Humans and Computers, Chess and Go
Saturday, May 10, 2003
Xbit Laboratories has a history of the highly publicized chess matches between the grandmasters and their silicon counterparts. It starts off with the Kasparov/Deep Blue match of 1997, a debacle for computers, and moves on to later matches with more ambiguous outcomes. Though the article tends to stack events in favor of humans, it's an interesting read nevertheless.
Towards the end, the article attempts to analyze the computers' weaknesses. Specifically, computers tend to rely on brute force calculations, which work well when there's only a few pieces remaining. However, the computer might not see an obvious, early checkmate because it can't calculate more than a few moves out on any piece.
The author takes some privileges with technology. He compares a multi-CPU computer to several different people exchanging ideas. Usually, this isn't the case. All of the CPUs are working on the same problem at the same time. The behavior the article describes would be more like a human analyzing the board using both halves of the brain.
Additionally, the computer is apparently a "cheater" if it tries to throw an opponent off balance. However, when the humans play classic openings out of sequence to confuse the computer, there's no such disparaging comparisons.
A final comparison is drawn between the controversial matches and the current state of Artificial Intelligence (AI). This comparison is fundamentally flawed as, by definition, chess programs do not represent AI. Intelligence goes into making the algorithms which decide the best move. However, as humans wrote all of the algorithms, it's human intelligence.
The computer's true weakness is its inability to identify patterns. We've spent our whole lives identifying patterns, from our mother's face to traffic congestion. After years, these patterns become very obvious to us, just as defensively weak patterns are obvious to chess grandmasters.
If you've tried to OCR a document (scan it to convert it to text) you know how poor computers are at recognizing something as simple as the letter "l." This is why Palm Pilots force you to scribble in a specialized "graffiti." The same is true of speech recognition software.
Computers have yet to attain the pattern recognition capabilities of a child. So, though faster computers with highly optimized algorithms may be able to make short work of chess grandmasters in the near future, Go may take more time. Of course, programmers are already taking up this new challenge.