Commentary for black move 16:

MURRAY CAMPBELL: Actually this is not the kind of position that would show off Deep Blue's calculating skill. It has taken -- in open positions, it can do some truly amazing things.

MAURICE ASHLEY: Any other questions from the audience? AUDIENCE MEMBER: I wanted to ask Mr. Campbell if he could discuss with us briefly the strategy that your team is using as far as time management. The computer's time management. We noticed particularly in game one that the computer seemed to move quite rapidly. If you guys were actually doing that intentionally to try to speed the game up somewhat and actually give Kasparov less time. DB MOVE: 16...Kh8.

MAURICE ASHLEY: Before you answer that, we should say that Deep Blue has played a very interesting move, the move from G -- the king from g8 to h8. And Yasser Seirawan is making laughing faces over there.

YASSER SEIRAWAN: I apologize, Murray. "Sorry, IBM!"

MAURICE ASHLEY: I guess you don't like this move?

YASSER SEIRAWAN: No, I love this move, because again I had made a prediction. My apologies. I thought that before the match Garry was a clear favorite. I mean I think he's a formidable, phenomenal player, and the first match he had learned a lot, I knew that he was going to be very well prepared. He was taking the match seriously, and I felt that meant death to Deep Blue. But I had also said in my prediction that the moment Deep Blue plays a move, a "null" move, a pass, if you will, the move Kg8-h8, the strategy is working. And this is exactly the kind of thing that Garry just loves to see, because this is a passive waiting move that says, "Hey, I don't know what I'm supposed to do, so I'll get my king out of there.

MAURICE ASHLEY: Well, let me defend the other side, play a little Deep Blue advocate here. The strategy that was mentioned earlier by Carl Hessle /R*ERBGS, I believe it was, the idea of bringing the bishop to g5, maybe will be is thinking of repositioning the knight to the g8 square and playing the bishop to g5 and felt that this was the most harmonious way of doing it. Yaz?

YASSER SEIRAWAN: I don't believe it, but it's a nice idea.

MAURICE ASHLEY: Well, we will --

YASSER SEIRAWAN: But it's a good point. Maybe the move Kg8 with the -- Kh8 with the idea Ng8 and G B -- Bg5 is on its table of possibilities, and we'll see. I feel that with my very limited computer experience, that is, playing against chess computers, normally speaking they make such moves just because they don't know, darn it, what they're supposed to do.

MAURICE ASHLEY: Murray ^ , is that ^ ; is that correct your feeling?

MURRAY CAMPBELL: Identifies that -- I've seen that plenty of times that problem. I can't predict until we see a couple more moves if Deep Blue intended that or not, if there was a point to it or if it's just waiting. Even in positions where it gets to that stage, where it's just waiting, it requires the opponent to do something, and eventually at some point the position is going to open up and become tactical. Now, if the position is overwhelming at that point, it's too late. But if there's still some counter play, still some chances for Deep Blue, it can often find some unforeseen and amazing defenses that can hold a seemingly bad position, hold it together.

MAURICE ASHLEY: Surprisingly, Kasparov does not look too happy, although -- and now he's even shaking his head a little bit, which we can only guess at what's going on in the genius's mind. But he doesn't seem ecstatic about the situation, he doesn't look like he's licking his chops to make the next move. We will see shortly what Kasparov does. To follow your point, Murray, in the situation, just the waiting move, where it makes waiting moves, what has been the nature of those waitings moves? Would it be a king move like this --

YASSER SEIRAWAN: Remember game one, Bd6-c7, that kind of a waiting move.

MAURICE ASHLEY: What do you see as a pattern?

MAURICE ASHLEY: Often a king move will be it, because you can often move your king without doing any damage fo your position -- to your position. So that's a typical one. There may be others. Rook moves, too, from one closed file to another closed file. As Deep Blue has gotten stronger and stronger we see less and less of those, but when you're playing a player who has spent the past few weeks preparing specifically to create situations where computers will have difficulties, it's quite possible that we're seeing one in front of our eyes right now.

MAURICE ASHLEY: Before we continue, we welcome to the stage International Master Mike Valvo. And I will take my leave.

YASSER SEIRAWAN: A nice round of applause for IM Maurice Ashley. Come back with lots of analysis! Excuse me, Murray. We've got some news, do we?

MIKE VALVO: I found out some interesting things in my travels.

YASSER SEIRAWAN: Okay.

MIKE VALVO: One that you should know, actually. The game between the winner of Aegon and MChess Pro followed this one very similarly.

YASSER SEIRAWAN: Yonakosh -- you're talking about the recent human vs. computer event that I played in The Hague just before this match --

MIKE VALVO: That one.

YASSER SEIRAWAN: Yonakosh hashvili from Israel was the winner, he played one of the strongest programs in the turn, and he defeated it in very grand style, Yona was very proud of this victory. What we call the Botvinnik pawn /TPORPLS accident -- formation, the triangle of pawns on e4, d3, and c4, and he won a very beautiful game.

