Google's DeepMind Just Won the Race to a Huge Artificial Intelligence Breakthrough

Wednesday, 27 January 2016 - 5:22PM
Wednesday, 27 January 2016 - 5:22PM
Google and Facebook are currently vying for the top spot in artificial intelligence right now, and Google's Deepmind has just won the latest battle, in the form of a game of Go. Learning this complex board game is often considered to be a major milestone for AI, and Google's Deepmind has not only learned it, but has become the first AI to consistently beat a human champion player.

This ancient game, which originated in China over 2,500 years ago, may seem an odd choice to test artificial intelligence, but it is incredibly difficult for a computer to master. There are many different variables and scenarios to consider, as Go has 10 to the 700 power variations of the game. Chess, which was once considered to be the holy grail for artificial intelligence, has only 10 to the 60 power variations. 

The program, called AlphaGo, was specifically designed to play and master Go using extremely sophisticated neural networks. Specifically, it uses "value networks" in order to evaluate the positions of the pieces on the board, and "policy networks" to plan its next move. The researchers took a novel approach in training the program, and combined both supervised observation of adult expert games of Go and trial-and-error from self-play. As a result, the program was able to beat other Go computer programs at a rate of 99.8%, and ultimately defeated the human European champion, Fan Hui, in five games in a row. 

Opening quote
"The game of Go has long been viewed as the most challenging of classic games for artificial intelligence owing to its enormous search space and the difficulty of evaluating board positions and moves," the researchers wrote in their paper, published today in Nature. "This is the first time that a computer program has defeated a human professional player in the full-sized game of Go, a feat previously thought to be at least a decade away."
Closing quote

Completing this milestone doesn't just give Google bragging rights; this research could have profound implications on several different aspects of artificial intelligence, most pertinently its predictive power and the speed at which it can search for a sequence of actions. This could lead to improvements in many different areas, such as facial recognition and predictive searching (not that Google needs to be much better at that, it's creepy enough as it is).

Following this announcement, DeepMind challenged the reigning world champion, Lee Sedol of North Korea, to a match. Sedol is considered to be the best human player in the world right now, so if there are any limits to this technology's ability to play Go, Sedol will be able to expose them.
Artificial Intelligence

Load Comments