The Chinese board game known as Go is generally considered to be the oldest and most complex strategy board game on the planet. Played by 40 million people worldwide, the game is more mathematically complicated than chess by several orders of magnitude.
If you've ever played the game, you know the deal. I tried once and sprained a frontal lobe.
Putting The Art In Artificial Intelligence: Photos
Go is so enormously complex, in fact, that even the world's most powerful supercomputers have been unable to play the game above the amateur level. Until now.
As reported in this week's edition of the journal Nature, researchers at Google have developed an artificial intelligence program that can compete with professional players at the renowned strategy game. In fact, the program - called AlphaGo - recently defeated the reigning three-time European Go champion Fan Hui, five games to zero.
As when the supercomputer Deep Blue defeated chess champion Garry Kasparaov in 1997, the achievement may eventually be considered a watershed moment in the development of A.I. and supercomputers. What's more, the AlphaGo system essentially learned to master the game on its own.
A.I. Takes A Stroll Through Amsterdam
In a blog post accompanying the release of the Nature article, Google DeepMind researcher Demis Hassabis writes that the AlphaGo program uses neural network technology - systems designed to function like the human brain - to navigate the game's subtle dynamics.
"We are thrilled to have mastered Go and thus achieved one of the grand challenges of A.I.," Hassabis writes. "However, the most significant aspect of all this for us is that AlphaGo isn't just an ‘expert' system built with hand-crafted rules; instead it uses general machine learning techniques to figure out for itself how to win at Go."
The details really are fascinating: Hassabis and the design team initially trained the program using records of 30 million individual movies played by human experts. After that, they cut the A.I. loose to create its own strategies by playing thousands of games between its own neural networks 'Ex Machina': Science Vs. Fiction
"These neural networks take a description of the Go board as an input and process it through 12 different network layers containing millions of neuron-like connections," Hassabis writes.
Once AlphaGo got the hang of things, as it were, it posted a 99.8 percent winning percentage against other Go programs, going 499-1 in a 500-game test run. Up next: AlphaGo will take on world champion Lee Sedol - considered the top Go player in the world over the last decade - in a five-game match in Seoul, South Korea, in March.
Get your tickets now. You can read more at the AlphaGo project page, or check out Google's video below.