DeepMind's AlphaProof Impresses at Math Contest

In a groundbreaking achievement, an AI developed by Google DeepMind has secured a silver medal score at this year’s International Mathematical Olympiad (IMO), marking the first instance of an AI reaching the podium. However, this accomplishment was not from a live competition, and the AI struggled with various mathematical disciplines essential for winning a medal, such as number theory, algebra, and combinatorics. To enhance its capabilities, Google DeepMind introduced AlphaProof, an AI designed to tackle a broader spectrum of mathematical problems, alongside an upgraded version of AlphaGeometry for geometry questions. When tested on this year’s IMO problems, the combined systems of AlphaProof and AlphaGeometry successfully answered four out of six questions, achieving a score of 28 out of 42 points—just one point shy of the gold medal threshold. Gregor Dolinar, president of the IMO, expressed admiration for the AI's rapid improvement, noting that its near-gold performance is remarkable. Timothy Gowers from the University of Cambridge, who evaluated AlphaProof’s responses, remarked on the AI's ability to discover solutions akin to human reasoning. The AI employs a trial-and-error method known as reinforcement learning, with researchers utilizing Google’s Gemini AI to convert mathematical problems into a programming language called Lean, facilitating the AI's learning process. While the results are promising, questions remain about AlphaProof's reasoning processes and its potential utility for mathematicians, as it cannot assist in identifying research problems. Additionally, a $5 million prize has been announced for an AI that can achieve a gold medal at the IMO, but AlphaProof is ineligible due to its non-public status.