A Computer Algorithm Just Beat a ‘Go’ Champion for the First Time

By Jonathan Zhou
Jonathan Zhou
Jonathan Zhou
Jonathan Zhou is a tech reporter who has written about drones, artificial intelligence, and space exploration.
January 27, 2016 Updated: January 28, 2016

In 1952, a computer program mastered tic-tac-toe. More than 40 years later, in 1994, a program became the checkers world champion. IBM’s Deep Blue beat chess Grandmaster Garry Kasparov in 1997 and today’s algorithms can easily defeat the world’s leading players.

But the game of Go has held out from being dominated by machines. The number of possible positions in Go, which is usually played on a 19 by 19 board with 361 intersections, is vastly larger than that in chess. This makes the traditional “tree search” model used by chess programs ineffectual for Go.

For years, the best Go computer programs could only defeat skilled amateurs at best, which is why the defeat of Fan Hui, the three-time European Go champion, by Google’s algorithm AlphaGo has been called a breakthrough.

The program didn’t defeat Hui by merely performing more calculations than its predecessors. It combined the traditional search for possible positions, as used by programs like Deep Blue, with neural networks utilizing parallel computing. The program trained on millions of games played by humans, then “self-trained” by playing against itself.

“This is the first time that a computer Go program has defeated a human professional player, without handicap, in the full game of Go—a feat that was previously believed to be at least a decade away,” researchers at Google’s DeepMind program wrote in Nature.

What makes the results of special significance to AI researchers is the program’s “intuitive” decision-making, which isn’t fully comprehensible to its creators.

“As shown by its results, the moves that AlphaGo selects are invariably correct. But the interplay of its neural networks means that a human can hardly check its working, or verify its decisions before they are followed through,” Nature’s editorial board writes.

“As the use of deep neural network systems spreads into everyday life—they are already used to analyze and recommend financial transactions—it raises an interesting concept for humans and their relationships with machines. The machine becomes an oracle; its pronouncements have to be believed.”

Jonathan Zhou is a tech reporter who has written about drones, artificial intelligence, and space exploration.