brainwagon

Kasparov fights to a draw in a six game match against Deep Junior. I’m not a very good chess player myself, but I spent a fair amount of time as an undergraduate studying heuristic search, so I am always interested when matches such as this occurs.

Whenever such matches occur, there is a flood of activity on newsgroups and weblogs. Invariably these fall into a couple of simple categories:

• It’s not really AI, the chess program just used brute force. I once heard that AI research was just research into writing programs for which we don’t as yet have good solutions: once we have a solution, it’s just software engineering. I find this comment kind of interesting though, because you don’t see people arguing about cars being able to go faster than humans can run being somehow “unfair”. Humans seem awfully testy when confronted with the idea that something could be smarter than them, although we’ve grown accustomed to the idea of stuff being stronger or faster than we are.

• Chess programs aren’t interesting. It’s all a solved problem.
  I recently became interested in computer chess about a decade of lassitude, and was shocked to find that a number of rather interesting improvements in chess implementation have occurred. They are very interesting programs that require a great deal of finesse and skill to create.

• 
  Chess is simply too easy. They’ll never beat humans at go!
  Go is usually used as the last stand for humans because
  • There are humans who are very good at it.
  • The large branching factor makes minimax and variations intractable.

  I think that this could be mistaken on a couple of levels. Perhaps even the top humans aren’t very good at go. It may be that they are only better than other humans. It seems odd to me to suggest that adding a machine capable of flawlessly examining millions of board combinations per second could in no way increase the quality of play of go players, perhaps at all levels.

Anyway, just some rambling thoughts. Some links:
the Computer Go Ladder, GNU go, and Computer Chess Programming references.