MIKE VALVO: And I also found out that way back in 1978 when David Levy played chess 4.7, that David employed exactly this strategy.

YASSER SEIRAWAN: Again --

MIKE VALVO: a3, d3, c4, and he was successful -- well, he won everything in those days against the machines, but that was his strategy, to do exactly what was done here. So there is some precedent. There's a very strong likelihood that Kasparov was aware of these games, especially the one in Aegon, and just emulated the play. Are you aware of these games, and did you consider them?

MURRAY CAMPBELL: We had seen the Aegon games, and I know that the -- the Israeli player you referred to. "Y -- Yona.

MURRAY CAMPBELL: I've used this pawn formation several times, I think two or three times in that tournament and was very successful with it so we did notice that.

MIKE VALVO: Did you do anything about that, what you noticed?

MURRAY CAMPBELL: We've done some preparation for closed positions of this sort. We're about to be tested.

YASSER SEIRAWAN: A reasonable attempt. Question, Carl again, if you will? AUDIENCE MEMBER: One thing from last night is Mr. Campbell talked about how in open positions it excels, and one of the things we talked about earlier is why the computer didn't play d5 right out of the opening and an effort to open up a file and open up the center a little bit, about what was in its opening book. If it favors open positions why didn't it naturally gravitate toward that?

YASSER SEIRAWAN: Did everybody capture that question, is it clear what he was asking? What he asked was in the opening there was the possibility of d7-d5 by the computer. That would have created the open positions that it excels, on why not?

MURRAY CAMPBELL: Well, it seems like a very simple question to ask why it plays a move or not plays a move. In fact, the system searches through many billions of possibilities before it makes its move decision, and to actually figure out exactly why it made its move is impossible. It takes forever. You can look at various lines and get some ideas, but you can never know for sure exactly why it did what it did. I really don't know. It was very close to playing d5, as I recall. It was a toss-up between d5 and d6, and I think it was very close.

MIKE VALVO: I have an interesting question to ask you, Murray. Over the years, I have been asked as a tournament director, how can computers play better than their programmers? Because there's the old saying that you put garbage in, you get bar gaj -- garbage out. How do you explain that?

YASSER SEIRAWAN: That's no reflection on you, Murray, don't take that personally. (Audience laughter.)

MURRAY CAMPBELL: In fact it seems paradoxical, but it's not at all. If I was programming Deep Blue to play chess the way I play chess, the human style of play, I could imagine that it might not play much better than me. It might have certain advantages. It wouldn't lose its concentration or so on. And wouldn't get tired. But basically it wouldn't be able to do much better than I could. But Deep Blue takes a completely different approach, and that's what we're -- was the goal of the project when it began at IBM Research was to create a very powerful computational engine that could search through millions or billions of possibilities and come up with a chess move in a completely different style than a human. So you can put a small amount of chess knowledge into a system like Deep Blue and very strong play will come out without it having to understand as much about chess as a Grandmaster understands about chess.

MIKE VALVO: Can we use the word "understand" when we talk about computers? How does a computer understand a position?

MURRAY CAMPBELL: No, we can't use the word "understand." It evaluates and chooses the one that leads to the best evaluation, is all I can really say. Reducing the position to a number is not ideal. There's so much more to a chess position than a single number. You really have to compare positions on various dimensions and make a choice based on very sophisticated decisions, and people have tried to create chess programs that mimic human reasoning styles, and in general, although there's been some interesting results, they can't play chess very well at all, and the only good chess playing programs have come out of the -- what we'll call the brute-force approach, the alpha-beta search as we call it, and that's searching through all the possibilities as deeply as you can, evaluate it, and choosing the move that leads to the best position. Now, it's much, much more sophisticated nowadays than it was when this approach became common in the early seventies, and the programs are much stronger both in terms of knowledge and in terms of the way they carry out the search, but, in essence, that's what they do.

MAURICE ASHLEY: Murray --

YASSER SEIRAWAN: Murray, we'll have one or two more questions from the audience, but I've got to ask this. You see, I watched the movie "Terminator," and what I want to ask is this. Do you think computers will become aware one day? You've just said maybe Deep Blue doesn't even know that it's playing chess. Do you fantasize or do you ever sit there and say to yourself 20 years, 50 years, a hundred years or is this ever going to happen? I've always want to ask that question to a computer scientist. Am I going to be around when this happens, you know?

MURRAY CAMPBELL: Every scientist who's thought about this has a different opinion and I have mine, and that is that eventually it will happen but it's not going to be for many decades that we'll have a computer system that's capable of intelligence, the same breadth and scope that human intelligence has. So I think that's quite a long ways off.

YASSER SEIRAWAN: And we do have a question. AUDIENCE MEMBER: What's the future for Deep Blue and its team? Is it to just beat the world champion, to develop commercial applications, to further refine the product?

MURRAY CAMPBELL: Well, the goal of the project has been to investigate the combination of a parallel supercomputer like the IBM RS/6000 SP, together with special purpose hardware to create this very powerful computer. And we believe we can apply the approach that we've taken to other problems and that are perhaps more practical than chess.

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