Sign In

Communications of the ACM

ACM TechNews

Deep Learning Machine Teaches Itself Chess in 72 Hours, Plays at International Master Level

View as: Print Mobile App Share:
An Intel Labs Seattle robot programmed to play chess by the University of Washington artificial intelligence group.

A researcher at Imperial College London developed an artificial intelligence machine that taught itself to play chess at the master level.

Credit: University of Washington

In the nearly 20 years since IBM's Deep Blue supercomputer beat world chess champion Gary Kasparov, the approach to developing chess-playing computers has changed little. Most successful chess engines continue to rely on searching through all possible future moves to find the best next move. However, Imperial College London's Matthew Lai has developed a new artificial intelligence (AI) machine called Giraffe, which he says can play the game at a master level while approaching chess in a more human way.

Lai's Giraffe is built on an artificial neural network and is designed to play chess similarly to the way humans do: not by considering all possible future moves, but narrowly considering the set of moves that are likely at any given point in a game.

Lai trained Giraffe on a dataset of 175 million chess positions he generated in part from a random sampling of positions from a computer chess database. He then had Giraffe play games against itself to improve its ability to evaluate future positions. Giraffe chooses the "best" move 46 percent of the time and places the best move in its top three ranking 70 percent of the time.

Lai says Giraffe is able to play at the level of an FIDE (World Chess Federation) International Master when running on a mainstream computer.

From Technology Review
View Full Article


Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA


No entries found