AlphaGo

AlphaGo is a computer program developed by Google DeepMind to play Go. It was the first Computer Go program to beat a professional human Go player without handicaps on a full 19x19 board. He also wion a 5-game match with a top professional.

The algorithm of AlphaGo is based on a first machine learning extensive trainning using previous data and simulated games, using mainly deep learning methods. All this trainning build policies (policy network) and incentives (value network) that are used in the moment of playing and taking the decision of each movement by a Monte Carlo tree search.

Go is a highly complex game due to the huge amount of possibilities that its possible moves offers. It is usually considered an intuition game.

See also

Artificial Intelligence

Material

• http://spectrum.ieee.org/automaton/robotics/artificial-intelligence/why-alphago-is-not-ai
• https://www.quantamagazine.org/20160329-why-alphago-is-really-such-a-big-deal/
• http://elpais.com/elpais/2016/03/09/ciencia/1457514101_562108.html
• http://www.bbc.com/news/technology-35785875
• https://github.com/Rochester-NRT/RocAlphaGo

Papers

• Silver, D., Huang, A., Maddison, C. J., Guez, A., Sifre, L., Van Den Driessche, G. & Dieleman, S. (2016). [Mastering the game of Go with deep neural networks and tree search].(https://research.googleblog.com/2016/01/alphago-mastering-ancient-game-of-go.html) Nature, 529(7587), 484-489.