Essays > Go AI
to Go Programming and AI
(Original Essay - Printable Version)
Chess has long been the focus of much research, and now, finally, computers are competative with the best of human players. Only a few years ago, the computer Deep Blue II defeated the World Champion Gary Kasparov. But now what is there for AI to do?
Go is the next frontier in computer board game Artificial Intelligence. The game is simple enough for one to learn all of the rules in a day, but so far, no computer has managed to even compete with any professional players. The best programs in the world are beaten regularly by smart school-children!
The computers lack ability because Go is a much larger game than chess. In an average chess position, there are about 15 to 25 legal moves. In Go, that number is usually around 250, making it virtually impossible to create a program based entirely on the minimax game tree in the way that Chess programs are written. Brute force no longer works so well.
There is also another problem. Go professionals often play by instinct. They often say such things as "this move feels good" or "that move seems a little slack" and "that seems to submissive." Needless to say, it is very difficult to consult with a Go professional about Go when the professionals themselves have a difficult time explaining the reasons behind some of their own moves!
Thus, Go programming requires a different route from chess programming. Instead of simulating the techniques human players use to play the game, the programmers must model human thought itself! Not only must a computer now act like a person, but it must also think like a person!!
Because of this, Go has been a testing ground for many different AI techniques including pattern matching, neural networks, and the genetic algorithm. Even with all of these new techniques, however, the minimax game tree is still at the heart of many Go programs, for it provides a framework for the rest of the programs and algorithms.
Go programming is the new frontier, one of the toughest human strongholds. What could be a more crowning achievement than to have a computer beat humans at their own game?