Mathematicians Challenge AI With 50 Hard Open Mathematics Problems
Epoch AI’s FrontierMath: Open Problems now features 50 significant unsolved mathematics problems designed for computer verification, testing whether AI can contribute genuinely new mathematical ideas.
Epoch AI has expanded its FrontierMath: Open Problems benchmark to 50 significant research questions that have resisted serious attempts by professional mathematicians. The problems were selected for their mathematical importance and designed so proposed solutions can be checked computationally.
The benchmark covers areas including number theory, topology and combinatorics. Its aim is not simply to measure whether AI can perform calculations, but whether machines can make meaningful contributions to unresolved mathematical research.
Among the challenges is the Sum of Three Cubes problem, including the still-open case of 114. Other problems include questions related to irrationality proofs and the long-standing Lonely Runner Conjecture.
Epoch AI says three problems had been solved by AI by July 31, when the benchmark expanded to 50. The organisation also distinguishes between solutions produced by AI, humans and human-AI collaboration when assessing progress.
The bigger question is what these results mean for mathematical discovery. Epoch AI says current AI breakthroughs have largely involved applying existing techniques, while developing genuinely new theory would represent a more significant step.
For now, the mathematicians’ blackboard remains firmly in place. AI has simply been invited to take a very difficult seat at the table.
