What you're reading is exactly what Maurice and Yasser said at the match - thanks to the fine work of the stenographers at the scene. Please forgive them a typo or two and enjoy the commentary!!
Commentary by Maurice Ashley and Yasser Seirawan
Game 4. February 14, 1996
Mr. Newborn Good afternoon. I'd like to welcome everybody to round 4 of the ACM computer chess challenge between Garry Kasparov and IBM's DEEP BLUE. We've played 3 games. The match is scored at one and a half to one and a half. And the next game is due to begin at 3 o'clock sharp. I have two quick questions I'd like to ask the audience. How many are here for the first time today?
We have a lot of new people today. Three-fourths of you haven't been here before. Well, this is possibly the most exciting sporting event going on in the United States right now. For that purpose we have the two best anchor commentators in the United States to cover this event for us. They've been doing a splendid job and the audience has had a terrific time and they bring the games to life like the Dallas Cowboys and I guess the Pittsburgh Pirates.
Mr. Seirawan Nice try.
Mr. Ashley You can tell you're from Canada. Steelers.
Mr. Newborn I'd like to first introduce Maurice Ashley who is an international master from New York City. And Maurice is one of the strongest chess players in the United States and has been a commentator on chess and computers for a while. We're delighted to have him with us.
(Applause.)
Mr. Newborn And his partner for the afternoon we have Yasser Seirawan who is a former U.S. open champion 3 times, as late as 1991. And we're delighted to have Yasser to also bring the games to life. Between the two of them you're going to have a great afternoon.
(Applause.)
Mr. Newborn I'd like to wish all of you a good time. The two of them are in charge, but they'll possibly bring a few people in here from time to time and the audience is free to ask questions. There will be lots of opportunities to ask questions. And I wish everybody a good time.
(Applause.) 1. Nf3 d5.
2.d4 c6.
Mr. Ashley Welcome to the ACM chess challenge. I am Maurice Ashley along with grandmaster Yasser Seirawan. The score is tied at one and a half, as you all no doubt know. Grandmaster and world champion Garry Kasparov is at the table now bringing himself together it seems, collecting his thoughts as his clock is indeed been punched to begin the game.
He took 19 seconds reflecting over his last move before playing knight to F3 to start the game. In fact he started with knight to F3 in game two and DEEP BLUE responded immediately with D5 which is what it also did in game two and Kasparov played D4 and now DEEP BLUE has played C6.
Mr. Seirawan The Slav defense. 3. c4 e6.
Mr. Ashley On move two as black. So we have a knight F3, D5, D4, C6, which is not what it played the last time, and then Kasparov instantly responded with C4 and DEEP BLUE with this triangle type formation, Yas, played E6 on move 3, E6. And it seems as if the DEEP BLUE team has prepared with a new setup versus Kasparov's opening moves. The last time Kasparov got something of an advantage.
Mr. Seirawan Well, clearly in game two Garry did it. I mean he had a very nice opening. He gained the two bishops. He ground the computer down. It was a model victory.
In fact I heard the expression an immortal modern victory. And today we're seeing that DEEP BLUE's team has started with a new opening. This opening formation is known as the Slav defense. Sometimes we see this move F5 and that move makes it a Stonewall Dutch.
The intent of the Slav defense is to take a stronghold in the center, especially this pawn D5 being supported at the moment by 3 black units, normally we also will seal the move knight F6 supporting the pawn on D5.
So what DEEP BLUE is really trying to do is just take a very nice stronghold in the center and keep it.
Now, I must say that I find this a bit surprising. First of all, Garry Kasparov has played against the Slav defense with incredible success. His score against the Slav defense must be somewhere in the high 80 percentile bracket. So would I say just about as well or perhaps even better than anybody Garry knows what to do against the Slav defense.
Mr. Seirawan There is one extremely sharp line connected with the move knight C3. This is a sacrificial line. D takes C4, E4, B5. And the intent here by Black is to win the pawn on C4 and maintain it. Garry might have been a little bit surprised by the move order of DEEP BLUE, and so he's probably wondering if they have this -- I think it's called the Bloomberg -- well, it's it's no bloom variation. Very sharp tactical line. But I'm sure Garry would love to get into the main lines which go knight C3, new variation, yes, knight C3, knight F6, bishop G5.
Then we get into what is known as the Bodvinick gambit. And again Garry is a leading, leading expert in both of these openings. And he's changed --
4. Nd2
Mr. Ashley And he has completely changed tactics. Instead of going into the absolutely crazy lines of the Bodvinick variation. He has instead played a move that there is no way that DEEP BLUE team could have predicted, and that is the move knight B to D2, and clearly Garry Kasparov considered all of the possible preparation that the DEEP BLUE team might have had regarding all the correct main lines, the super sharp main lines, and after some thought, approximately 3 minutes Garry spent on the last move, he decided let me play something extremely quiet involving no sacrifices, very solid, all the sharp stuff, you know, you put out.
If you look and remember the game, game 1, Garry went sharp. Garry got killed. Simple straightforward. The computer is so adept at handling sharp complicated positions that even the best human in the world finds it extremely difficult to keep up, and Garry even gave the computer a rating of over 3,000 in sharp positions. In quiet positions sometimes Garry says the computer plays way below that level, even down as far as 2300. And if you consider Garry Kasparov has the highest rating in the world at 2775, then clearly in those kind of situations Garry Kasparov will win every time, as he did, for example, in game 2. So Garry Kasparov trying to lead the game into more quiet positional channels.
As a matter of fact, I said the key for Garry Kasparov to win this match is to play boring chess, and it seems as if that's exactly what he wants to do, Yas, make sure no monkey business, no crazy stuff happens in the games.
Mr. Seirawan Absolutely. And I was just going to say that the move knight B D2 is a clear example of Garry Kasparov edging away from the computer strengths. Garry is known in the chess world for never turning down a challenge and certainly a challenge like the slav defense he would go into the sharpest most sacrificial lines against a human competitor, and so you're seeing here in this just, the fourth move of the game, a clear adjustment by Garry. And mind you, I concur with this strategy by Garry for all the reasons you said, the calculative abilities of the computer are simply too damn impressive. So he's had to make an adjustment. We're seeing it here on move 4.
I'll just talk about the move knight B D2. We're very fortunate to have with us Frederick Freidel. In the data base of Fritz do we know if Garry Kasparov has played this before?
Mr. Seirawan All right. Frederick is just going to get a microphone so he can join us in the conversation.
Let me just address the points of the move knight B D2. In the first part, in the opening the thing that you have to do is develop your pieces. You want to bring them out to the most active squares. And the move knight B D2 has a clear drawback. This bishop on C1 is suddenly blocked in. It cannot move. So White will probably have to play a move like B2 B3 in order to fianchetto the bishop and get it out. So the move knight B D2 blocks the bishop kind of happily perfecting development. However, in its favor, the knight protects the pawn on C4, and it also supports the freeing central break E2 E4. So the move knight B D2, some pluses and minuses, and it's not considered the most active move in the position.
Mr. Ashley Wouldn't you say that if Garry Kasparov were playing a guy like Kramnik or Cherepkov, who are sharp specialists on the Black side of this position, that he would never play a move like knight B D2 because they wouldn't be afraid of that move. They'd say what's the deal? Why are you ducking all the fancy stuff in this line? Especially since everyone knows Garry Kasparov is a specialist in these positions on the White side. But against the computer this move knight B D Garry just played it calmly. This is what I'm going to do.
Mr. Seirawan It's a clear compromising of his style. Nonetheless, as far as a secondary move goes, it's not terrible. And again what you had pointed out earlier seemed that the DEEP BLUE team had it come prepared with something special in the openings. It's a very sharp line. Garry deducted the challenge.
4. ... Nf6.
Mr. Ashley Well, look at this. So far DEEP BLUE, after playing very quickly in the opening, practically 15 seconds, I believe, on each move in the opening, has spent over 5 minutes, and finally has decided to play the move knight to F6, which would seem like a very normal average-looking move. So clearly Garry Kasparov has taken the computer out of its book, clearly, and it has spent some time now over this surprising move for it knight B D2 playing pretty standard, knight to F6, nothing wrong with that move. Can't fault that last move by DEEP BLUE in any way.
Mr. Seirawan Absolutely. The move E2 E4 is now prevented because the knight on F6 controls the E4 square twice, along with the pawn on D5. So knight F6 would be the expected move. Now, again it's a development question. And Garry has to really take a deep look at this position and determine how he wants to develop his pieces. 1 of his favorite schemes of development that he's used to tremendous effect in his world championship matches is the fianchetto. He enjoys to play moves like G2 G3 in these types of positions or even B2 B3 with the idea of bringing his bishops out to the squares G2 and out to the squares B2, thereby kind of like making an incision in the center, controlling the center from afar. I wouldn't be surprised if we see a slow move like these. Another possibility for Garry would be to play the move queen C2. The intent there, once again, is to buttress that center push E2 E4. In general it's not a good idea in the openings to develop your queen early, so I'd bet that Garry is probably going to go for either a kingside or queenside fianchetto, maybe both.
Mr. Ashley How would you argue this, though, that, Garry Kasparov is being forced to play, so to speak, against himself? He really can't continue with the sharp lines that he's known to excel at. Garry Kasparov in a sharp position is a fish in water, no question. Garry Kasparov will come up with fantastic ideas, paradoxical ideas, sacrificial ideas that are so deep and profound in extremely complicated positions where the average person is trying to figure out what's going on here Garry and he'll just whip out the variations. I'll just do this this this this this this. But against the computer, such a discussion would be fruitless because the computer could show him all the moves possible, all the best moves possible in those kind of situations. So he's forced really to play anti-Garry Kasparov chess in a way if he's going to be successful against the computer.
Mr. Seirawan Yes and no. I agree that he has to back away from his big weapons, but the truth of the matter is Garry did not become the world chess champion by being one-dimentional. In fact, the problem that I found when I've played with Garry is that, gosh darn-it he does so many things so well that when I feel I've done something wonderful I've forced him to take up a defensive position, let's say, I force Garry on the defensive. I've got him now. Well he plays beautiful defensive chess. And then the same point that we were having discussions in the earlier games. He's so wonderful in the opening and middle game you forget that he can play the end game extraordinarily capably. And the 2 games that I lost to Garry were in the endings where he outplayed me beautifully in a king and pawn end game, for example.
So the reality is Garry is not a one-dimensional player and he can fall back on some of these strengths that are a little bit hidden or overshadowed by his wonderful attacking skills.
