An 80-year-old mathematical conjecture, known as the plane unit distance problem, has been solved by OpenAI’s advanced artificial intelligence model. This breakthrough is igniting discussion around the immense mathematical capabilities of AI.

“This is a problem I never expected to see solved in my lifetime,” states Mischa Rudnev from the University of Bristol, UK. “It’s a groundbreaking achievement.”

Tim Gowers commented that the solution represents a “significant milestone in AI mathematics.” He noted in a paper that had it been submitted by a human, it would be accepted without hesitation, highlighting that such groundbreaking evidence of AI-generated solutions is rare.

The plane unit distance problem was deemed by the 20th-century mathematician Paul Erdős as his “most important contribution” to geometry. The challenge lies in determining the maximum number of lines of equal length that can connect numerous dots placed on an infinite paper.

Erdős believed that the optimal arrangement for maximum connections would be a grid layout, suggesting that the number of lines would surpass the number of points only marginally. Persistent efforts to prove his assertion or to discover alternative patterns yielded minimal advances, with the most recent improvements occurring over 40 years ago.

OpenAI’s model revealed that Erdős’s hypothesis was significantly underestimated, demonstrating that a more irregular arrangement of points can yield far more connections.

“Initially, I could hardly believe the results,” remarks Will Sawin from Princeton University. “It convinced me that this achievement is the most remarkable in the realm of AI mathematics so far.”

Details on how the OpenAI model diverges from publicly available AI technology and its training methods remain undisclosed. However, researchers have mentioned that the model is “general purpose” and was not specifically trained for mathematical applications.

AI employed strategies from algebraic number theory to establish extensive lattices in dimensions far exceeding two-dimensional configurations. By creating these more sophisticated shapes, AI translates them into two dimensions to generate representations of the higher dimensions.

“The counterexamples produced by AI are intricate, and while the foundational ideas exist in existing literature, the synthesis to achieve them required ingenuity,” explains Kevin Buzzard from Imperial College London.

While the magnitude of this result is undeniable, it underscores a gap in mathematicians’ consideration of Erdős’s original predictions. Samuel Mansfield from the University of Manchester, England, notes that the complexity of executing an experiment to challenge the conjecture would have eluded many geometry experts lacking advanced number theory knowledge. “This highlights a necessity for interdisciplinary expertise,” he adds. “In hindsight, AI’s adaptability in this area may not be surprising.”

Rudnev emphasized that the problem’s allure lies in its “pure intellectual challenge,” and although it may not directly influence other unresolved questions, it has inspired further research. Sawin, after examining the proof, utilized the techniques unearthed by AI to advance the number of connectable points.

“Similar to many other AI milestones, it didn’t take long for human researchers to internalize, comprehend, and extend upon the AI-derived arguments,” commented Buzzard. “This contrasts sharply with human discoveries that often require extensive validation periods.