The US company Google DeepMind created a computer program that defeated for the first time, a professional oriental strategy game Go, a study published Wednesday in the journal Nature.
One of the great challenges of artificial intelligence was get a win against a professional Go, an oriental game whose success is rooted, among other things, the ability of the player to evaluate positions and movements on a board.
The new software, the AlphaGo, defeated in October, in London, in five games session, the Chinese professional Fan Hui, based in France and Europe champion.
Up To date, the Go computer programs more successful “played” with amateurs, but had never won a professional called “Oriental chess”, with ancient origins of ancient China.
The AlphaGo, according its creators, will enable significant advances in the fields of artificial intelligence, which at the moment seem inaccessible to researchers.
For the experts from Google DeepMind, the computer program uses “value networks” to evaluate positions on the board and “tactical networks” to select the movements of the opposite color pieces.
The artificial neural networks AlphaGo are “trained” in this game through an educational process overseen by a human being which, in turn, is reinforced by learning the machine gets to play against itself.
In games against other Go programs, AlphaGo won 99.8% of the clashes, including with professional Fan Hui, say the researchers in an article published in Nature.
The next challenge is to sample the software in a game with the world champion, the South Korean Lee SEDOL in March in Seoul.
Go game confronts a board two players, which should place alternately black and white stones on the free intersections of a grid.
The aim is to dominate the largest possible surface of the board, to which players must choose between putting the stones together, and secure them to prevent them from being captured, or separate them so that they can gain ground in the box.
No comments:
Post a Comment