His name is AlphaGo , a software designed by Google DeepMind and he is the new European Go champion . For the first time in history, a machine has defeated a professional gamer in this ancient oriental strategy game.
The Go (which arose in ancient China about 3,000 years ago), represents one of the best challenges for artificial intelligence due to the diversity and complexity of movements and possible positions on the game board. It is based on the fact that two players alternately place pieces in the shape of small stones – in black and white – on a grid-shaped board in order to occupy more territory than the opponent at the end of the game.
The algorithm developed by the artificial intelligence division of Google has achieved a success rate of 99.8% against other Go programs and in its duel against Fan Hui (three times professional Go champion in the European competition) AlphaGo won in a five game tournament. The human has been defeated by the machine.
How does this software operate?
AlphaGo uses “value networks” to evaluate table positions and “political networks” to select its own movements ; all this was the result of a learning process supervised by humans and reinforced with games that the software played against itself (impossible not to remember “War Games” from 1983, right?).
“After the chess match between Gary Kasparov and IBM’s Deep Blue computer in 1996, the goal of some AI researchers to beat the best human Go players has been an exceptional challenge, as the complexity of Go outweighs the chess, both in number of moves and in game planning, ”said Jon Diamond, president of the British Go Association .
The next challenge will be to beat the world champion, Lee Sedol . Will it succeed? We will see it next March.
The work has been published in the journal Nature .
