Google DeepMind AI Achieves Silver Medal Level in International Math Competition
Google DeepMind introduces AlphaProof and AlphaGeometry 2, two AI systems that successfully solve complex mathematical problems requiring advanced reasoning. The breakthrough allows the AI to earn the equivalent of a silver medal at the prestigious International Mathematical Olympiad.
Google DeepMind introduces two specialized AI systems, AlphaProof and AlphaGeometry 2, that successfully tackle complex mathematical problems. By working together, these systems solve four out of six problems from this year's International Mathematical Olympiad, achieving the equivalent of a silver medal. This milestone marks the first time any AI system reaches such a high success rate on these highly challenging problems.
While AI models easily generate text, they traditionally struggle with the logical reasoning required for advanced mathematics. Math problems demand difficult cognitive tasks like forming abstractions, hierarchical planning, setting subgoals, and backtracking. Additionally, there is significantly less formal mathematical data available online compared to informal natural language, making it hard to train systems to verify their own mathematical proofs.
To bridge this gap, DeepMind builds AlphaProof as a reinforcement-learning system that teaches itself to prove mathematical statements using the formal programming language Lean. The system utilizes a fine-tuned version of the Gemini AI to automatically translate informal, natural language math problems into formal, verifiable logic. This approach represents a major step forward in the ongoing quest to build AI systems capable of genuine, complex reasoning.