Remember Deep Blue? Back in 1996, IBM created a chess-playing computer that beat the then world Champion Gary Kasparov. That machine’s impressive results were the result of it being able to evaluate over 200 million positions per second. Naturally, Kasparov didn’t take that approach. So it was really a triumph of brawn over brains. Since then chess programs have gotten better, though usually because the programmers encoded strategies in the software.
Fast forward to today and playing chess is still a worthy problem to tackle. This time however the emphasis is on learning rather than brute force evaluation. Matthew Lai has created a system that learns the rules and discovers patterns from analysing previous games by humans and machines. Much like a human chess player does. It also played itself, trying to improve it’s techniques.
Using an Artificial Neural Network learning system has been a successful approach and Lai’s system has now reached International Master Level. It’s quite possible that this AI could even discover new strategies and tactics for winning as it continues to learn.