Mr. Ashley A very good point. And for the moment however he is thinking about how to continue the game, what channels to take the game into. He has several options at this point and every single option leads the game into different kinds of positions, which Garry will try to figure out what is the most favorable kind of position he can get as a human where his intuition, his experience, his strategical insight will play a bigger role while the calculative ability of the computer will be much more diminished. But he is spending sometime here in the opening moves.
The DEEP BLUE team did say that they hoped that this is where they would give him the most problems is in the opening moves because as the game goes along, DEEP BLUE plays so quickly that the pressure will be on Garry to keep up in the middle game.
Mr. Seirawan Absolutely. That's a key point. We're talking about some of Garry's strengths. Let's talk a little bit about the computer's. The computer absolutely marvelous calculative ability. But beyond that, that calculative ability allows it to make decisions extremely quickly. And we've heard some of the programming team saying hey that's the 1 spot where we know we've got an edge, which is our guy plays really quickly, very fast, and it will put a lot of pressure on Garry. If we can put Garry in time trouble we've got a real chance in this match. And that's something that Garry has noticed because sometimes we see DEEP BLUE playing instantly and other times, confusingly enough, you'll get into a situation where DEEP BLUE has a forced move. It's the only move in the position and it will suddenly take a couple minutes thought. I find that very interesting.
Mr. Ashley Well, what do you think about today's game though? How significant is the result of today's game? Win lose or draw? Clearly a win would have tremendous impact on the next 2 games.
Mr. Seirawan Absolutely. Victory from either side today impacts the course of the match dramatically and it sounds to me like we need a poll.
Mr. Ashley No, no. This is an early poll. We ought to wait a few minutes. Yasser Seirawan's poll, folks. He always likes to do 1 of these polls, get some audience participation.
Mr. Seirawan Since we got Garry on the first move, he's on serve so to speak as White. All of those who believe that Garry is going to come in today and beat the computer, raise your hands.
Yes, we got a lot of cheerleaders in the audience today.
All of those who think the computer is going to win.
Boo in the back, boo in the back.
And who thinks it's going to be a draw?
Well, we've got --
Mr. Ashley A couple of faint-hearted souls.
Mr. Seirawan We see a lot of blood.
Mr. Ashley People want to see Garry win today. 5. e3 Nbd7. 6. Bd3. That's interesting. We should point out it has been passed on to me, Yas, by an eagle-eyed member of the audience that Garry Kasparov has removed the watch. The watch is off. Garry has taken off his watch and people who have not seen that before may think what are they talking about? Whenever Garry Kasparov takes off his watch, it means time to get down to business. It doesn't mean that he's in any trouble. There's some other things that he does, and we will point that out when the time arises. But whenever he takes a watch off, it means you know I really have to think about this position and it's time to get to work and he has already done so very early in this game. In the meantime after some thought he played the move E3. Garry played the move 5 E3 and instantly DEEP BLUE responded with knight B to D7, to which Garry Kasparov also immediately replied bishop from F1 to D3. So the last couple of moves obviously after that nice little think by Garry Kasparov and another 4, 5 minute think on his part, he has decided how to develop. He has eschewed the development of his bishop on the G2 diagonal. The idea of fianchettoing instead of centralizing his bishop putting it on the square D3.
Mr. Seirawan Also a little bit of a surprise because the moves E3 and bishop D3 are perfectly natural moves. Garry is preparing to break E3 E4 to get more of a control in the center. However, the drawbacks to these types of positions because literally they are some of the most attractive positions in the game of chess today, a lot of the international grandmasters favor both sides. They enjoy playing both sides of the position. These types of positions have an opportunity somewhere notice middle game of just blowing up, literally blowing up and you see a lot of trades, lots of forced exchanges and I think that that's a type of a game that the computer plays quite well. Mr. Seirawan And that's why I thought the slower G2, G3 building a house, protecting the king might have been something we would have seen by Garry. So Garry's pause by the way I think was very significant, even though it was openly 6 minutes, I really think he was thinking very, very deeply, not only about what kind of patterns, formations he wanted to set up for himself but also just what kind of a struggle he wanted, whether he wanted to delay the struggle into the late middle game or to have an early promising central game again based on the move E3 E4. So he made his choice. Deep Blue moves 6. ... Bd6.
Kasparov responds 7. e4.
Mr. Ashley Indeed. Remember from game one that fateful game for Kasparov losing this game in the kind of position where he really didn't have to lose. He could have played much more solidly. He could have defended the position, kept strong trumps in the situation but instead of doing that he went over in Kasparov type style on to the attack and after the game after he was crushed by the computer, refuting the attack because it simply did not work, Kasparov says you know it's funny if you make -- if you go on the attack like that against grandmasters I would have won. Against the average grandmaster I would have won the game. But the computer doesn't believe in ghosts. It says does it work or doesn't it work?
And once it's calculated everything to the last possibility within a certain range it says I don't see it. What do you have? And it just goes right for whatever it feels like going for. And we saw this unbelievable move in game one where the computer ripped off a pawn on the left side of the board when it seemed as if clouds were gathering around its king, and cool calm defense by the computer completely refuted Kasparov's attack.
So this is a very good point you're making. If Kasparov is going to go for the king move and open up the game he's going to go in the strengths of the computer and it could be testy.
Well, we, as I speak are now seeing this happen. The computer responded to Kasparov's last move with bishop to D6 and now Kasparov in -- with what seems to me like a very brief pause is blasting open the central position with the move E4, the move that you mentioned.
His king is still in the middle, Yas. Is this wise? Isn't this the kind of position the computer is going to think, yeah, now we get to mix it up, baby?
Mr. Seirawan Well, one thing -- baby. One thing is for sure. The move E4 has to make the programmers of DEEP BLUE's team quite happy because what it causes is an immediate eruption in the center where exchanges take place. And every pawn that is exchanged, the position becomes a little bit more open. There are many strategical themes in chess.
One of the most important strategical themes is what we call the open position. That's where you have exchanges of two pawns or more. Then the diagonals, the ranks, the files are all opened up. And that's really a strength of the computer in open positions it can calculate extremely well. However, in semi-closed or closed positions, closed positions are those positions where all 8 pawns for both sides remain on the board. There you see a lot of exam perfecting and maneuvering as people try to position their pieces in the best possible places. And then you have your semi-open positions where you have an exchange of a pawn or more. The more the position opens up I think it's in the favor of the computer because then the computer can neutralize Garry's early advantage.
However, Garry probably has something very specific in mind when he played this move. We're likely to see the move D takes E4 now. Just try to run a little quick variation across the board. We'll probably see D takes E4, knight takes E4, knight takes E4, bishop takes E4, castles by black. So the computer's king will be nice and safe. Castles by white, and then there's an old trick that goes E5, E5 for black, the idea being to open up the center the then you'll see D takes E5, knight takes E5, knight takes E5, bishop takes E5, bishop takes H7 check, king takes H7, queen H5 check, king G8, queen takes E5 snaring a pawn. It's very interesting. This is an old trick that works for black when the pawn is on A3. Because of the opposite color bishops. But this is -- in this particular variation white wins a pawn full out. And now we're seeing the moves as we're expecting them. 7. ... dxe4
8. Ndxe4
Mr. Ashley So in fact black did play D takes E4 and afterwards instantly Kasparov captured with his knight on E4, knight from D2 takes E4. So the position suddenly down somewhat. I think DEEP BLUE will avoid that little tactical trap you mentioned, Yas. You know, I think it's --
Mr. Seirawan Good chance it is will.
Mr. Ashley I think it sees things like that. I just have a feeling. But at the moment the position -- okay. The position is opening up a little bit. But maybe we shouldn't play that song too much. After all white does have a good center and pretty decent development.
Mr. Seirawan Well, it goes back to something I mentioned earlier. The move knight B D2 did rob the bishop on C1 of its activity. So the move E3 followed by E4 certainly opened up the bishop on C1 allowing Garry to activate the bishop. In the position at hand DEEP BLUE's bishop on D6 is under attack. That's why the move knight takes E4 is pretty much forced. The black bishop could go to B4 with check in this position, but the problem after that is that after the simple move bishop D2 white will have a very comfortable advantage in space and also a slight lead in development. So it's likely that we'll see the exchange of nights on E4.
Let's just take a look at that if we will. Knight takes E4, bishop takes E4. Again castling for both sides. And we could say pretty much after this point that the opening battle is resolved. And now we're going to get into the middle game battle. And in this middle game battle Garry will have an advantage which Fritz 4 point 0 concurs. And that advantage is based upon space.
Space is an extremely important element in chess. Most important elements are material. We all look at the material first. I'm a pawn up, I'm a pawn down, I'm a rook ahead. I'm winning. Material is very important. The second is development or time in chess. You get your pieces out as quickly as you can. The third is the structure of the pawns, pawn structure, are my pawns weak or strong. And the fourth is space. What I mean is space, when you start the game of chess each side has 16 pieces. 64 square board. But you would say that white has 32 squares. His first rectangle of the board and black has his 32 squares. If I control more space than my opponent, I'm going to have greater mobility. 8. Nxe4.
And we'll see now -- we haven't -- DEEP BLUE has played the move knight takes E4. I'm sure as soon as Garry returns to the board he'll recapture the knight.
Mr. Ashley We should make that clear to the audience not in the theater with us is that Kasparov often leaves the boardment on this occasion he did.
Mr. Ashley And usually when he leaves the board is because he feels extremely comfortable with the position. He does have a dressing room in back he can go to that has a TV monitor like the 1 we have here and he can see if there are any changes on the board and he will promptly come back as he's done and he's back at the table now. As you were saying. Kasparov plays 9. Bxe4.
Mr. Seirawan And Garry immediately, as soon as he did write down DEEP BLUE's move, played bishop takes E4. So we now have a position on the board which has opened up a bit but it's opened up to White's advantage. Keep in mind that the pawns on D4 and C4 control 4 squares of Black's, E5, D5, C5 and B5. Those 4 squares right across the center there, those 2 White pawns are in good position. The bishop on E4 is an excellent bishop. It controls that diagonal B1 H7 as well as the diagonal H1 A8. So White has better mobility, controls more space and I think has emerged with a slight but annoying advantage for DEEP BLUE's parts. He has a very solid, if compact structure.
