Mathematicians issue a major challenge to AI (www.scientificamerican.com)

Mathematicians have launched a new initiative, “First Proof,” challenging AI systems to solve a set of unsolved mathematical problems relevant to their current research. The exam was designed by a group of 11 renowned mathematicians, including a Fields Medal winner, and consists of encrypted lemmas—smaller theorems that contribute to more significant results. The AI systems will have one week to tackle these problems, with the solutions being unveiled on February 13, 2024. This experiment is notable because the problems are unique and not available in any large language model's training data, offering a more rigorous test of AI's mathematical capabilities than previous assessments. The significance of this initiative lies in its potential to better evaluate the role of AI in pure mathematics, an area that has seen increasing interest from the tech industry. Recent advancements have seen AI systems achieve impressive feats, like scoring high on the International Mathematical Olympiad and solving various Erdős problems. However, past evaluations often lacked control and clarity, leading to skepticism about the originality of AI-generated proofs. The success of AI in solving these lemmas could underscore its value as an auxiliary tool for mathematicians, streamlining routine tasks and enhancing the overall research process, which many experts believe is where AI's most immediate impact on mathematics will be realized.