Commentary for black move 16:

YASSER SEIRAWAN: Well, I don't think that the IBM team would want to get in a car with Deep Blue driving. I don't think so. I don't think so. But one of the beautiful parts about Deep Blue's calculation, its unconventional approach to chess is because the computer considers every possibility, it also considers the paradoxical possibility.

GK MOVE: 16 Nf3

DB MOVE: 16...exd4

MAURICE ASHLEY: We're anticipating Kasparov to play Nxd4 because the other response would give him an isolated pawn and that's considered in the textbooks as a weakness, that Kasparov would rather not have. So we're respecting him to play Nxd4, but he is considering the move. We look at the clock time, Kasparov has used one hour -- he has left one hour and 26 minutes, I believe it says, while Deep Blue has over an hour and a half. There's one hour and 25 minutes now for his first 16 moves. Remember, the time control is 40 moves in two hours, 40 moves in 120 minutes. You can use your time as you wish. You can distribute that time, you can spend 20 minutes on one move, if you like. You can spend ten seconds on another move. And so Kasparov will judge exactly what he needs to spend on each move depending on how critical that situation is in the game. Deep Blue we've seen regularly just /TKREPL out moves, less than three minutes, just about every time. It's been rare that it's spent more than four minutes on a move, and we've noted those times. It's like almost with shock when you see Deep Blue spending that kind of time because for a program to not think that long except in very, very dicey situations. We do have another question from the audience. 67899 gentleman -- the gentleman over there --

AUDIENCE MEMBER: The gentleman over there asked if Deep Blue searched every possibility, and even the best programs don't search every possibility. Even Deep Blue, who is 200 million moves a second, if it were to search to a depth of 12 ply, would take it like 2,000 years for it to actually calculate every possible move to a depth of 12. I'm sure there's people who know more about this than I do --

YASSER SEIRAWAN: They're going to be on stage soon.

AUDIENCE MEMBER: Probably. But in general --

MAURICE ASHLEY: In five minutes actually.

AUDIENCE MEMBER: -- very simply, there is a search algorithm called alpha-beta search which basically does prune off certain branches that are not as good as others, if there's -- Deep Blue -- well, not Deep Blue necessarily, but an alpha-beta search would look down at sort of the ending of moves after it's generated everything and would throw out the ones that are really stupid and wouldn't be used for anything, which of course could lead to the possibility that there's like checkmate next turn, which it wouldn't necessarily see if it was at the very lowest level. But Deep Blue also can use selective deepening where it would go down further in the tree and sort of expand out like the gentleman was saying about Fritz going down like 32 ply for certain things, just to clarify that.

MAURICE ASHLEY: Thank you, we have a very knowledgeable awed /SKPWREPBS we take advantage of the vast store of knowledge.

YASSER SEIRAWAN: Absolutely. And the idea of what the gentleman was saying was that if Deep Blue kept looking at every possibility and every counter possibility and every counter possibility and so on, very quickly you have an exponential explosion of numbers, and then it just gets overwhelmed so that at some point it starts pruning its tree and stops looking at too much garbage so that it can look deeper. But one of IBM's computer scientists I believe is going to be joining us and giving us some insight to that kind of thing.

MAURICE ASHLEY: If you look for example at this position that Kasparov is in right now, Kasparov is thinking about this situation. But how many moves can Kasparov possibly be considering right now in this position? All of two, I would say.

YASSER SEIRAWAN: Indeed. He either has two possibilities.

MAURICE ASHLEY: Now, we say only two possibilities. Deep Blue would only come in and say no, he doesn't he has about 45 and after that it's about 200,000 and after that he has another hundred million, right? (Audience laughter.) But we organize "No, he has only two, and that's all he's looking at right now is what comes after this move, these two moves. Deep Blue will look at all the other junk. Deep Blue will look at random moves from a3, a4, okay, and all the possibilities that stem from that, b2, b4, all the possibilities that stem from that, even crazy queen sacrifices, Qxe7. It sees everything and just looks and says, "Maybe that works, who knows, and it keeps looking. So you've got to understand that although it has amazing calculating power, it looks at a lot of junk with that power. 99 percent of the stuff just wouldn't occur to us, couldn't come to our minds, eeb. I know it's astonishing, chess Grandmasters, whenever a decent move says what about this move? How quickly the Grandmaster says, "I didn't even see that move, and you know why? Because it's bad."

YASSER SEIRAWAN: Exactly. It's true. That's why you have many talented Grandmasters who understand chess and it's very natural to them. But it is a very, very difficult. It also sees a very, very great deal of hidden beauty, and some of the computer's ideas are -- some of the computer's ideas are quite attractive. And I'd like to get a question from the left.

AUDIENCE MEMBER: I don't know if it's Hal usenating or if something was meant to be this way but I think on the side of the table over in the room it says Garry Kasparov with two R's, and I'd just like to point that out.

YASSER SEIRAWAN: We're going to fire the stage hand immediately after this question was asked. The young gentleman had pointed out that Garry Kasparov's table, his name is misspelled.

AUDIENCE MEMBER: He does have two R's.

YASSER SEIRAWAN: "Garry." I thought you said "Kasparov" had two. No, Garry does spell his name with two R's. Phonetically, it's a Russian name so when it goes from the Cyrillic to our alphabet, it /HARZ the Garrrrry," or "Garry," --

MAURICE ASHLEY: Either with, Garry with two R's, or Garry with one R, it's Garry.

YASSER SEIRAWAN: We won't worry about that too much. But we will wonder, Yaz, in this position, why is he thinking so long about this move? I mean Nxd4 is very, very natural, and cxd4 is a little ugly. Not necessarily a bad move, but it is a little ugly. Why is Garry thinking so long about this option?

YASSER SEIRAWAN: Well, this is one of the things that Grandmasters try to do in top-notch, tournament level chess. And this is just kind of a carryover from the way Garry plays against human beings. He has an obvious move. It's obvious he's planning Nxd4. But rather than simply play the move, what he would do if I was his opponent is try to come as far as he can, calculate as far as he can and try to set up an ambush, or to trick me down the line. And when he's playing the computer and there's a forced capture like this there's no reason not to make the capture quickly because it's going to be very hard for the human to predict the human's moves. He could not have predicted Bb5, could not have predicted Bb4+. He just says these are a bad move, they're not a part of my consciousness, I'm not going to worry about them. So in this moment he's trying to understand the position as much as he can so he can set up a tactical shot /AOR an -- or an ambush down the road.

MAURICE ASHLEY: He doesn't look like he's trying to understand anything, though. Look at the head and the hands in this posture.

YASSER SEIRAWAN: He's suffering.

MAURICE ASHLEY: He doesn't look as if he's too happy about the situation.

YASSER SEIRAWAN: We have a question from the audience. Yes, sir?

Real-time text commentary is made possible by LiveNote, Inc. and Vincent Varallo Associates