Mr. Ashley In the meantime we do see that it is much, much more easy for White to develop his pieces considering the open nature of the game on his side of the board than it is for Black to do so particularly. Yas, there is some concern, as often happens in these positions for Black's light squared bishop, that minor piece on C8, is blocked in not just by the knight. The knight, of course, can move very easily and get out of its way, but much more durably so by the pawns on B7, C6 and E6 which prevent that bishop from just jumping into the game, which as you so ably pointed out, you have to develop your pieces in the opening. If you don't develop your pieces, the guy will just overrun your position and come at you with all guns blazing and your pieces will be asleep on the back rank. Right here for Garry Kasparov it's very easy to envision some moves like castles, rook to E1 or the queen going to E2 C2. It seems as if Garry Kasparov has met flexible options in this position and they're easy moves. What will happen at the end of it we don't know. But his moves are very easy. Deep Blue plays 9. ... O-O. Kasparov responds 10. 0-0.
Mr. Seirawan And again that flexibility comes from the fact that Garry Kasparov has more space. His pieces can go to more squares. We have seen the move castles by DEEP BLUE, and I'm sure Garry will do likewise, which he also has done very, very quickly. So they've now reached the position that we've had on screen for some time and this is where I recommended the -- not recommended, but pointed out the old trap E5. Because of what you were saying, Maurice, the bishop on C1 literally -- pardon me, the bishop on C8 is literally a prisoner of its own structure. So that was why the move E5 has been played in the past in positions to open up that bishop. But in chess terms, at least in Seattle we used to call such bishops Septilicus, to talk about a bishop that is septic, that is impotent. So this septic bishop on C8, what we try to do is play a move like knight F6, for example, in this position, attacking the Bishop on E4. Now, if we can just see that on the board, Frederick. Okay. So now the bishop on E4 being under attack, giving White the choice of bishop G5 or moving the light squared bishop. Assuming that the light squared bishop pulls back a bit, say bishop to C2, then we would see a move like B7 B6 and the idea again being to bring that bishop on C8 into the game by playing bishop to B7 or bishop to A6 or bishop to B7, followed up by the moves C5. All right. But in any case it's clear that Black -- DEEP BLUE has a problem it's going to have to solve, what it's going to do with the bishop on C8.
Mr. Ashley The Septilicus bishop. I don't know how Nimsovitch would have felt about a term like Septilicus, but a very interesting term. I don't think I'm going to be able to forget that term when I have 1 of those bishops. But indeed the bishop is locked in on C8 for the moment, and that's going to be a concrete problem for DEEP BLUE to solve. And knight F6 is certainly not a bad way to do it. That bishop on E4. On the other hand White's bishop looks very potent. Nice square to be on. So Garry Kasparov I would think feels pleased about this situation he has on the board. He can't complain. He's in no danger. His king is safe. The tactics that we talked about certainly are not present in this position, the position is not opening up with sharp crazy play. I think it's a position Garry Kasparov has under control.
Mr. Seirawan Indeed. And again I think it's a type of position he wants where it's up to DEEP BLUE to figure out how to neutralize his advantage. And that's the best kind of position to have, where you're on the sunny side of the torture and it's up to your opponent to neutralize the advantage. 10. ... h6.
Mr. Ashley Look at this last move. Very interesting I'll say. DEEP BLUE has played the move H6, which avoids a few things in the position. It avoids a piece jumping on to G5. You mentioned for example that a bishop might end up on G5. Potentially a knight could also do so. It also avoids this tactical shot that you talked about, Yas, the idea of bishop takes H7 against the E5 move, but it looks highly suspicious to my eye that in a situation where 1 ought to be developing pieces or thinking about what's happening on the left side, for example, we've been really focusing on that bishop on C8, that suddenly we find this weakening move around Black's king, the move H6. It has its point. How tremendously weakening it will turn out to be is not yet clear. But certainly a loss of time on some level. This move H6 develops no pieces whatsoever and Garry seemed to somehow leap up in his chair a bit and just lean over the board. His eyes now trained on that king to see if -- a frown on his face as if to say no, not that much. I mean move. That seems a move if a human made it against Garry made in how many moves? I'll be out of here how soon?
Kasparov moves 11. Bc2.
And look at this. Before we could speak on it Garry Kasparov has instantly responded with the retreat bishop to C2 setting up the old okey-dokey, maybe bringing his queen to D3 and threatening mate on H7. And that's something Garry Kasparov did instantly. Now, kind of a basic attacking idea. But clearly Garry Kasparov is pleased with seeing the last move H6 and this diagonal he's targeting instantly. No other move his mind but attack attack attack. I think Garry Kasparov is now like a bull that sees red. And he just wants to know what the computer is going to do about this attacking, this simple brutal as somebody called it, the Fred Flintstone school of chess. Just cave man killing style.
Mr. Seirawan That's a thing that you really got to love about Garry. I mean most bulls need a red flag to get excited. Garry doesn't need anything like that. He just needs to see a single opening and he'll jump right on it.
Just as you pointed out the retreat bishop C2 although it cost White a tempo, what he is intending to do is to bring his queen to D3 to create that bishop and queen battery and of course it's a caveman threat. Deep Blue responds 11. e5.
Mr. Ashley And instantly DEEP BLUE, just after about a minute's reflection DEEP BLUE, with no development to speak of, has decided to open up the position, explode in the center of the board with the move E5 and boy, you know, if I were playing Black I'd be scared in a position like that. It's almost anti-positional. You know, but the thing is with DEEP BLUE, DEEP BLUE knows no fear. So basically DEEP BLUE is saying either it works, either you make me or you don't. And we've seen when Garry Kasparov tried to attack before when there was no mate there was no mate. DEEP BLUE just refuted the attack and Garry Kasparov now has got to be thinking this thing doesn't know anything about basic prince. But at the same time it sees easily 6, 7 moves deep.
Tremendously.
Mr. Ashley If he thinks there's a mate he's got to be doubting himself. Did they pull the plug on this thing or something? Are the processors not working?
Mr. Seirawan Some of the processors forgot to handshake one another.
It's very interesting, actually, I'd just like to go back just one move here because this move H6, I wanted to say at the turn of this century we had one of the founding world champions, if you will, an unbroken line of world champions. And the original champion which Wilhelm Steinatz said what we would refer to as Steinatzian principles. And one of his principle is don't unnecessarily weaken the pawn structure in front of your king. And this move H6 to the human mind would be hey you're violating one of the cardinal rules of chess. And the computer did it for a specific reason. He wanted to cover the square G5.
The second point is what the computer has now done which is E5. And another Steinatzian principle, do not go on to the attack when you're underdeveloped. And again we have the situation with this bishop on C8, this rook on A8, this knight on D7, DEEP BLUE is not fully developed and yet it's making a breakout in the center.
Obviously it's doing it for good positional reasons, which is to bring out this bishop on C8. But let's take a look at the Fred Flintstone.
Mr. Ashley Queen D3. What if Kasparov plays queen D3 and threatens mate on H7?
Mr. Seirawan Okay. Obviously what black would have to do is block the threat. But he couldn't block it with the move G6 because now that pawn on H6 has been tender iced and eliminated. So we're going to have to go back. After queen D3 I think there's only one response, F5. You couldn't defend with knight on F6 because that would leave -- would leave the pawn on E5 under protected. So queen D3 forces this weakening move F5, which --
Mr. Ashley Looks horrid.
Mr. Seirawan Looks a little suspicious, yes. I really dislike it. I mean I'm a player that enjoys having a safe kick. I figure that if I have a safe king I can make lots and lots of mistakes and I'm not going to lose the game. But after F5 there are a lot of weaknesses around black's king.
However, there is the followup threat of E4. So Garry is undoubtedly looking at a position like F5, C5, in order after the move E4 to escape with a check against the black king. So F4 C5. The bishop might drop back. And still it's hard to say if white is going to emerge out of this with an advantage. But in the long term black's king is quite weakened.
Hans, did you have your hand up?
(Audience commentary.) Mr. Seirawan Okay. Dr. Hans Berliner mentions in passing that instead of the move bishop C2 -- is that right, Hans?
(Audience commentary.)
Mr. Seirawan What Hans suggests is that in this position, in anticipation of the move queen D3, Garry Kasparov might play the move C5 immediately in order to attack Black's bishop, force the bishop to drop back to say save the square C7, and then follow it up with the move queen D3. Now creating the battery. And again we have pretty much the same position, a transposition, if you will. Black would probably be forced to play the move F5. And the question being then -- and after the move F5 we have a rather sharp position. Will Black's weaknesses on the kingside tell or will White find a way of developing his pieces very quickly to bring out an attack?
Mr. Ashley Yasser, I'm going to put you on the spot here.
Mr. Seirawan Again?
Mr. Ashley What's your gut feeling on Black allowing this kind of position to occur on the board?
Mr. Seirawan Well, the gut feeling is I'm sure the same as Garry's. Yum-yum, you know. The computer is definitely doing a no-no and I'm going to punish it. On the other hand, because the position is going to be opening up and is going to be very sharp type of position, the computer may be able to ward off Garry's threats before he can really get his shot in. And 1 of the things that I like to do when I have an advantage is not necessarily force it home. What I like to do is try to build it up a little bit more. And in looking at Garry's position, if we'll go back to the position at the board, 1 of the things that I have to say to myself is I don't like my bishop on C1. That move H6 was an annoying move because it stopped me from playing that move bishop G5. So I'll give the computer credit and I'll say I've got a bad bishop. That is to say my bishop on C1 is inactive. So let me try to do something to activate it. Let's give the following idea a shot. C5, as Hans Berliner would suggest. Let's see if Garry -- C5, bishop C7. And now the move B4, B2, B4. The idea behind that move is to support the advance pawn on C5 but far more importantly to bring that bishop to B2 because I know that with my bishop on B2 it will roar down that diagonal from A1 to H8. And look at the treasure on H8. So if we can, we have the position after B4, we might see a line, for example, like E takes D4 by Fritz -- by, pardon me, IBM. And now we see a move like bishop D2. So what will happen, this is a typical open position where lots of things can happen but you've got to like White's bishops. Right now they kind of sweep down the board. And this is a type of position that Garry can really bring home.
Mr. Ashley Well, we will see if the computer does have defensive possibilities against this what looks to be an impressive positional advantage for Garry Kasparov, although, as you said, it's not crystallized yet. What's interesting to me that I've realized over watching the computer it's very difficult -- chess players love to come up with systems and ideas and principles that you can use in many, many positions.
Steinetz of course was the initiator of the whole thing by coming up with so many principles, 1 you mentioned earlier about not weakening your position unnecessarily with the side where you're being attacked. Many of these principles about the 2 bishops and controlling central squares, all of these things help us to really play a game of chess. We follow those principles sometimes blindly, but hopefully elegantly and with a certain degree of specificity.
