Google DeepMind’s AI systems have been trained to solve complex math problems that require logical reasoning. Two specialized AI systems, AlphaProof and AlphaGeometry 2, worked together to successfully solve four out of six problems from this year’s International Mathematical Olympiad (IMO). This is the first time any AI system has achieved such a high success rate on these types of problems.
AlphaProof is a reinforcement-learning-based system that trains itself to prove mathematical statements in the formal programming language Lean. It can automatically translate math problems phrased in natural, informal language into formal statements, which are easier for the AI to process.
The Gemini model works alongside AlphaZero to prove or disprove millions of mathematical problems. The more problems it has successfully solved, the better AlphaProof has become at tackling problems of increasing complexity.
AlphaGeometry 2 was optimized to tackle geometry questions and successfully solved one such question. It also attempted two combinatorics questions but left them unsolved.
Two renowned mathematicians checked the systems’ submissions, awarding each of their four correct answers full marks. This is the first time any AI system has been able to achieve a medal-level performance on IMO questions.
The achievement could pave the way for exciting human-AI collaborations, helping mathematicians to both solve and invent new kinds of problems.
Source: https://www.technologyreview.com/2024/07/25/1095315/google-deepminds-ai-systems-can-now-solve-complex-math-problems/