Google Develops First A.I. To Beat Professional Go Player

AH 2015 A New Google LOGO 1

Those who have played the ancient board game Go will agree; it’s impossible to completely master the millions of possible strategies, let alone to analyze an opponent’s moves and predict which move will go best, which involves millions more possibilities. A great many researches and companies have been trying for some time to develop an A.I. that can consistently win at Go. It has long been considered the game of all games for an A.I. to learn and master. It may seem like a mere novelty, but the kind of analysis and prediction an A.I. would need to do in order to master the game of Go would have incredible implications in just about every A.I. related field, though it would take time to develop. For a while now, Google-owned DeepMind, who managed to build an A.I. that absolutely slays Atari games before Google bought them, and Facebook have been neck and neck, both very close to solving the puzzle. According to an announcement today, Google has managed to pull it off. Facebook also happened to post yesterday that they were “very close” and had only a “lone researcher” working on it.

DeepMind’s A.I. apparently managed to best a national champion Go player in 5 games in a row. Using what were called value networks and policy networks in conjunction with the same sort of tech Google uses for search, the A.I., called AlphaGo, was able to pull out a victory. This was not the first time it had beaten a human professional, however; back in October, AlphaGo was able to outplay the three-time champion of Europe’s Go scene.

A win rate of 99.8 percent against other Go playing A.I.s was also achieved by AlphaGo, cementing it as the clear first and best to meet the insurmountable A.I. challenge of mastering the millennia-old game of Go. The implications here may not be entirely clear to most, but this was called a historic breakthrough by researchers. The foremost application that comes to mind is personal assistant technology, but such complex search and prediction algorithms will doubtlessly have tons of other uses in the near future.