Mathematicians Puzzled Over a Famous Problem for 80 Years. Now, They’ve Used A.I. to Identify a Clever Solution

In 1946, the mathematician Paul Erdős posed the unit distance problem—and suggested a winning strategy. An A.I. model has now landed on a better one. Why didn’t humans get there first?

Ellen Wexler | Writer and Special Projects Editor

June 3, 2026 12:14 p.m.

Share

Copy linkEmailSMSFacebookXRedditLinkedInBlueskyPrint

Add as preferred source

Place any number of dots on a two-dimensional plane—say, a piece of paper—and measure the distance between each pair. If you rearrange the dots, how many pairs could be positioned exactly the same distance apart?

The Hungarian mathematician Paul Erdős first posed this question in 1946. Even as the number of dots grew, he argued that the best way to maximize the total pairs was to draw a grid-like arrangement. Erdős also speculated that the number of pairs could only be slightly higher than the number of dots. For the next 80 years, most mathematicians assumed he was correct.

But now, an artificial intelligence model has challenged this approach. OpenAI recently announced that one of its internal models had identified a strategy to produce more pairs than the arrangements proposed by Erdős.

“This is the unique, interesting result produced autonomously by A.I. so far,” Daniel Litt, a mathematician at the University of Toronto, tells Scientific American’s Joseph Howlett.

This particular puzzle, known as the planar unit distance problem, was “one of Erdős’ favorite problems,” Noga Alon, a mathematician at Princeton University, writes in a companion paper to OpenAI’s announcement. Alon had heard him mention the problem “multiple times” in his lectures. In 1982, after decades with few breakthroughs, Erdős offered $300 for a proof or disproof of his proposed upper bound. He increased that amount to $500 around 1995.

Born in Budapest in 1913, Erdős was the son of two math teachers. Both of his older sisters died of scarlet fever while his mother was in the hospital for his birth. Soon after World War I broke out in 1914, his father was detained in a Siberian prisoner-of-war camp. He wouldn’t return for six years. In the midst of this anguish, Erdős’ mother became very protective of her surviving child, keeping him at home most of the time. “I fell in love with numbers,” he later recalled. “They were my friends. I could depend on them to always be there and always behave in the same way.”

After graduating from a Hungarian university with a PhD in mathematics, Erdős lived an itinerant lifestyle, staying with mathematicians around the world who he wanted to collaborate with. He would turn up, often without notice, uttering his signature greeting: “My brain is open!”

“He lived and breathed mathematics and could fall asleep at a dinner table if mathematics wasn’t the topic of conversation,” the Los Angeles Times’ Hector Tobar wrote in Erdős’ 1996 obituary. He co-authored more than 1,500 papers, making him one of the 20th century’s most prolific mathematicians.

In addition to his published papers, Erdős posed hundreds of problems for fellow mathematicians to solve, often offering monetary rewards. He priced some as low as $10 or $25. Others came with bounties in the thousands. (His largest was $10,000.)

In practice, this reward system was rather informal. “I solved a $250 problem, but I only got $50 because Erdős didn’t like the proof,” the late András Hajnal, then a mathematician at Rutgers University, told Science magazine’s Charles Seife in 2002. Because Erdős had posed so many problems, he didn’t always know when someone had solved one.

“Some years ago, he was visiting Athens, Georgia,” mathematician Carl Pomerance recalled in 2002. “He was going back into town, and my colleague Helmut Maier was giving him a ride.” When Maier mentioned a theorem he’d recently proven, Erdős wondered whether he’d offered a prize for it. The two men rerouted to the library, where they discovered that Erdős had, indeed, promised $100. “I said to Erdős, ‘That’s a pretty expensive taxi ride,’ and he found that hilarious,” Pomerance told Science.

When OpenAI announced its recent breakthrough, “everyone in math lost their minds,” writes the Wall Street Journal’s Ben Cohen. The idea of such a discovery would have been laughable even a year ago—or, as OpenAI’s Sébastien Bubeck tells the publication, even “a month ago.”

OpenAI’s models had attempted to tackle Erdős problems before. Last year, Kevin Weil, then a vice president at the company, announced that “GPT-5 just found solutions to ten (!) previously unsolved Erdős problems.” In reality, the model had identified existing solutions that were buried in the literature. The Erdős Problems website listed these problems as “open,” but “the ‘open’ status only means I personally am unaware of a paper which solves it,” Thomas Bloom, a mathematician at the University of Manchester who maintains the site, wrote on social media. He called OpenAI’s claim a “dramatic misrepresentation.”

This time, OpenAI appears to have been more careful. In the companion paper, a group of mathematicians—including Bloom—reviewed the breakthrough and reflected on A.I.’s role in the process.

OpenAI’s strategy involves arranging the dots using techniques from a branch of mathematics called algebraic number theory. Soon after the announcement, Will Sawin, a mathematician at Princeton University who also contributed to the companion paper, drew on the same approach to find an even better solution. Bloom, Sawin and their colleagues also used similar techniques to crack another open problem, called the sum-product conjecture, which Erdős had posed in the 1970s.

“It was a surprise because I had thought about the problem quite a bit,” Bloom tells New Scientist’s Alex Wilkins. “Once you know that something might be possible, you’re willing to try a bit harder to actually get it to work.”

OpenAI, however, didn’t release the model’s original output. Instead, it published a “rewritten summary” of the model’s chain of thought, along with a proof rewritten by mathematicians. As Bloom writes in the companion paper, “The human still plays a vital role in discussing, digesting and improving this proof, and exploring its consequences.”

On June 2, a group of mathematicians published a series of recommendations concerning that still-vital human role. Dubbed the Leiden Declaration on Artificial Intelligence and Mathematics, the document lays out guidelines for researchers, policymakers and organizations on the ethical use of A.I. in mathematics.

The declaration’s authors warn about flashy headlines publicizing results that haven’t yet been vetted by the standard academic process. They urge researchers to disclose which A.I. tools they’ve used, take responsibility for their work’s accuracy and ensure they’ve properly cited existing scholarship that A.I. may have drawn from.

In announcements like OpenAI’s, “basic information needed to assess the scientific meaning of the result is kept secret,” co-author Rodrigo Ochigame, an anthropologist and historian of computing and artificial intelligence at Leiden University in the Netherlands, tells the New York Times’ Siobhan Roberts. “The company disclosed nothing about the methods, human-written prompts, training data or computational resources consumed.”

The authors are also concerned about A.I. companies’ motivations. OpenAI’s researchers direct resources toward Erdős problems because they’re trying to improve models’ reasoning abilities more broadly. Some of these models, the authors write in the declaration, are trained on mathematicians’ papers and then “commercialized for applications that raise grave ethical concerns,” such as warfare and mass surveillance.

Did humans need A.I. to crack the unit distance problem? In the companion paper, the authors explain that the model pulled from knowledge across domains, whereas most human mathematicians are focused on narrow specialties. Researchers had also assumed that Erdős’ strategy was correct, so few had devoted significant time hunting for ways to disprove it. While humans’ time is limited, A.I. can tirelessly hammer away at any given problem.

“Certainly, this is an idea that, as far as we can tell, humans did not come up with. This is not an idea that humans couldn’t have come up with,” Sawin tells Gizmodo’s Gayoung Lee. “It’s not that A.I. solved an impossible math problem, but it’s not nothing. It’s somewhere in between.”

Ellen Wexler | | Read More

Ellen Wexler is a staff writer and special projects editor, digital, for Smithsonian magazine.