Man vs. Machine

The Kasparov - Deep Blue Rematch

Six games, May 3-11, 1997, at the Equitable Center in Manhattan.
Reports by Hector Kim.

Jump to: About me | Last year in Philadelphia | What is Deep Blue? | The games | Why I still think Garry wins | Thursday night at the club

About me and this page

I'm an actuary in northern New Jersey, which means I spend my days on probabilities and my Thursday nights on chess, at our club in the basement of Grace Lutheran. My USCF rating has wandered around 2050 for six years: Expert, on a good night. I can tell you that the knight on the top of this page is the piece I misplay most often.

I don't write move-by-move notes here. Plenty of people who are much stronger than I am are doing that. These are reports for people who want to know what happened and what it felt like, written by somebody who has spent thirty years losing to people and is now watching the best player in the world try not to lose to a refrigerator.

Last year in Philadelphia

In February 1996, in Philadelphia, Deep Blue beat Garry Kasparov in the first game of their first match. It was the first time a computer had beaten a reigning world champion in a game at regular tournament time controls, and half of our club thought the end had come. Then Kasparov won three and drew two of the next five games and took the match 4-2, mostly by steering into quiet, closed positions where calculating ten moves ahead counts for less than understanding the position. The other half of the club said they'd told us so. I've been waiting fifteen months for the rematch.

What is Deep Blue?

It's an IBM RS/6000 SP supercomputer with hundreds of special chess chips built for this one job, and it looks at about 200 million positions a second. That's roughly twice as fast as last year's machine. A grandmaster looks at a few positions a second, at most, and throws most of the possibilities out without looking at them at all. The machine doesn't get tired, doesn't get nervous, and doesn't care who is sitting across from it.

The project goes back to Carnegie Mellon in the 1980s, when the machine was called Deep Thought; Feng-hsiung Hsu and Murray Campbell were there from those days and are still on the team. This time they brought in grandmaster Joel Benjamin, a former U.S. champion, to help with the openings. A member of the team sits at the board and moves the pieces for the machine.

The players sit in a small room upstairs at the Equitable Center on Seventh Avenue, with a few cameras. The audience watches downstairs in an auditorium, on big screens, while grandmasters talk through the game on stage. The winner gets $700,000 and the loser $400,000, which is more than the club treasury, but not by as much as you'd hope.

The games

Game 1, Saturday, May 3, 1997: Kasparov wins.
Kasparov had White and played it slow and quiet, the way he finished last year: no early fights, nothing sharp for the machine to sink its teeth into. He outplayed it slowly, and the machine never got any real counterplay going. Near the end it played a rook move none of the commentators could explain, and a move later the IBM team resigned for it. One game, one point, and I went to bed happy.

Game 2, Sunday, May 4, 1997: Deep Blue wins.
I was there for this one. I took the train in and got a seat in the auditorium. Deep Blue had White and played a Ruy Lopez, and then did something I didn't expect from a computer: it played like a patient positional grandmaster. At one point the commentators were sure it would grab a pawn, the kind of thing computers always do. It didn't. It played a quiet move that kept Kasparov tied up, and it kept squeezing. Kasparov resigned on move 45, and the auditorium made a sound I have never heard at a chess event, as if somebody had let the air out of the room.

Then came the part that hurts. Over the next day, analysts went back over the final position and found that Kasparov might not have been lost at all. There seemed to be a perpetual check in it, which is a draw. The world champion may have resigned a drawn game. I can't think of the last time that happened at that level. He hasn't hidden that he's unhappy about the game, and he has asked to see the machine's printouts from it.

Game 3, Tuesday, May 6, 1997: Draw.
Kasparov opened by moving his queen's pawn one square, something you'd expect from a beginner and almost never see from a grandmaster. The idea, as far as anyone could tell, was to get out of the machine's opening preparation right away and into a position nobody had programmed for. It was a long, careful game, and it was drawn.

Game 4, Wednesday, May 7, 1997: Draw.
Deep Blue had White again, and Kasparov answered with the Caro-Kann, a solid defense he hardly ever plays. It went deep into an endgame, and it was drawn. Then came two days off, and everyone at work wanted to know what was going to happen. I told them what I tell them about everything: it depends.

Game 5, Saturday, May 10, 1997: Draw.
Today. Kasparov had White and went back to the quiet setup from the first game. He pressed for most of the afternoon, and for a while I thought this was the day. The machine held, and it was a draw. Two and a half points each, with one game left.

Game 6, Sunday, May 11, 1997.
Deep Blue has White. Report here Sunday night.

Why I still think Garry wins

He has Black tomorrow, so a draw is the realistic hope, and a draw ties the match. A win would be a gift. But look at where we are. After the second game he looked shaken, and some of the sports pages sounded ready to give up on the human race. Since then he has played three draws, and in two of them he was the one pressing. He's had five games to learn how this machine plays. He's the best there has ever been at learning his opponent.

Speaking as an actuary, here's my honest table for tomorrow: a draw, 55 percent. Kasparov wins, 20 percent. Deep Blue wins, 25 percent. Speaking as a chess player who has lost a lot of games to people who calculated better than he did, I'm with the human, and I don't care who knows it.

Thursday night at the club

This Thursday we skipped our ladder games and set up the final position of Game 2 on three boards. Greg Palmer had run it through Fritz on his laptop at home and brought in a printout of the drawing line, and we went through it twice. Bruce Mazzola, our club president, said he'd have resigned too, which made us all feel better about Kasparov and worse about Bruce. Melvin Cox, who has been coming on Thursdays since the club started in 1971, said that every champion before Kasparov at least got to play somebody who had to go home and sleep. Then he beat me in nineteen moves, which I'm choosing not to report in detail.

There's a pool going on Game 6, a dollar a square. I took "draw by move 50."

How I'm following it

IBM has the moves on its Web site as they're played, with commentary. I keep a board set up on the kitchen table and play the moves out as they come in. My wife Christine says the kitchen table belongs to the match until Monday, and then it belongs to her again.

Game 6 tomorrow, and I'll be at the board with my coffee.