But the computer on the other hand doesn't really care too much about those principles. And we saw that in game 3, I thought, in a very big way. Instead of playing a natural move in game 3 the move C4, and instead played the move rook to C4. And this stunned us. Afterwards Garry Kasparov said it stunned him. It didn't occur to him at all. And he said of course a human being would have played the move C4 without much thought. C4 was so natural. And afterwards I gave it some thought and I said you know it's funny the computer doesn't care about our principles. What it cares about is what's happening in this position, what's going on right now. And in some sense the computer showed us a thing or 2 about how to defend a worse position of that type. Now we would love to bring a system of principles to explain how the computer defends and maybe we can use it in our own games.
But the problem is that it's so specific to the position that it's hard to make principles surrounding those moves. Garry Kasparov mentioned that if he had just 1 more move, if only his pawn were on H8 instead of H7 he would have been completely winning, but it so happened his pawn was on H7 that he wouldn't win the game. He was a tempo short. How do you bring a set of principles towards that way of defending?
In this position, for example, if the computer manages to defend we'll say wait a minute, now, we thought it was worse from all the principles that we understand, but in this given position again the thing was showing that it's still equal. It can still do what it did and survive. It seems to me very difficult to bring any kind of principles to this because everything for the computer depends on a specific position.
Mr. Seirawan And it's calculative skill. And last night I had a lot of trouble sleeping because of the previous game. And what happened in that previous game really shook me a great deal. And Garry said it very well. He said against any other player in the world I would have won. And I think he's absolutely right. He would have won. And it was only these very strange defensive moves, these awkward defensive moves, as Joel Benjamin described it, Joel Benjamin in the trainer for the openings in DEEP BLUE. He said when the computer makes a move it's sort of like a guy taking a 30-foot shot in basketball and you go no, no, no, no. And it's a swish and you go yes yes yes yes. And that's what the computer did. Yesterday it was making the moves that Joel was going no no no and they worked and they saved a game that I think anybody would have lost.
And yet on the other hand I was absolutely certain that Garry had a winning advantage despite the computer's inventiveness in saving the position. So I couldn't get to sleep and I was thinking about the positions, thinking about it more and more and I finally decided where Garry went wrong and that was the move rook A5 after rook B4. And I think had he played rook A6 he would have won the game. But it would have required him to understand this crazy move rook C4 was coming. And because he had no anticipation of seeing something as crazy as that, he didn't play the move rook A6. And I'm sure he made a deeper study of the position as well and I know he felt he had advantage, as I know he feels has an advantage now.
The computer has made this time wasting move H6 and he wants to take advantage of it. Should we turn and invite our audience to ask us some questions?
Mr. Ashley If you have any questions feel free to put your hand up.
Mr. Seirawan I'll just say at 4 o'clock in a short while we'll have some folk from IBM's DEEP BLUE team to explain some of the hardware. And I mean it is extraordinary what these engineering folks have done.
Mr. Ashley Joe Hoane of the team is planning to come at 4 which will be in a few minutes. So you will be able to pick their brains there. But we invite with you any questions now.
(Audience commentary.)
Mr. Seirawan The question was the move C5, although it does attack the bishop on D6, doesn't it leave the square D5 weak.
Mr. Ashley I know that Kasparov may have had similar thoughts to the same question because instead of going on the unbridled attack queen D3 he instead played the move rook to E1 which now carries on that 3 on two, those of you who are basketball fans you know 3 on two is usually pretty good and right now Kasparov has 3 things attacking the E5 pawn the knight on F3 the pawn on D3 and the rook on F1 with only two players defending the E5 pawn for black the bishop on D6 and the knight on D7.
So that is a definite threat to rip off a pawn for free on E5. So DEEP BLUE will have to respond to this move. Again though the move C5 you said may have some drawbacks.
Mr. Seirawan Let's go back to C5 for just a moment so I can answer the gentleman's question. After C5 this is a typical move what Bobby Fischer described. To get squares you got to give them. Okay. The move C5 attacks the bishop. It forces the response C7. But it does have the disadvantage of losing control of that D5 square. And the gentleman suggested that maybe in the future black could play knight F6 and bring the knight to the D5 square. And that's absolutely right. I'm willing to sacrifice control of the D5 square for the temporary advantage. But that's something that Garry thought better of. And therefore the move rook E1. And the move rook E1 I must say is quite in line with the nature of the position. Rooks lust after open files. They're like a cannon and they need an avenue of attack. They need an opening.
And the rook on E1 is beautifully poised because after rook E1 as you say you got your 3 on two, hit against the E5 pawn, and that open file, white would then control a lot of space. For example, if we see the move E takes D4 now, let's just take a quick look at this, queen takes D4, attacking that bishop on D6. A move like bishop C5 would be in the position.
And then after a move like queen D3, knight F6. This is the type of position that the player who gets his pieces out quickly is going to get an advantage. But it's quite drawn because of the nature of the structure. The structure of the pawns being 3 versus 3 on the queen side, 3 versus 3 on the king side, would you say that DEEP BLUE, if DEEP BLUE achieves this particular position has done a good job of neutralizeing the open advantage.
Mr. Ashley You got to admit that was such a natural variation you played and as you played it I'm thinking white's doing well, but knight F6, white is not doing anything.
Mr. Seirawan Yeah, it stopped.
Mr. Ashley And DEEP BLUE is in fact initiated its first move E takes D4. Unless Kasparov has something very specific in mind I'm concerned that he won't have any advantage when he allows the black knight to land on the F6 square. DEEP BLUE playing very quickly the knight on D7 was tide down to the pawn on E5, defending it. So when queen D3 came the knight couldn't go to F6 because E5 was under attack. Kasparov has played queen takes D4 in response.
Now, the fact that the knight can get to F6 quickly, not just on the next move, but even now makes me really wonder. I mean this variation surely Kasparov saw, very forcing line.
Mr. Seirawan Yes.
Mr. Ashley So he must have something planned and he must feel that he had some advantage in this position when he went into it, but the fact that black has solved the main problem of the bishop on C8, now a move like knight to F6 or as you pointed out the sequence bishop C5 then knight F6, what does Garry want? Is he playing for something completely different here? It looks as if that position -- I'm sure you have the feeling as you mentioned that you'd be okay here.
Mr. Seirawan Definitely.
Mr. Ashley What's the big danger?
Mr. Seirawan Again, if this type of position that we seal on the middle screen with the Fritz screen, I think Fritz has given Kasparov just a very minimal advantage of point 06.
So we would have to say that in this position white's neutralized although he has a slight, slight edge in development. As soon as that bishop on C8 whose diagonal is beautifully open as soon as that bishop on C8 develops well you cannot talk about any advantage whatsoever.
Mr. Ashley And we should just say the middle screen, the position you're discussing is now leave out.
Mr. Seirawan So maybe Garry has seen something a little bit more deeply. I will definitely give him credit for that. But if he doesn't come up with something very quickly his advantage of the white pieces are just going to disappear.
Mr. Ashley It seems to me as if his advantage has evaporated. This position here -- of course Kasparov has been very happy to get into end games with the computer. He has had the suspicion that in the ending he'll be able to out play the computer. But you know looking at this any kind of resulting queenless ending in this situation probably favors black even because that pawn on C4 can sometimes get in white's way unless white can get like a spatial bind on that left side of the board, that pawn on C4 the weak squares on the D line, okay, I can't say it favors black but certainly it doesn't seem like a kind of position that is bad for black and indeed DEEP BLUE has carried out this move bishop to C5 attacking Kasparov's queen, forcing it to move to some square.
We see many options for it, queen D3, queen E4, maybe even queen C3 in some weird situation. And in fact Fritz -- Fritz 4 the computer analyzing with us feels that maybe queen C3 -- yeah, queen C3 might be the best option with an idea like B4 and putting the bishop on B2. That's interesting. That's more positionly inclined than the other variation which seems to automatically give DEEP BLUE equality.
Mr. Seirawan Indeed. Queen C3 at least has the one advantage that after queen D3 the queen is going to be opposing black's queen. The draw back -- I didn't want to suggest queen C3 myself. I'm sure Fritz has an answer. But the draw back I feared was queen B6.
Mr. Ashley And Kasparov indeed has played queen to C3. And that move, that move is very interesting. And Kasparov playing very subtle chess, one might say in not immediately going after the weakened diagonal, the B1 to H7 diagonal where his light squared bishop is posted and that occurred to us very quickly as the way he would want to go.
But here Kasparov has played much more subtly and it seems his idea is to get rid of the dark squared bishops in some way or maybe to bring his bishop to B2. And this move, Yas, you said you had some concern about this move.
Mr. Seirawan Yeah. I had some concern about this. Obviously the move queen C3 is nice. What white wants to do is white has good development and he just wants to keep that good development and improve on it if he can. And again mentioning the fact that the bishop on C1 really wasn't in the game. The idea behind queen C3 is to try to play the move B4, attacking the bishop, forcing the bishop to retreat say to B6 and then to follow up with the move bishop B2 and the queen and bishop are beautifully placed on that diagonal for a nice battery. So B4 is the positional threat. The thing that I didn't like about queen C3 immediately I didn't -- was a little bit unsure of the tactics of the queen B6. Queen B6 attacking that pawn on F2 and in some cases the computer has that nice little securing idea of bishop B4 and the queen and the rook are on the same diagonal. I had assumed that bishop E3 would be pretty much forced. And then after an exchange of bishops, bishop takes E3, rook takes E3, knight F6. Well, a minimal advantage, but certainly an advantage for white.
Mr. Ashley Although Fritz believes that it's a huge advantage.
Mr. Seirawan Well, that's nice.
It's hard for me to believe that white has a huge advantage, but certainly development counts. And we can go back to the position at hand. If --
(Audience commentary.)
Mr. Seirawan Yes, back to the position at hand. Rook E8 would be a move that tries to neutralize the advantage. But again we would likely see a scenario like rook takes E8 check, queen takes E8, B4, attacking the bishop, the bishop would have to drop back, for example, bishop to E7 and then we might see -- let's see. Can I get him -- bishop B2, bishop F6. And typical Kasparov type sacrifice, queen to D3, queen to D3 sacrificing a bishop. That is little convoluted but I'm sure these are the ideas that Garry is looking at. Bishop takes B2, bishop taking B2, accepting the sacrifice, queen H7 check, forcing the response king F8 and now we play rook to E1, cutting off black's king. So this is obviously working out ideally for Garry.
