Mountain View
Agency dpa
Google achieved a milestone in the advancement of artificial intelligence: a software developed by the Company won the European champion Chinese strategy game go.
The computers take 20 years to win chess games human opponents, but go, created more than 2 thousand 500 years ago in ancient China, it was far too complex.
The current software gets instead predict the opponent’s movements and learn constantly. On the go, the player reciprocates black and white pieces and is a major way that chess challenge because you have to calculate many more potential movements.
The program solves the problem AlphaGo to predict the most likely movements human opponent and therefore ready for them. The software has “two neural networks” with millions of similar connections to nerve cells, he explained Google in a blog post on Wednesday night.
First, the machine received the information on 30 million movement of experts, after which he could predict the next step that would give human opponent in 57 percent of the time.
Then, networks AlphaGo played thousands of times against each other to go getting better. In tests against other programs go the show won 499 of 500 games. In October AlphaGo he managed to beat the current European champion, Fan Hui, in five out of five games. His next challenge will be in March in South Korea against the best in the world, Lee Sedol.
The defeated champion said that unlike humans, the machine does not make serious mistakes. “Sometimes you are tired, sometimes too eagerly want to win because we feel that pressure,” he told the science journal “Nature”.
The program however, is different. “It’s very strong and stable, is like a wall. For me it’s a big difference. ” If I had not known it was a machine, I would have thought that was a very strong opponent but a very rare person.
The go is a very popular game among developers of artificial intelligence. Within hours of the statement of Google, Facebook founder Mark Zuckerberg wrote that a team of the social network is also working on software for the game.
The world of technology heavyweights have long behind the development of computers that can learn, take on simple tasks that no longer need to make human beings and to organize vast amounts of information.
Google is using them for example in the automatic recognition of the content of photographs in its photo service. Facebook is focused on speech recognition, according to Zuckerberg.
AlphaGo was developed by the British company Deep Mind, bought by Google two years ago. Demis Hassabis cofounder stressed that the software continues to develop itself. “It’s different from a program that one has built and who knows in all its nuances what you can do. Here learned things himself, so it is shocking to see how power is appropriate, “he told” Nature “.
Hassabis created Deep Mind with the ambitious goal of teaching thinking computers.