OpenAI Researchers Solve Decades-Old Geometry Problem Using AI
OpenAI researchers have solved the Erdős Unit Distance Problem, a combinatorial geometry challenge that has resisted mathematical proof for decades. The breakthrough came through an AI model that identified a solution path human mathematicians had not found, then converted the result into a formal proof.
The problem asks a deceptively simple question: what is the maximum number of points that can be placed in a plane such that every pair of points is exactly one unit apart? Despite its intuitive framing, the problem has defeated mathematicians for years, including Paul Erdős, who called it one of geometry's most important unsolved problems.
How the AI Found What Humans Missed
The OpenAI team-including Mark Sellke, Mehtaab Sawhney, and Lijie Chen-used an AI model to explore the mathematical space surrounding the problem. The model's initial output surprised the researchers enough that they initially doubted it.
Sebastian Bubeck, an OpenAI researcher, said the result "sounds like too good to be true." The team spent months converting the AI's output into a formal proof before accepting the solution.
Previous human attempts had relied on intricate geometric constructions and advanced number theory, but the approaches were fragile. Sellke said earlier solutions were "too delicate to execute." The AI found a more stable path by drawing on deep tools from algebraic number theory-connections that might not emerge from conventional mathematical reasoning.
The model could "explore all of these possibilities much more comprehensively" than a human mathematician working through a single line of reasoning. That breadth of exploration proved decisive.
What This Means for Research
The solution demonstrates that AI can contribute to fundamental scientific discovery, not just assist with computation. The capability to synthesize information across different mathematical domains and identify novel solution paths suggests a shift in how AI might function in research.
Lijie Chen said his timeline for AI's contribution to the problem "got shorter" than expected. The speed and effectiveness of the AI's approach outpaced researcher expectations.
The implications extend beyond geometry. If AI can crack long-standing theoretical problems by finding non-obvious connections, similar approaches might accelerate progress in physics, biology, engineering, and medicine-anywhere researchers face large solution spaces and interdisciplinary challenges.
For research professionals, this suggests AI's role is shifting from tool to collaborator. The Erdős solution shows AI can identify patterns and connections that complement human intuition, potentially opening research directions that would otherwise remain closed.
Learn more about AI for Science & Research or explore AI Research Courses to understand how these capabilities are being applied across scientific fields.
Your membership also unlocks:
