An international research team that included Korean mathematicians in the development process used Google DeepMind's math artificial intelligence (AI) agent Aletheia to present answers to 13 world-class math problems.
According to the mathematics community on the 8th, an international joint team that includes Kim Sang-hyun, a professor at the Korea Institute for Advanced Study, Jeong Jun-hyeok, a professor at Brown University in the United States, Tang Luong, a senior researcher at Google DeepMind, and Tony Feng, a professor at the University of California, Berkeley, used Google DeepMind's math AI agent Aletheia last month to present answers to 13 world-class math problems. The research results were first released on the preprint site arXiv on the 29th of last month.
Aletheia, which means "truth" in Ancient Greek, is a math AI agent based on Google's large language model (LLM). It is a math research agent built on Deepsync, the advanced reasoning mode of Google's Generative AI Gemini.
Using Aletheia, the researchers reviewed about 700 unsolved problems among the Erdős problems for about a week in Dec. last year. The Erdős problems are a collection in combinatorics and number theory posed by Erdős Pál, a Hungarian mathematician regarded as one of the most influential figures of the 20th century. Of the 1,179 in total, about 700 remain unsolved. Solving the Erdős problems is used as a benchmark to gauge AI's problem-solving ability in mathematics.
According to the team, Aletheia produced meaningful solutions to 13 problems. In particular, for Problem No. 1051, the solution Aletheia proposed is considered significant enough to form the basis of a separate paper.
Kim Sang-hyun said, "AI and mathematicians exchanged ideas and produced researcher-level results," adding, "Building on the order established by the mathematical community over thousands of years, using AI will be beneficial for humanity."