Computers have already beaten the best humans at Checkers, Chess, and Jeopardy, but mastery of the ancient Chinese game Go has eluded computer scientists for the longest time. Deepmind, Google's artificial intelligence division, claims to have made a big breakthrough using neural networks, and it recently took on and defeated a Go world champion.
As Google's blog posts puts it, "There are 1,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000 possible positions" in Go, which makes the game very complex. Mapping out every possible move won't work for Go, making it the one game computers can't beat humans at. Performing well at Go is considered by some to be a benchmark for artificial intelligence.
Deepmind's AI, called "AlphaGo," recently took on reigning three-time European Go champion Fan Hui and won all five games. In March, AlphaGo has a throw-down scheduled with Lee Sedol, who has been the top Go player in the world for the past decade.
Deepmind is the same group that put a neural network to work on classic Atari games like Breakout. Unlike previous computer game programs like Deep Blue, Deepmind doesn't use any special game-specific programming. For Breakout, an AI was given a general-purpose AI algorithm input from the screen and the score, and it learned how to play the game. Eventually Deepmind says the AI became better than any human player. The approach to AlphaGo was the same, with everything running on Google's Cloud Platform.
The group says that because its approach is general purpose, "our hope is that one day they could be extended to help us address some of society’s toughest and most pressing problems, from climate modelling to complex disease analysis." For now, though, Google still needs to defeat the world's greatest Go champion. Hopefully there will be a livestream where we can watch it all go down.
Source: Arstechnica
0 comments:
Post a Comment