ExplAIning A.I.

The cutting-edge challenge for artificial intelligence is a game that is over 2,000 years old. Known simply as Go, the game is played on a board of 19 rows and columns where players alternate placing back or white markers, called stones.

For many years, computer scientists labored over chess and backgammon algorithms. Each of these ancient games has largely succumbed to the power of artificial players. In the case of chess, the grand-masters of the world have been defeated since about 2007 by machines that learned from classic matches between humans. the reigning Backgammon A.I. camp is a program by Gerald Tesauro named TD-Gammon, which uses something called temporal difference learning as well as the well-known neural network technique; it plays against itself and learns from the outcome!

Go is the new frontier, and it appears that the manifest destiny of A.I. is to conquer this space as well. The game is quite different from previous challenges because it is easy for the person who is ahead to lose that advantage. It was thought that this difference would keep A.I. players at bay for some time. But in recent months, machines have started beating the best human players, sometimes.