Taking after four losses, one of the world’s top Go players – Lee Se-dol – has beaten Google DeepMind’s AlphaGo program. Lee Se-dol, who has been defined as the Roger Federer of Go, has so far just figured out how to beat this AI once out of his four played games, so finally, AlphaGo has already won the set.
Back in October, AlphaGo played against and defeat the three time European Go champion Fan Hui, winning every one of the five games. It also looked like Lee Se-dol would lose each of the five of his games too yet figured out how to turn AlphaGo’s triumphant streak on its head by causing the A.I. to make a fault that it couldn’t recover from.
The AlphaGo AI project is different from “expert” systems which use hand-created rulesets. AlphaGo rather uses machine learning methods to work out how to enhance its playing style, actually playing games of Go again and again until it got to a state where it can play intensely against world class players.
Building up an AI for Go was thought to be a difficult task because of the enormous measure of positions a player can make. In Chess, the measure of moves a player can make is around 20 for the normal position (current state of the board), in Go anyway it’s around 200. In Go there are numerous positions the board can be in. Indeed, there are more positions than there are atoms in the Universe – or 2.08×10^170.