Mr. Ashley Except queen H8 mate being -- you're just going to lose your queen.
Mr. Seirawan Exactly. So we have to go back to the position at hand. So this is what Kasparov is dreaming of, is setting up his bishops on C2 and B2 and then raking black's king side
Mr. Ashley This is indeed very subtle and seemingly very potent. This last variation by Garry Kasparov, the move queen to C3, no, we didn't like queen D3 and it looked very equal. And the variations started coming out quickly, and then we noticed this queen C3 move, and it certainly offers Garry Kasparov some reason to play for an advantage. I mean, it's not orthodox, but it's certainly consistent with controlling key squares in the middle and putting his pieces on very solid squares.
Mr. Seirawan Exactly. Well, the entire match has been unorthodox. This is an historical first, of course. There's been world championship matches in the past, but never 1 of man versus machine with classical time controls. I suppose in this position a fairly good neutralizing move might be queen F6 and again maybe we'll take a couple questions from our audience. Yes, we have a question in front.
(Audience commentary.)
Mr. Seirawan We'll go back. The gentleman asked the question that --
(Audience commentary.)
Mr. Seirawan The gentleman asked the question that in his view well Garry should really be aiming his queen at the king side. Maybe he should have tried the move queen E4.
Mr. Ashley Instead of the move queen E3.
Mr. Seirawan With the idea that after knight F6 to stay on the king side with the move queen H4. This is a very standard type of attacking pattern because in such circumstances there's the threat to play bishop takes H6, sacrificing a piece. And I think that was the absolute problem. Against a human player Garry might well have chosen this. But after what happened in game one Garry doesn't want to sacrifice anything.
Mr. Ashley Well, you know, I'll continue looking at this position. I've noticed something very specific and that's the thing about working out plans. You have to have some kind of specificity to it otherwise you won't end up succeeding. And this move knight to G4 after queen H4 threatens bishop F2 check, threatens to exchange queens. In fact might be a big advantage --
Mr. Seirawan A good neutralizeing move.
Mr. Ashley -- for black immediately. So this would be the reason why Kasparov would not want to do it specifically, specifically, as well as not want to go carry out just barbarous attacks. You know, that's really what we're seeing here.
I mean we saw, you know, Laurel Christenson still well respected. The Fred Flintstone school is not so subtle. You know, you can't just beat the guy up. You can't just go in there and say, well, I feel like checkmating you here I go. 14. ... a5.
And what Kasparov has shown here is his subtly in handling the attack. And instead of carrying out that other manuever, he has instead gone for something else which leads to interesting play. In this position here DEEP BLUE after some thought has decided on a pawn move. We saw a couple queen moves. We thought about it but it has instead decided on a restraining pawn move. It has played the move A7 to A5. A5 has been played. And that sets up a threat right now of bishop to B4 which would neutralize a lot of white's play. Because after bishop to B4 I think white would be force today play queen D3 and then knight F6 and then we're back in the line where white didn't want his queen on D3.
So that's a threat right now Kasparov has to deal with to move bishop to B4. I think if it's consistent with Garry's plan he might play a move like A3, but it's not the same. You don't get that B4 push the way you wanted to and get that space on the left side.
Mr. Seirawan With tempo. Let me just introduce to the public Frederick Freidel from Germany. Frederick has been working with computers I think most of his life. A very long life too I might add. Frederick, you had a comment that you'd like to share with the audience.
Mr. Freidel I was just amused because we're discussing all these variations and even before. And if you just look here and you can see Fritz after a few seconds came up with A5 which was in 5 seconds. So Fritz really understands DEEP BLUE quite well. This has happened a number of times while we've been discussing something else. Fritz is saying no, in one, this one, watch it.
Mr. Seirawan Well, Garry had described this as in game two. What he said is that he, Garry, has a Fritz program and other chess software programs that he uses in his own training and he clobbers them very frequently. But he also told us that he felt like he was beating up on the children of Big Daddy.
But still the children can point out the possibilities. And A5 again is a very good neutralizing move because if we can back just before A5 was played we do see that there was a very specific threatened position and that was white playing the move B2 B4. So DEEP BLUE's last move neutralizes this threat while introducing, as Maurice has stated, the threat of its own.
Now, we'll see the move A5 as played by the computer and now the computer's threat is to play the move bishop to B4. Bishop B4 would skewer or kabob the queen and the rook on the diagonal E1 A5. Okay. So now Garry's threat has been neutralized and the computer introduces a threat of its own.
One way of dealing with this is A2 A3. Garry could play A3, but again just as Maurice was saying this wasn't exactly what Garry had in mind when he dropped his queen back. So let's look at something much more specific. Again it's an opening situation. You wouldn't bring out your bishop -- your pieces as fast as possible. Let's just take a look at what would happen after the move bishop to F4.
Mr. Ashley Let me stop you for a just a second. I am noticing Kasparov shaking his head over this last variation and typical of Kasparov we mentioned before how expressive he is at the board and he tends to give everything away. Kasparov plays 15. a3
I remember you're mentioning that you would love to play him some poker because he is so expressive. And now shaking his head. It seems that he had not considered the nuances of this last move delicately enough and it's very possible with the symmetry in the position, with the very secure square on C5 for that bishop, Kasparov -- for the black bishop that Kasparov is not really ecstatic about what he has done and what he has allowed happen on this chess board. And it seems as if the computer will be able to do some kind of neutralizing possibilities. Well, in fact Kasparov has played A3. Remember you had mentioned the move B bishop F4 instead which looks like a reasonable move also.
Mr. Seirawan Again I, I just wanted to mention if we'll go back Kasparov has played the move A3. One of the moves he had to have considered is bishop F4 in order to bring that rook on A1 into the center of the board. But then after you see the move bishop B4 as a response, queen D3, knight F6, rook E2, queen takes D3, bishop takes D3. Black is complete has completely neutralized the position and no end game advantage whatsoever for Garry. So Garry had to in fact choose the move A3 or move his queen away.
Mr. Ashley The thing about A3 though is that -- does it give you the feel of here I come here I come I'm going to knock you out. You know that's not that kind of move, is it?
Mr. Seirawan No.
Mr. Ashley That's kind of boy, I got to wait. I can't just go hit the guy. I got to wait a move and play A3. And if you remember that move of DEEP BLUE's, when it played H6 we said oh, time waster. He lost a tempo. And you now you look at A3 and go oh, time waster. You gave back a tempo.
Mr. Seirawan That's a weak jab, absolutely.
Mr. Ashley And I'm not -- the last -- rook E1 and queen C3 seemed interesting but now it looks as if DEEP BLUE is neutralizing a lot of what Kasparov had.
Mr. Seirawan Indeed. Not to double guess Garry this early a stage. The move C5 for white a little bit earlier in the game was suggested because again as you say that bishop on C5 is in a very secure position. I'll just point out a tactic in the position. The tactic that I would like to point out is that black could play the move bishop B4 now, B bishop B4 looks like a really tricky move. Because after A captures B4, A captures B4, it looks like white's in trouble because the rooks are attacking one another and the queen on C3 is being attacked. But the simple response queen D3 would win the game for white because of the threat of queen H7 mate. So the immediate move bishop B4 isn't on yet. If black could just get in the move knight F6, then the move bishop B4 is a trick that Garry is going to have to watch out for.
More questions from the audience and in a very short while I think we'll be bringing in one of our IBM crew.
(Audience commentary.)
15. ... Nf6
Mr. Seirawan I'll invite Frederick to answer the question. What is the information, the graphs on Fritz's screen.
Mr. Freidel Do you mean this graph here?
Mr. Seirawan Just the whole thing.
Mr. Freidel Okay. I'll tell you. This is 1 engine which is Fritz 4. Fritz 4 has a lot of engines, Fritz 3.1, hire, a, et cetera. And we are using Fritz 4, the latest and strongest. This is the evaluation plus over equals means White has an advantage. It is 56 hundredths of a pawn. This is a little White which tells you watch out something has changed. When you're entering games or following them, watch that light. It will light up if someone made a blunder. 16. Be3. .
Mr. Ashley Let me give pause for a second. The move played by DEEP BLUE has come in. Knight to F6 has been played and Garry Kasparov quickly responded with the move bishop to E3. And I'll let you get back to our point in a second. But I'm very fascinated by what Garry Kasparov is doing in this game. Maybe you can talk about this a bit, Yas. The fact that Garry Kasparov is so intent on controlling the dark squares, now he's played this move bishop to E3 trying to exchange off Black's bishop. And if there's anything we can do in Black's position, it's the fact that the pawn from C7 has gone up to C6 and also the pawn on A7 has gone to A5 and has allowed for some potential of dark square weaknesses in the Black position, particularly a square like D6. I don't know how significant that would become, but in a lot of these positions we've seen the light square bishop. We see the light square bishop for Black. Even when it gets in the game, it doesn't have much to do. I mean this happens from time to time in this opening.
And it seems Garry Kasparov is carrying out a very simplistic strategy of just controlling the dark squares, getting rid of the bishop and then carrying out an assault on Black's dark squares.
Mr. Seirawan Okay. I'll follow up on that and I'll let Frederick get back to his answer. Good point. I think that's a little bit too long term. I think Garry has a very short-term concept in mind, and that is his rook controls the E file. And the crucial square on the E file is the E7 square. So the key piece that defends that E7 square is the bishop. 16. ... Bxe3.
17. Rxe3.
Mr. Ashley And DEEP BLUE, by the way, has captured on E3 and Garry Kasparov with the bishop and Garry Kasparov has captured rook takes E3 and his other rook is going to get right in back of it. The refreshment stall is it?
Mr. Seirawan The refreshment stall. The idea is that White, Garry, wants to follow up with the move rook on A1 to E1. This is known as doubling rooks on the open file and then penetrating the position with rook E7. And let Frederick just complete his explanation of all of the wonders.
Mr. Freidel Okay. I'll continue. This is move 17. It's considering B6, now B . There are 35 legal moves in the position. It's considering the second 1 of them. You know, Fritz just like DEEP BLUE goes through every single legal move. You can watch the counter going up. It's looking at an average of 9 ply deep -- no a brute force of 9 ply, looking at everything to 9 ply. And certain lines which look dangerous or strange or promising it's going down to 25 ply.
Mr. Seirawan Just explain the concept 6 ply a bit.
