Google DeepMind AI Solves Olympiad Math at Silver-Medal Level
Two Google DeepMind AI programs successfully solve four out of six problems at the 2024 International Mathematical Olympiad, achieving a score equivalent to a human silver medalist.
Google DeepMind introduces two artificial intelligence programs that achieve silver-medal-level performance at the 2024 International Mathematical Olympiad. The AI systems successfully solve four out of six complex problems, earning a total of 28 out of 42 possible points. This impressive score outperforms the vast majority of the more than 600 human students participating in the competition.
The technology relies on two distinct systems named AlphaProof and AlphaGeometry 2. AlphaProof utilizes reinforcement learning, a technique previously used to master board games like chess and Go, but adapts it for abstract mathematical reasoning. By repeatedly testing strategies and learning from failures, the AI gradually discovers correct solutions to incredibly difficult problems.
To accurately verify its own logical reasoning, AlphaProof integrates specialized algorithms known as proof assistants. These tools meticulously check each step of a mathematical argument to ensure total accuracy before accepting a solution. While proof assistants exist for decades, their integration into modern machine learning provides a major breakthrough for AI's ability to comprehend and solve formal mathematics.