discuss income tax forms. This happens often in science. The question is simple. The answer has machinery.
Why is the sky blue?
Simple question.
Suddenly you are inside light scattering.
Why does a fever rise?
Simple question.
Suddenly the immune system enters with files, clerks, police, miscommunications, and a chemical civil war.
How many dot-pairs can be one unit apart?
Simple question.
Suddenly number theory arrives wearing boots.
That is one of the great joys of knowledge. The small door opens into a large room.
And perhaps that is the reason this story stayed with me. I am sitting in Calcutta, not in Princeton, not in Cambridge, not in some glass building where the coffee machine has more funding than most Indian research labs. Outside, the day has the gray patience of a wet newspaper. Inside, the laptop is warm, the tea is ordinary, the body is not particularly cooperative, and the mind is trying to make rent out of fragments.
Then comes this news: somewhere, a machine found a new path through an eighty-year-old human puzzle about dots.
It is absurd.
It is beautiful.
It is also a little rude.
Mathematics, which we imagined as the last quiet room of human thought, now has a strange new visitor. Not a replacement. Not a messiah. Not a demon. More like a tireless, socially clueless assistant who may occasionally discover a staircase behind the bookshelf.
The proper response is not panic.
The proper response is not worship.
The proper response is to sharpen the pencil, check the proof, and keep asking better questions.
Because Erdős is gone, but his problems are still wandering the earth.
And now, apparently, some of them have met AI.
P.S. References: OpenAI, “An OpenAI model has disproved a central conjecture in discrete geometry”; arXiv, “Remarks on the disproof of the unit distance conjecture”; arXiv, “An explicit lower bound for the unit distance problem”; The Guardian report on OpenAI and the Erdős unit distance problem; Live Science report on mathematicians verifying the AI result.