Mr. Freidel Ply is a half move. 9 ply means it's looking at I play this; you play that; I play this, et cetera, until 9 times. So I'm surprised that it's pretty deep, so this must be quite a fast computer.
Mr. Seirawan So when you say it's looking at 25 ply, that means it's 12 and a half moves deep into certain variations, 12 and a half.
Mr. Freidel Yes. Now it is going ten ply. It's decided it's completed its 9 ply brute force search, and it starts from the beginning. It has a move for 9 ply. Now it's looking at ten ply and it's still considering B5 to be the best move. Notice that Black -- White is better by about half a pawn. This is how the game progressed. However, this is not very accurate because we have been analyzing stuff. If we say -- if he plays, this I have a mate in 1. Fritz marks a wonderful advantage, which, you know, White didn't take in the game. So this is not very accurate. I could easily remove it and I should perhaps remove it for a while. And this is the main line. It is looking at it. You can see it has worked for 1 minute and 22 seconds. I can tell it to show me 2 lines. This is very useful to find out if a line is forced, if it shows me 1 line just retains the value and the next 1 loses a huge amount, it means that move is forced
Mr. Ashley All right. Thank you, Frederick, for that. 17. Bg4.
Mr. Freidel And this is the commentary here. It chatters at you. But we have a move.
Mr. Seirawan Careful now. We do have a move on the board.
Mr. Ashley And DEEP BLUE has played bishop to G4. Bishop to G4 on move 17.
Mr. Freidel Fritz was just considering that.
Mr. Ashley And bishop to G4 has been played. Now with the opportunity and maybe the desire on DEEP BLUE's part to exchange off that bishop, Garry Kasparov walking back into the room. He had stepped out for a moment. But again if DEEP BLUE could get in 1 more move, something like rook E8 and exchange off a pair of rooks, Garry Kasparov's advantage would be reduced somewhat, I would think. I wouldn't think that Garry Kasparov would be able to hope for too much in a position like that. But of course it's White to move. So he seems to have a very logical and forceful move, knight to E5 in this position, avoiding the exchange of his knight for the light squared bishop.
In addition to which those dark squares, Yas, you know you talked about E7. I mentioned the other dark squares in the position and it seems as if White's pieces are ready to take up residence on the dark squares. He's traded off the bishop on C5 and now his minor pieces, his knight can go to E5 easily, but also his major pieces they don't have to worry about that bishop anymore. The bishop on C5 that used to be on C5 is now gone, so a rook can go to E3. The other rook can go to E 1. The queen can sit on C3 and I think that's going to give White some real possibilities for advantage.
Mr. Seirawan Well, again, exactly. A move like knight E5, grandmasters just love to play moves like this. The knight we know is a very, very powerful piece when centralized. The deeper in the center and the deeper into the opponent's position the knight goes, the better for the attacker. The knight on E5 would control a wonderful asortment of squares, if we could just click on the square C6, B7, F7, G6. So you note that the knight is -- wow, the whole screen lit up. All of those beautiful squares that the knight could jump to. So grandmasters really understand how very powerful a knight in the center of the board could be. It's Christmas. Look at that. Okay.
And certainly that's the type of a move that Garry would pounce on. But if we'll go back just a half a ply, if you will, and go back with the move bishop C8, 1 of the things that DEEP BLUE had to consider was neutralizing the E file. And that was with the move bishop E6, stopping the rook from going to the refreshment stall. So bishop to E6, if you will, Frederick.
The problem with the move bishop E6 immediately is that Garry could have played knight D4, forcing the bishop to go back. So now let's go back to the position at hand, if we will. We'll go back to the game position. And now after the move knight to E5, then the move bishop E6 comes and the bishop can no longer be disturbed. So that's the counterweight to jumping out. I still think that the move knight E5 is an excellent move, occupying the center, preparing rook E1, sometimes preparing the move knight G6. And I think it's something that we're about to see.
Some more questions.
(Audience commentary.)
Mr. Seirawan Well, my answer to that is looking over here at Fritz, in the opinion of Fritz, it gives Garry Kasparov 53 percent -- hold on. Let's do this. 5300's of a pawn advantage. That is to say Garry Kasparov has a clear advantage from the opening still. So the answer is no, the move H6 has not completely neutralized Garry Kasparov's advantage. But I think what's going to happen is Garry is going to have to strike while the iron is hot. It very often happens that if you don't use your advantage it slips away. 18. Ne5
Re8.
Mr. Ashley And he has indeed gone for the move knight to E5, and before I could even remark on how wonderful the knight was on E5 striking out like an octopus in all directions, DEEP BLUE has responded with the move rook to E8, just instantly. It had anticipated knight to E5, and it decided that its best move, instead of bishop to E6 would seem positionly sound, it has played rook to E8, allowing the exchange of the White knight on E5 for the bishop on G4. This is an interesting idea. If DEEP BLUE can truly neutralize the position, that is if knight takes on G4 now, knight takes bishop and DEEP BLUE gets the opportunity, an exchange there on G4 and, B, maybe get rid of all the rooks, all the rooks, not just the pair of rooks on the E line, but potentially all the rooks in the game, then that could be an interesting situation for Black.
Black would then have the combination of queen and knight against queen and bishop, and that may be a position that DEEP BLUE could survive. And maybe even look to do something. I don't think we can speak of an advantage for Black in this situation, but that's an interesting positional decision.
Mr. Seirawan Well, there's a nice tactical variation that's quite forced, and it doesn't work. I'm going to get myself mated. So hold on here, folks. But the line goes knight takes G4, rook G3 and now knight F6 going back because I was threatening the pawn on G7 with mate, rook D1, queen E7, the only move. If the queen goes for example to C7 I have a shot like queen takes F6. That still doesn't work, I understand. But let's say queen E7 instead of queen C7. Now I'm going to get myself mated, but the idea was to play rook to D7. And after knight takes D7, I would mate with queen takes G7, but of course it's queen E1 mate. I know I'm going to get mated. But if queen takes D7, again I'm playing this as if I didn't have a king in position, queen takes F6, queen takes F6, I have this wonderful mating attack. It all doesn't work because my king on the back rank will suffer until a rook E1 mate.
Mr. Ashley But that's a very good point. It may seem facetious, but I think it's a very profound point you're making in that whenever you're attacking, you can't just go for what you have and think that you're not playing against an opponent. You have to deal with king's safety also. And the fact that you may get mated in some variation when you're going on the attack is a profound thing 1 has to think about. Sometimes what 1 has to spend a tempo in order to make luft and give your king some air. So that is a profound point. And there is the possibility of Garry Kasparov playing a move like H3 with tempo at some stage, maybe even now in order to deal with that mating idea. And by that deep variation you just pointed out where you ended up getting mate that had might give us the opportunity to say maybe the best move here is H3 and we might see Garry Kasparov do that instead of taking the bishop on G4.
Mr. Seirawan There is, however, in luft, making lufts with H3, luft meaning a German word to create air. So the idea of a move like H3 would give White's king on G1 the H22 square a square to breathe. More important than making luft, however, is development. And if we could go back, I think a very predictable move from Garry, Garry believes in bringing everybody to the party and that means the rook on A1 has got to have a role too. So a move like rook to E1 with the threat of knight takes G4, by the way, an immediate tactical threat would force the decision because, for example, White threat is knight takes G4 and then there's a discovered attack against the rook on E8. So this type of a move would fit in very nicely. And after, for example, a move like bishop to E6 it's quite clear that White -- Garry is building up his outlook towards Black's king. And there's a possibility that he again could go all out for an attack, dare I say it, with a move like F4.
Mr. Seirawan There was a deep breath there. There's a lot of people sucking air after I suggested F4. With the idea of playing F5, forcing the IBM on E6 to back up and then that E line is opened up. So we might see Garry start burning some bridges if he feels that he can get something immediate out of it.
Mr. Ashley Yeah. There are always, as you said, strengths and weaknesses to every move you make, every move you make in chess, maybe checkmate is the lone exception. But every time you make a move, you create a vacuum somewhere else. You cause an opportunity for someone to just step right in and say here I come, you didn't see this, did you? You thought you were winning on this side, but the last move you made weakened you over here. Every move in chess does that. Not just pawn moves as is standard we know, weakens the squares they left behind but piece moves too. Always one has to be careful of every aggressive move because in some way it may create a weakness.
This kind of aggressive move you mentioned F4, very aggressive. Want to control some space, rush your pawn down the board, chase a bishop back, at the same time weakening the white king. So your eyes might light up when you see a move like F4 but it also depends on the kind of player you are. The more aggressive players would think F4 no problem. I don't want to hear about the weaknesses. The more defensive players, difficult to play F4, man, my king is a little weak. It is there's always two ways to see these moves and Kasparov, as you said, I think realizes that this position here is super critical. His next few moves will say if he has an advantage. Does he really have anything with his space or is he going to be neutralized in a few short decisive moves by DEEP BLUE?
I would say we would know in the next 4 or 5 moves what is really happening in this position. And Kasparov knows it. I think he senses it that it's right now that he has to be extremely concrete or lose whatever advantage he had already built up in space. We know that lead in development is usually temporary. So I would anticipate -- I'll make a guess that this is a 20 minute think position.
Mr. Seirawan You know, the old 20 minute think position.
Any questions, please? Are we doing such a good job? Come on folks.
(Audience commentary.)
Mr. Seirawan Joe, the gentleman asked me about yesterday's game and the programmer Frederick might just find that position so I could do so. We're really off base with today's game, but we're looking at yesterday's game. And this was. And now the idea at this moment in yesterday's game Garry came up with a very --
Mr. Ashley At the moment we are where DEEP BLUE played the move 22 rook to B4 Kasparov had two moves to make. He chose rook to A5 but you had mentioned rook to A6 as a response.
Mr. Seirawan Right. Rook to A5 is the natural move. Every human being will make the move. But there's a very deep problem in the position and that is that DEEP BLUE managed to save the game by posting his bishop strongly on D6. So I was looking at the move rook A6, rook C4, castle and now comes the trick, bishop D6 and now B5, bishop takes F8, B takes C4, bishop C5, knight to E4, rook to B1, G5, room to B7 P, knight takes C3, rook takes A7, rook takes A7, bishop takes A7, knight to many B5, bishop to C5, C3 and the pawn runs home. This is what I --
Mr. Ashley Is this what grandmasters do in their sleep?
