A mathematics artificial intelligence agent developed with participation from Korean mathematicians has provided solutions to 13 world-class mathematical problems.
According to the mathematics community, an international research team including Professor Kim Sang-hyun of the Korea Institute for Advanced Study and Professor Chung Jun-hyuk of Brown University used Google DeepMind's mathematics AI agent "Aletheia" last month to present solutions to 13 major mathematical problems. The research results were first published on the preprint server arXiv on May 29.
Aletheia is a mathematics AI agent based on Google's large language model. The research team used Aletheia to review 700 unsolved problems from the Erdős problems in December last year. The Erdős problems are 1,179 unsolved mathematical problems left by Hungarian mathematician Paul Erdős, considered one of the most influential figures of the 20th century. His problem list encompasses number theory, combinatorics, and graph theory, among other fields, with approximately 400 problems reported to have been solved. For this reason, solving Erdős problems is also used as a benchmark to gauge AI's problem-solving capabilities in mathematics.
In this study, Aletheia found solutions to 200 problems. After excluding cases with errors in problem descriptions or issues with the answers themselves, the researchers concluded that mathematically meaningful solutions were presented for 13 problems. The solution to Problem 1051 in particular is considered significant enough to warrant a separate academic paper based on the approach Aletheia presented.
However, the researchers cautioned against over-interpretation, stating, "This research shows AI can make substantial contributions to solving mathematical problems," while noting that "in some cases, the AI misunderstood the questions."