Google’s Alpha-Go: Software kicks the Go champion – Handelsblatt

man vs. machine

“8”

Google’s software Alpha-Go has beaten top players Lee Sedol surprisingly in the first game of the Asian board game.

(Photo : Reuters)

Seoul 1: 0 for Artificial intelligence: A computer program from Google on Wednesday surprisingly won the first five games of the Chinese board game Go against the South Korean champion Lee Se-dol. The remaining four games to be discharged by Tuesday.

The software AlphaGo won on Wednesday in the South Korean capital Seoul, the first of five planned games. The 33-year-old Korean Lee Sedol shed around three and a half hours on Time. He was previously hopes of overall victory, but acknowledged that people made more errors than machines.

Experts for Artificial Intelligence had until recently predicted it would take another decade to computer professional Go could beat poker players. However, Google’s Alpha-Go program then had last year already defeated a top European player.

In chess, the former world champion Garry Kasparov was measured in two competitions with the IBM computer Deep Blue in 1996 and 1997. While the first match even 4: 2 won for themselves, Kasparov had in the second race with 2,5: 3,5 beaten. Since then, other competitions have underpinned the result, even the strongest chess players today have against powerful computer programs no chance more.

Go has at first glance simpler rules than chess. Nevertheless, the Chinese board game is a bigger challenge for computer, because there are many more possible moves: Do the figures more than 64 fields are moved in chess, so the Go players are 361 fields are available to place his checkers

.

through clever set of white and black stones try the opponents, to gain as much space on the game board. The problems encountered are so complex that the successful chess “brute force” method, when the computer goes through all possible sequences of moves with sheer computing power, promises no success.

The technology developed by Google software Alpha- Go solves the problem in that it predicts the likely features of the human opponent and locks to that. This is made possible by two artificial neural networks with millions of compounds, similar to the human nervous system

Both networks have been optimized by the developers in different ways for their task. While a 30 million trains from encounters human Go poker players evaluates, makes the other thousands games against itself.