Mr. Seirawan Absolutely. And in fact not only do grandmasters do this in their sleep, they oftentimes remember. And what happened was one of the great American champions, Frank Marshall, every night when he went to bed he took with him a notebook and a pencil because he said you never know when you might wake up with a good move in mind. And he in fact credited a good night's rest to his finding the Marshall gambit, one of the most profound defenses in modern chess theory.
So, yes, I mean I don't know when you're going to get an idea like this, but that was what I was thinking about last night. And you might go back and try to find other defenses for DEEP BLUE, but I'll be darned, I think you're going to discover what we just saw was a main line.
Mr. Ashley So instead of 22 rook to A5 it may be that 22 A rook to A6 was the most accurate, although an awkward looking move indeed but may have been the most accurate move in the position.
Mr. Seirawan And what you'll recall, for those of you who were with us yesterday, Garry lamented, he said, I was a tempo from winning, a tempo behind from winning. And I think that move rook A6 would have given him that crucial, crucial tempo.
Anyway, we'll go back to today's game. A little applause there. You like that?
(Applause.)
Mr. Seirawan We had expected somebody from IBM's team a little while ago.
Mr. Ashley Maybe they're a bit concerned about the computer's position and the nervousness is keeping them away. I'm sure that we will get others here shortly. But for the moment, for the moment we see Kasparov, he's been thinking now for over 5 minutes or approximately 5 minutes. We had talked about him putting in the 20 minute think in this position. It is very easy to envision his doing so because the position is critical. We called this kind of position a critical position. It is so important for Kasparov to take a long look at this position and make sure that he plays the most precise move. Otherwise in about 4 or 5 moves time we won't be talking about Kasparov winning the game at all.
Mr. Seirawan We have a little bit of a pause because Frederick will have to rekey the game and it will be no problem.
Mr. Ashley Yeah, while he does so, Kasparov has found himself in these kind of critical situations before and you talked about the importance of this game. If DEEP BLUE were to win the game.
Kasparov not having an advantage, not winning.
Mr. Ashley All heck would break loose.
Mr. Ashley What's the value in this game and the rest of the match?
Well, first of all, it's very strange. I have a great deal of empathy towards Garry in this particular match because I really want to see him clobber this thing. And I was thinking about it. What Garry had said before, that the computer literally will not let you -- is remorseless, will not let you back into the game.
If you make a mistake and the computer has an advantage, it's going to punish you for the rest of the game. So Garry is in a situation where a single mistake may cost him the first prize of $400,000. Now, he knows that he's a big favorite to win the prize, but at the same time he has to stay with the fact that he really cannot afford a mistake.
And that really kind of puts a lot of tension on Garry where he may not play in his normal style. But spend a lot of time double-checking and rethinking his analysis. And that's I think what we're seeing a bit with Garry at the board today. He's doing a lot of recalculation. We are fortunate because Joe Hoane will join with us very shortly to explain the engineering marvel of IBM's DEEP BLUE.
But again I think it's very crucial game today. If Garry can win obviously the match dynamics are much in his favor. If he loses, well, then he's up against the wall. At this moment I'd like to say welcome to Joe Hoane of the IBM team. Please join me in a round of applause.
(Applause.) Mr. Hoane I'm glad you're all here. This is great.
Mr. Seirawan Joe, a couple of questions to start us up.
Mr. Ashley Yas, before we continue --
Mr. Seirawan Do we have some moves?
Mr. Ashley No, we haven't. But I have to do an interview for radio, and I will be back shortly. So thanks, Joe. I'll be back and joining you with the rest of this game.
Mr. Seirawan Maurice Ashley taking a break. Don't stay away too long. Thank you very much, Maurice.
(Applause.)
Mr. Seirawan Let me key you up here. How many years have you been working on the program yourself?
Mr. Hoane It's been since '91, '90, something like that.
Mr. Seirawan And what has been the breakthrough -- I'm trying to coax you to talk about the chips because I know they were recently pressed. What breakthrough has these new chips meant for DEEP BLUE?
Mr. Hoane I can answer that, I think, very well by talking about the process that we've gone through. We've had 3 iterations on the hardware, I guess 4, if you count CB's hand-wired circuit board. I mean what happens in computer science terms, it's just a computation you want to perform. And you know how do you do it fast is the question. I mean, how we do computer chess, how Frederick does computer chess. It's all how do you do it fast? I mean it's just -- there's a few things that go into the problem. There's chess knowledge, which is of course really important, but actually makes it go slow, as I think Frederick has been remarking in general. The more chess knowledge you put in, the slower it goes because you have to execute code to -- that embodies the chess knowledge. And so you can approach it by having good algorithms. You can approach it by having work on computer architecture and you can wait for technology to advance. I mean when the new Pentiums come out, of course, Frederick is going to demonstrate his program on the new Pentium because technology has advanced and the program is all of a sudden a lot better. And that's a very interesting thing. If you just sit and wait, it's going to happen.
So we've taken advantage of all of these things. But in particular we don't have the bottleneck that all software programs have. We're doing all the chess knowledge and parallel all at the same time. When we will say we're doing 500 million positions, we're not doing that. When we say we're doing 50 million or a hundred million positions per second, that means we're doing that many evaluations to which would be utterly impossible just in software.
Mr. Seirawan I think so, but to be very clear because parallel processing to me, maybe I have the wrong impression. I sit at home, as you say, with my laptop Pentium and I'm bouncing around and I've got a single CPU, but DEEP BLUE doesn't have a single CPU. It's got how many and how do they all work together?
Mr. Hoane Okay. What it has is the code processors that are specialized to chess. As you alluded to, we have a new co-processor which is a third or fourth generation. What you do is take what you know about chess and push it down into the hardware. Just having 1 chip makes it fast. Now it goes faster just because you've got this piece of hardware that does exactly the thing you need quite a bit faster than the general purpose computer would do. Then what you've got to try to do is get many of them to work together. And that's where the win comes in this case. Alphabeta algo is hard to paralyze, but to the extent you do it, you get a tremendous advantage from a having a parallel processor.
Mr. Seirawan And you have a situation then where basically you have all of these slave ships, if you will, all working parallel processing, and there's some central chip giving all of these other chip commands, package of information, go solve this problem and come back with it. So you have a lot of handholding between the chips. Is there communicating packages of data beta back and forth? Is that a quick or slow process and are you going to try to heighten that communication between the chips?
Mr. Hoane Well, that's the problem to work on, is both how they communicate and generating any parallelism at all. I got to make this concise somehow. There's 2 problems. How do you get things to work together well, but in alpha beta, in chess, it's also how do you generate parallelism and that's part of the problem because you know when you have a chess position, you've got, you know, 36 moves on average and you can just break it up and search them all at the same time. That's 1 aspect of how it works. I hesitate to go into more detail.
Mr. Seirawan But right now obviously we know, we've been told that you're looking at 200 -- that DEEP BLUE is looking at 200 million positions a second.
Mr. Hoane Yes.
Mr. Seirawan Do you believe that just with the technology you have now you could probably increase that by increasing the communications?
Mr. Hoane There's no doubt, just working on the communication and the software aspect of the communications, it could go quite a bit faster.
Mr. Seirawan Tell me quite a bit faster.
Mr. Hoane Double should be easy.
Mr. Seirawan So you're saying with the hardware, with the software you have now just tweak it a bit and you're looking at 400 million --
Mr. Hoane You're talking about 6 months work, but yeah, you can do it.
Mr. Seirawan Okay. At this point I'd like to open up the audience, some questions.
(Audience commentary.) (15 &16) Mr. Seirawan How many extra plys do you get from the various 256 processors?
Mr. Hoane I don't have a good measure. I didn't have time to instrument anything. I think we go through our iteration 12 or 13 fairly often now. In DEEP THOUGHT two where there was 11 host with 16 co-processors we would go to our depth. So we're talking about on average -- well, sometimes ten. Sometimes 11. It just depends. I'd say two more ply.
(Audience commentary.) Mr. Seirawan What is the alpha beta.
Mr. Hoane I think there's a way to say it fairly simply, which is in a given position, in a given chess position, you've got all these moves and the alpha beta essentially says pick one, find out what the score is, 42, okay, you know that's the score.
And then what you do is you just prove all the other moves aren't as good. And, you know, chess master may fall back on this strategy when he doesn't know what to do, which is, you know, eliminate all the moves that lose and then you got one or two left.
Mr. Seirawan And then you toss a coin in the air.
Mr. Hoane Well, the computer sort of works that way. I might mention though that one of the differences between a human chess player and the computer is as far as I know chess players -- I've got a quote from you here.
Grandmaster Seirawan said I see the position and I see the essence of the position. I take that to me mean that he also sees like one or two moves that are going to be made. I mean it's just look at the position, there you go, it's this move. I mean I've heard that many times in the commentary. And in general it's a little different from the computer. The computer doesn't look at the position and know that all these 36 other moves are bad, you know, as Grandmaster Seirawan does. I mean it he just knows. He's got all these years of experience playing chess and that leads him to say all these moves are bad and maybe this other move is okay. Maybe this move is good. And the computer just spends all its time saying, well, no this move is bad, no, this move is bad. Well, this move is bad too and this move is bad.
And that's where you spend all your time. That's why you don't really go any deeper that be the grandmaster. It's only as the speed of the computer has come up and up and up that you get now we're just about as deep. I think grandmasters look at two or 300 positions in a search. Well, that ends up in 11 plies. It's about the same depth. Whereas the computer you do the same thing. It's looking at an average of 5 point 6 moves in the position and it's going to about depth 11. It's the same number, the same time. It works out.
Mr. Seirawan We have a question to the right.
(Audience commentary.)
Mr. Hoane Well, you have to really get into the problem in order to do well at a problem. You can't play chess well by knowing nothing about chess. And so the answer is there's a lot of work there. But you look at the new problem and you say well I've got these techniques that are fairly novel and I know how to do the search and I know how to do the communication and you can just bring the whole thing up much quicker than you could if you hadn't had the experience with chess.
And furthermore it leads to you even tackle problems that nobody can figure out how to do. So there's a number of aspects about why this would proceed much more smoothly based on our experience and our techniques. It's more applying a way of doing things rather than any specific piece of code. 19. Re1.
Mr. Seirawan Let me just interupt for a brief moment, we do have the expected move by Garry Kasparov rook on A1 to E1, bringing everybody into the party as we said. It does set up a tactical threat of knight takes G4 winning a piece.
So DEEP BLUE will have that problem to solve. My expectation as I mentioned before was bishop E6 and then we have this very bold thrust F4.
