Mathematics serves as the basic language for understanding the universe, covering areas like physics laws and computer logic. Google DeepMind thinks artificial intelligence (AI) can help mathematicians create new ideas and find answers faster.
The AI for math initiative, backed by google deepmind and google.org, joins Imperial College London, Institute for Advanced Study, Institut des Hautes Études Scientifiques, Simons Institute for the Theory of Computing at UC Berkeley, and Tata Institute of Fundamental Research. The partners aim to spot math problems ready for AI help, build needed tools, and achieve new findings.
Google gives money from google.org and shares advanced tech like Gemini Deep Think, an improved thinking mode for AI; AlphaEvolve, an AI agent that finds new algorithms; and AlphaProof, a system for finishing formal proofs. This setup creates a cycle where basic math research improves AI, leading to stronger ties.
A pivotal moment for ai and mathematics
This initiative starts during quick growth in AI's reasoning skills. In 2024, AlphaGeometry and AlphaProof reached silver-medal level at the international mathematical olympiad, a worldwide math contest for young people. Later, the newest Gemini model with Deep Think hit gold-medal performance, solving five out of six problems perfectly.
AlphaEvolve worked on over 50 open math issues in areas like analysis, geometry, combinatorics, and number theory. It bettered old solutions in one-fifth of cases. It found a faster way for matrix multiplication, a key computer calculation, using fewer steps for 4x4 matrices than a 50-year-old method. In computer science, it uncovered new structures showing some problems are harder for computers, clarifying limits and guiding future work.
This fast AI progress shows its growing power. The initiative hopes to use AI for harder math, exploring its full potential. By mixing mathematicians' deep insights with AI's new skills, it could open fresh research paths, boost knowledge, and lead to breakthroughs in many sciences.