Fascinating Documentary About Kasparov vs. Deep Blue

The subject matter and most of the content is good, but you have to ignore the stupid whispering of the narrator. I think the director was trying to portray "thoughts in the head" of Kasparov or perhaps someone watching at a tournament, but it sounds more like a golf announcer broadcasting from the edge of the eighteenth green at a major championship. "Looks like he's got an eight iron there, Bob." Completely unnecessary.

The other irritating thing is the contrived big corporation conspiracy theory angle. I think they had to play that up for plot, for purposes of drama, and much of the material along this line is of the terribly trite variety: "the heartless, evil corporation cares about nothing but money." Kasparov had a legitimate beef about the way IBM handled some things, but my guess is they egged him on with questions designed to elicit philosophical-sounding answers about corporate power. The documentary isn't filled with it, however, but it is sprinkled here and there with the narrator hinting at times about unknowable evil.

Here is what happened: Deep Blue played the first game like a computer and got blown away. During a critical move in game two, after spending 15 minutes calculating at 200 million positions per second, Deep Blue made a very human-looking move. The thing is, a few moves later it missed a very basic move, but in typical computer fashion. By then, however, Kasparov was blinded by frustration, and he missed it. Regardless, how could this happen? That question becomes the thing that defeats Kasparov in the end. After game two, Kasparov demanded documentation of the computer's analysis. IBM refused, and Kasparov couldn't shake his anger and suspicion.

The best thing about the documentary is that you get a glimpse of Kasparov's inner world, his anguish and, maybe, a touch of regret about the way he handled things. Most are familiar with what appeared to be spoiled-brat rudeness at the end of the final game where he resigned and stormed away from the table. That clip was played over and over on television. Kasparov came off like a jerk. Well, it turns out there were real reasons for his response that went beyond simply reacting angrily from losing. There was a certain amount of jackassery from the IBM team that had been building over the course of the week, and Kasparov couldn't recover from that. He describes what might be referred to as "intrusive thoughts" that centered on the difference between the machine's performance between game one and game two. Kasparov couldn't get past his emotional reaction, and it wore him down. The IBM team knew that. He was defeated before the last game even started.

As they explain in the documentary, playing a human is one thing, and playing a computer is another, but playing a human and a computer is impossible. That is, a human helping a computer at this level is unbeatable. The person need not be a grandmaster either. They simply need to let the computer know what lines of attack to abandon. This is because of something called the Shannon number. The Shannon number is the number of possible moves in the typical 40-move chess game: 10¹²³ variations. For comparison, the number of atoms in the observable universe is only 10⁸¹. From Claude Shannon's paper: "A machine operating at the rate of one variation per micro-second would require over 1090 years to calculate the first move!" And therein lies the problem for computers and programmers and the way they can be beaten. A computer can't just look at the board and eliminate dumb lines of analysis. So they have to find ways to program around that.

One of the knocks against Kasparov is the argument that he was the best in the world, so it wouldn't matter if a human was helping the computer. But the suspicion wasn't that a human was playing the game; it's that a human was eliminating fruitless lines of analysis. It's a compelling idea, and the documentary could have spent more time developing it by explaining the way in which moves were relayed from the computer to the playing board and back again. This was just hinted at: that Deep Blue was behind some locked doors surrounded by all these IBM people--including hired grandmasters--and private security, and who knows what was going on?

There are other important factors. For example, IBM hadn't provided a record of Deep Blue's prior games. It turns out the contract stipulated providing a record of "public games." Deep Blue had only played private games. So the Deep Blue team had every game Kasparov had ever played in a tournament, and Kasparov had no Deep Blue games to examine. In a match against a human, Kasparov would have been able to examine prior games.

Anyway. It's worth a watch.

I think Kasparov's suspicions are justified. He knows what he's talking about. An experienced chess player--especially at that level--can tell computer from human and estimate within a few points another player's chess rating just by watching them play for a little while. In the same fashion, for example, an experienced highway patrolman can glance at a car and estimate its speed within one or two miles per hour. It's what they know.

IBM wouldn't give him a rematch.

In any future computer vs. human game at this level, the computer should be required to sit on the stage across from the grandmaster, and the person relaying the moves must not be given any opportunity to adjust the computer's analysis. Better yet, require the computer to "see" the chess board so that no human input is required at all--so that the computer just displays its response. That would eliminate any dispute, and we could see that hired grandmasters aren't poking away at keyboards. And all that security? No one is going to stick the computer in their pocket and run away with it. Just put a cop nearby. Good grief.

Follow me on Twitter.

posted by rdbrewer at 04:58 PM

| Access Comments