overview
What is AlphaGo?
AlphaGo is an AI program developed by DeepMind that enables researchers and strategists to master the game of Go at a superhuman level. It combines deep neural networks with advanced search algorithms to achieve this. AlphaGo is an artificial intelligence (AI) system that utilizes deep neural networks with advanced search algorithms, specifically a Monte Carlo tree search, to master the ancient Chinese game of Go. Its primary function is to play Go at a superhuman level, demonstrating the capabilities of deep neural networks and reinforcement learning in highly complex strategic domains. The system employs two main neural networks: a "policy network" to select the next move and a "value network" to predict the game's winner from any given position. Initially, AlphaGo learned from thousands of expert human Go games through supervised learning, then refined its skills by playing millions of games against itself via reinforcement learning. This self-improvement process allowed AlphaGo to discover novel and creative strategies that often surprised human professionals. While AlphaGo itself was developed for Go, its underlying principles have inspired applications in healthcare, biology (notably AlphaFold 2), robotics, finance, climate science, mathematical reasoning (AlphaProof, AlphaGeometry 2, Gemini's Deep Think), and algorithm discovery (AlphaEvolve).