All systems Go with artificial intelligence

All systems Go with artificial intelligence

Artificial intelligence has taken a leap forward thanks to researchers at Google, who have created a program that can beat a human champion at the board game Go.

It's a step up from the game of chess, leaving the computer with a larger board space and possible moves.

Based on the traditional Chinese pastime, Go is a two-person strategy game with black and white pieces in which players aim to claim more territory than the other.

Go has been regarded as a "grand challenge" for artificial intelligence because of its complexity.

Standard Go boards have a 19x19 grid, compared to a standard chess board, which is an 8x8 grid.

The results of the program have been documented in Nature today, in which researchers say the findings "provide hope that human-level performance could potentially be achieved in other seemingly intractable artificial intelligence domains".

David Silver, Aja Huang, Demis Hassabis and colleagues developed the AlphaGo program, which uses value networks to evaluate the board positions and policy networks to pick its next move.

Those networks are trained using a combination of supervised learning from human expert games and from playing against itself.

Prior to the Google's more advanced program, the most successful computer players have been at the level of an amateur human player able to be defeated by a professional human opponent.

AlphaGo had a 99.8 percent win rate against other Go programs and beat the human European Go champion 5-0.

The convincing win is the first time a computer has beat a professional player on a full-size Go board with no handicap -- an achievement boffins thought was more than a decade away.

But another challenger is on the horizon for AlphaGo.

Lee Sedol -- considered the world's best Go player over the past decade -- will challenge the program in a match in Seoul in March.

3 News