Google says that its AlphaGo Zero artificial intelligence program has been successful in chess against the world’s leading software specialist within the hours of teaching in itself the game from scratch.
The company’s DeepMind division, says that he played 100 games against Cod 8, and won or drew all of them.
The research has yet to be peer-reviewed.
But experts already suggest that the achievement to strengthen the firm’s position in a competitive sector.
“From a scientific point of view, is the most recent in a series of dazzling results that DeepMind has occurred,” the University of Oxford’s Professor Michael Wooldridge told the BBC.
“In general, the trajectory of DeepMind seems to be that in order to solve a problem and then demonstrate what you can really ramp up performance, and that is very impressive.”
DeepMind has already defeated several of the best human players of the Chinese board game Go, as well as the teaching itself how to play video games including Pong and Space Invaders.
The team in London is currently trying to develop a system that can beat humans in the space strategy game Starcraft, which is seen as being even more complex challenge.Think of the time
Google is not commenting on the investigation until it is published in a journal.
However, the data published in the Cornell University Arxiv site status that an algorithm nicknamed AlphaZero was able to beat Stockfish only four hours after you have received the rules of chess and is said to learn playing simulations against himself.
In the 100 games that followed, each program is given a minute to think about the time of each movement.
AlphaZero won 25 games in which he played with white pieces, making the first step, and the other three in that it plays with the black pieces.
The two programs drew the remaining 72 games.
DeepMind describes the level of performance achieved as “superhuman”
Google stressed that Stockfish 8 earlier he had won the year 2016, the Top Chess Engine Championship. The software was first launched in 2008 and has been built by volunteers in the years since then.
The open source project has been hit by another program, Komodo, two great chess team of the challenges of this year.
Even so, one of the great masters of chess was still enormously impressed by DeepMind victory.
“I’ve always wondered how it would be if a top species landed on earth and showed us how they played chess,” Peter Heine Nielsen told the BBC.
“Now I know”.Open v closed
AlphaGo Zero recent achievements will not rest in the chess only.
The paper says he was also triumphant in the Japanese board game of Shogi in front of one of the main artificial intelligence program named Elmo, after two hours of self-training.
The AlphaZero algorithm won 90 games, drew two and lost eight.
In addition, after eight hours of self-training was also able to defeat the previous version of itself on Go – the winner of 60 games and losing 40.
Prof Wooldridge pointed out that the three parties were quite “closed”, in the sense that it had limited sets of rules to deal with.
“In the real world we do not know what is around the corner,” he explained.
“Coping when you don’t know what is to come is much more complicated, and things are set to get even more exciting when DeepMind moves in the most open of the problems.”
Princeton University AI expert Prof Joanna Bryson adds that people should be cautious about buying too deeply into the signature of the hype.
But he added that his ability for the good publicity had put him in a strong position against the applicants.
“It is not only about hiring the best programmers,” she said.
“It is also very political, because that helps Google as strong as possible when negotiating with governments and regulators looking at the AI of the sector.”