AI SCIENCE
From napkin sketch to research breakthrough
For nearly 80 years, mathematicians have been stuck on a question that sounds annoyingly simple: if you place dots on a flat surface, how many pairs can be the same distance apart?
This is called the unit distance problem.
In 1946, mathematician Paul Erdős proposed a possible answer, but no one had been able to prove or disprove it.
Now, researchers say that conjecture has been disproved.
A reasoning model was given the problem and asked to either prove it or find a counterexample.
Instead of confirming what many mathematicians expected, it found a way to beat Erdős’ proposed arrangement.
The proof used algebra and number theory, two areas that do not seem like obvious tools for a geometry problem about dots on a plane.
Harvard mathematician Melanie Matchett Wood said that is part of what makes the result interesting, because it shows how ideas from one area of maths can unlock progress in another.
The counterexample is not exactly doodle-friendly. It involves creating a complex grid in a higher-dimensional space, then projecting it onto a flat plane. Very casual behaviour from a dot problem.
The result is being treated as a real mathematical breakthrough, but researchers are being careful about what it says about AI.
The discovery matters because:
It challenges a long-standing maths conjecture that had remained unresolved since 1946.
It shows how algebra and number theory can be used in unexpected ways to solve geometry problems.
It raises fresh questions about how AI-generated proofs should be checked, credited, and shared.
Dots behaving badly
Some experts said the proof relied more on persistence than a sudden flash of genius.
It has also raised bigger questions about how AI-generated maths should be checked.
In this case, experts said the proof was relatively easy for humans to verify.
But future AI-generated proofs could be much harder to assess, especially if they run to hundreds of pages.
That concern has already led a group of experts to call for stronger guardrails around AI in mathematical research.
Their concerns include reliability, transparency, proper credit for ideas, and access to powerful private tools.
For now, the takeaway is simple: AI may become a useful research tool, especially for testing routes humans might not have the patience to follow.
But mathematicians still want the receipts.
I wonder what they’ll say when I tell these geniuses I use the calculator when splitting the bill at the restaurant.- MG