KASPAROV AND DEEP BLUE
by Tim Dees
It was roughly ten years ago that the final Kasparov-Deep Blue match took place. If you don't remember, that match was the second of two matches that pitted the world's greatest chess player against an IBM supercomputer, nicknamed Deep Blue. In the first match, the computer put up a strong challenge, but eventually crumpled.
The second match, however, went quite differently. The IBM programmers made demands that Kasparov found tough to accept, such as the ability to tweak Deep Blue's software between games. Kasparov eventually relented. During the match, Kasparov noticed that the computer was making moves of exceptional creativity and originality. He had never seen a computer make such moves before. He accused the programmers of cheating, either by using a human to make some moves, or by reprogramming Deep Blue in mid-game. To prove they were cheating, Kasparov asked to see the log files. The programmers refused.
To this day, Kasparov maintains that the Deep Blue programming team swindled him. But the more interesting thing is that both Kasparov's earlier win and later loss against Deep Blue demonstrates something profound about the human mind.
When a programmer teaches a computer to play chess, he essentially has it analyze every possible board state. So it takes every possible move and analyzes it based on the fallout from that move. This takes enormous processing power. That's why Deep Blue had to be a supercomputer, and that's why computers have gotten better at chess as they've gotten faster. But the human mind works nothing like that. The brain has nowhere near the processing power to compute trillions of possible board states. So it must be playing by some other system, and a system that is far smarter than anything we've come up with on a computer.
There are other games, however, for which we understand how the brain works. Backgammon, for instance. In backgammon, a computer that uses the same processes that Deep Blue used (looking at each possible board state given trillions of possible moves) will lose to a below-average player consistently. In the 1970s, however, computer scientists started using neural networks to play backgammon. Neural networks are systems that work very much like neurons in the brain. After using the neural network programming, the machine was still terrible at backgammon. But then the programmers tried something different: they allowed the computer to play a few hundred games to train the neural network to play the game. After that, the computer could handle even the best opponent.
Neural networks can run on slow computers (like the brain), so computers have gotten no better at backgammon since the '70s. But neural networks have been unsuccessful at playing chess. So we're still left wondering what's going on in Kasparov's brain.
Comments Requested - I would love your thoughts on the relationship of brains to computers, and the differences between minds, consciousness and computational machines. Also, if anyone has knowledge of pattern recognition vs. sequential processing, that would be cool as well.