Alphago (Chinese name for Alphago (Alfa Go or Alfa Dog) is an artificial intelligence Go program developed by Google DeepMind in London, England, which became the first computer Go program to defeat professional Go players on a full-size 19×19 board without the aid of a concession, and is a major challenge for artificial intelligence and human intelligence.

In March 2016 in a five-game Go In March 2016, AlphaGo defeated professional Go player Lee Se-dol in the first three games of a five-game Go tournament, becoming the first computer Go program to defeat a professional ninth-degree Go player without the aid of a concession.

In terms of terminology, AlphaGo uses a combination of Monte Carlo tree search and two deep neural networks, one of which is a valuation network to evaluate a large number of picks, and a move network to select discs. With this design, the computer can combine the long-term inference of tree diagrams with the intuitive training of spontaneous learning like the human brain to improve chess playing strength.

AlphaGo has shown significant improvements over previous Go programs. In 500 games against other Go programs like Crazy Stone and Zen, AlphaGo (running on one computer) lost only one game. The October 2015 distributed computing version of AlphaGo uses 1,202 CPUs and 176 GPUs.

&p> ;lt;p> However, Google has not publicly explained what hardware or software improvements were made to enhance its power from October 2015 to March 2016, so it may further utilize more hardware in the March competition.

AlphaGo uses Monte Carlo tree search, with two deep neural networks, the valuation network and the move network, to evaluate a large number of picks through the valuation network and select landing points through the move network. AlphaGo's database contains about 30 million moves. Once it reaches a certain level of proficiency, it begins to play a large number of games against itself, using reinforcement learning to further improve it. Go cannot be solved just by finding the best move; a game has an average of 150 moves and an average of 200 options for each move, meaning that there are too many possibilities to solve.