Google’s AlphaGo beats professional player in go board game

Spread the love

Google has developed a computer program that is able to beat a professional player in the board game go. The AlphaGo program defeated the European champion of the game, Fan Hui, five out of five times.

Go is considered one of the most challenging classic artificial intelligence games to learn, due to its large search space and the difficulty of determining board positions and moves. The software was developed by DeepMind, a company that Google bought in 2014. How AlphaGo was able to win in the complex game is explained in a study in Nature.

AlphaGo has two networks, one predicts the opponent’s next move, while the other tries to predict the outcome of different moves on the board. These networks are combined with an artificial intelligence algorithm to look ahead in the game in search of possible moves.

AlphaGo uses value networks to record board positions and policy networks to choose its next moves. The program features neural networks trained by learning games played by experts and learning the games the program plays itself. The policy network suggests the best moves to make, while the value networks evaluate the positions taken. Then AlphaGo chooses the move that is most successful in its simulations. The self-played games come from a search algorithm that uses the Monte Carlo method, with which the program simulates thousands of games.

Go is a complex game in which two players try to surround each other and gain territory. In the game, a player has a choice of about 200 moves at any time, compared to 20 moves in chess. As a result, according to Google researcher David Silver, it is not possible to explore all possibilities with brute force. Instead, AlphaGo uses a more human approach, limiting the possibilities. This makes the moves seem more based on a system similar to human intuition.

The combination of these ensured that AlphaGo was able to beat European go champion Fan Hui five times in a row. Also, the program has a 99.8 percent success rate against other go programs. In March, AlphaGo will compete against one of Go’s world champions, Lee Sedol.

You might also like