Garry Kasparov went in for attack on in game one was in his own words soundly massacred. So he will approach a move like F4 with trepidation. But in the meantime DEEP BLUE is still thinking and we'll continue our discussion.
Another question, please.
(Audience commentary.) 19. ... Qc8. Mr. Seirawan The question referred to yesterday's game that the move castles was the first original move and no it wasn't. In fact, what DEEP BLUE did was avoid playing the move H3 and came out with knight E5. So it was in its book. DEEP BLUE, if I understood correctly, played 15 or so expected moves that your team had prepared.
Mr. Hoane Yeah. The book was quite deep.
(Audience commentary.) 20. ... Be6
Mr. Seirawan It was prepared differently. The programmers had prepared it in a different variation.
And again just to interrupt the discussion, we do in fact have the expected move bishop E6 and now Garry has to make his decision. I mean I feel that Garry has built up his game beautifully. All of his pieces are coordinating. Please keep in mind that these rooks are nicely positioned on the open E file. DEEP BLUE still hasn't done anything with this rook in the corner on A8.
Superbly centralized knight. This knight is a little bit more defensive, a little more passive. Garry has also developed his queen and DEEP BLUE's queen has yet to move. So from the perspective of piece activity, Garry is like a spring coil. He's all ready to go. But again, he's got this dilemma, should he play a bridge burning move like F4 or should he try to increase his advantage in a slow, methodical way with a move like H3 or perhaps even B4 trying to gain space on the queen side?
I vote for F4, but it's something that Garry is going to have to worry about.
Again I will invite you to ask questions of Joe. It's an honor to have him.
(Audience commentary.)
Mr. Hoane I don't know that it's not as good -- 20. f4
Mr. Seirawan Just for a moment on that. That is actually something that I'm not actually sure is accurate. I know 5, 6, 7 years ago the grandmasters like myself, we used to laugh with gleeful happiness. Look at this silly move the computer has played in the end game. And it did play end games quite badly. Yes, we've got it. And it did play end games quite badly, but an enormous number of improvements have been made. I'll let you answer the same question, Joe.
Garry has played the move F4. He's showing his confidence, isn't he? This is just beautiful. I love it. F4. The idea is he wants to play the move F5, forcing the bishop on E6, driving it away, so that his rooks could have command of the open E file. This move F5 is going to be a really difficult one for the computer -- the computer err is going to have to meet that. And DEEP BLUE may have to play the move queen C8.
Audience Member It did, instantly.
Mr. Seirawan The computer did in fact play the move queen C8. Because of the threat of F5 as such a strong threat, he had to blockade the F5 square. And we'll go now --
Let me just say offhand, Garry has got several ideas, not the least of which is again an expansive plan on the queen side to play a move like H3 and perhaps follow it up with G4 and then with F5. It's not the sort of thing that Garry would like to do.
At this moment I'd like to introduce Joel Benjamin, working with the IBM DEEP BLUE team, grandmaster Joel Benjamin.
Joe, could you please answer the gentleman's question about the ending? How good is DEEP BLUE's ending and is it going to be improved?
Mr. Hoane Well, as far as I know DEEP BLUE is fine in the ending. I think the traditional weakness of computers has to do with the very great consequences of depth, of things that happen very, very far away in terms of -- you could move a pawn and make a whole bunch of other moves and move the pawn again and make a bunch of other moves. And the consequence of moving that pawn is so far away that computers have trouble seeing it, but we're get to go the point that we have the raw depth in search to cover those. 21. h3
Mr. Seirawan Just a quick question. We know that computers can take certain select positions like 4 piece or 5 piece endings where there's only a maximum of 4, 5, 6 pieces on the board and then using a data base technique -- oh, we have seen it, H3 again in the same motif, the idea being to ensure the advance of the kingside pawns. But going back to the question of the end game, we're now seeing that the computers have had a marked effect in grandmaster play because they've discovered positions that can be won after 200 moves and chess players have said in the past, well, if we can't win them in 50, we'll call it a draw.
Mr. Hoane Isn't that still the rule though? It doesn't really matter what you find. Still 50 moves, it's a draw.
Mr. Seirawan They've tinkered with it and have sometimes for specific positions allowed 75 moves of play.
Also, I'd like to involve Joel into the discussion.
Joel, are you satisfied with the opening today?
Mr. Benjamin Well, I'd have to say that this position looks a bit ominous for Black. White certainly is building up an attack. However, I will say that the hardest way to beat DEEP BLUE is to overwhelm it with an attack because it's just tremendously resourceful, much more so than any human player. And if you slip up in the slightest in the exec ution of the attack, then it will find a way out. So what we want to do is we want to make Garry -- we want to force Garry to play as accurately as possible on every move because he's -- he may not be able to play the perfect move every time. He may not have enough time to figure out what the perfect move is. I think I would have preferred to have a more active position than this, but it's still possible that DEEP BLUE will find ways to make it difficult to conduct the attack.
Mr. Seirawan You want DEEP BLUE to play positions where Garry is provoked into attack with the idea that he may miss something and then the computer comes back with a countershot?
Mr. Benjamin Yeah. Well, we don't mind being attacked, just as long as we're able to counterattack or to create problems along the way. So that the attack won't be simple and smooth and just play itself. We want there to be problems along the way. This position is perhaps a little more methodical than I would want to see. 21. ... b5.
Mr. Seirawan Again, this is 1 of those 30 foot shot moves. But I think it's a good 1. DEEP BLUE has played the move B5, putting pressure against White's C4 pawn, the idea being that if DEEP BLUE can get rid of the pawn, either by forcing its capture, then there's this move knight D5 forking queen on C3 and rook on E3. Probably after the move B5 Garry may ignore it with the move B3, and then the computer comes up with A4. Just what you're saying, trying to make the position as tricky and as complicated as possible.
Mr. Benjamin It's just not going to make it easy for White to conduct the attack. It's going to throw as many obstacles in Garry's way as it can. And I think any human opponent would be very intimidated to have to play this position with Black, but --
Mr. Hoane It's 1 of the strengths of the computer. It's not intimidated and, second of all, it does defend really well. The lack of intimidation is justified. And so it's doing something over on the queenside with this big attack.
Mr. Seirawan Making a diversion.
The gentleman in White.
(Audience commentary.) Mr. Seirawan The question was what would happen if DEEP BLUE plays itself, as I'm sure it does. I mean you test various engines against one another. Does white win all the time or does DEEP BLUE always lose to itself?
Mr. Hoane No. Actually what happens is you just -- whoever loses basically usually exposes a weakness in the program.
Mr. Benjamin Well, I've seen Fritz play itself constantly and seems like something kind of random happens. Sometimes white Fritz wins and sometimes black Fritz wins and it's probably kind of like that with any program.
Mr. Seirawan But just as a followup to the gentleman's question, does DEEP BLUE's program play score a higher percentage against itself when it's playing white?
Mr. Hoane I don't have those statistics.
Mr. Seirawan All right. Another question.
(Audience commentary.)
Mr. Hoane Do I change the nature of --
(Audience commentary.)
Mr. Hoane Yeah, you have to. The pawn threats are very different in the end game and there are lots of small adjustments but the pawn is treated differently is the big one.
(Audience commentary.)
Mr. Hoane Almost everything.
Mr. Seirawan So that's very interesting. So it's almost like you might say you have different modes of play. How would the computer know when it's an end game? What defines the end game that the computer then says to itself well I'm going to switch off that mode of play or that brain and play in a different -- use a different algorithm.
Mr. Hoane Well, getting that exactly right is tricky.
Mr. Seirawan For me too.
Mr. Hoane Sometimes when the queens are on the board but there's just a couple pieces -- there's a couple players of an end game.
Mr. Seirawan Take a question.
(Audience commentary.)
Mr. Seirawan What the gentleman had asked is it's clear that Kasparov's style is being compromised to agree. That is to say he's holding back changing his attacking style and so he's saying well I'm going to play in a different mode of play. Are you yourselves also making tweeking your strategy --
Mr. Hoane I think I have a good answer for this, for the real thrust of the question. It didn't about any particular opening. It's that the opening is a negotiation. You know, and you figure out what positions are comfortable for you and uncomfortable for your opponent. It's really negotiating. You have this knowledge you go through and you find the path that's good.
Mr. Benjamin Let me just elaborate. That very question, that's one of the main reasons why I'm here, is because I can add a knowledge of Kasparov that they don't have. And I know a lot about how he approaches matches and a lot about how he would approach this match in particular. And I expected that he would play positional openings, would he play solidly, he wouldn't go for wild positions, even though he often does those in tournament play against humans because that particular style is much more effective against human opponents and he knows that. So we expected him to play in that manner and in response I gave a lot of thought to what we could do to make sure that we don't play too much of that kind of strategical game where he can just grind us down and try to get to a position where DEEP BLUE can flex its muscles.
Mr. Seirawan Well, again, just to buttress what both of them said, in chess, you literally have a choice of thousands and thousands and thousands of openings and what grandmaster Joel Benjamin has done is come to the DEEP BLUE team and say listen, I know Garry's style of play and if you play these types of opening positions against Garry you're dead mate. So you better make some adjustments here and you've got to tell me what DEEP BLUE is going to feel most comfortable in because I can tell that you if you give Garry these types of positions he's going to kill you.
Mr. Benjamin Yeah. But that thinking, it's a little bit different than if you're preparing, you know, a human opponent for Kasparov. Would he say, you know, the passions that you would say not to go into are completely different. In this case, you know, we weren't that afraid about really sharp positions.
Obviously we looked at them and we gave it some attention, but the biggest fear is that would he just get to some position that the computer doesn't understand. That's always a big fear, that that kind of thing can happen. And we always feel more comfortable when the computer has a position where it's a little bit easier for it to generate aggressive play for itself. It's rather obvious that's the right way to go.
Mr. Seirawan I'd like to ask two more questions and then we'll get back into the analysis of the game.
(Audience commentary.)
Mr. Seirawan We'll be getting into the game. I just want to take advantage of the presence of the two people on stage here.
(Audience commentary.)
Mr. Seirawan What are some of the limitations of what DEEP BLUE can understand? Positions?
Mr. Hoane I think it's dangerous territories for us to answer any question like that.
Mr. Benjamin Let's just say, you know, that in general there are some things that it doesn't understand that well. There are some things that perhaps we haven't detected. There are some things that we have a fair idea about, but, you know, we haven't either figured out exactly how to correct them or haven't had time to correct them. And, you know, we know that there are, you know, these weaknesses that tech attack and, you know, we just hope that they don't come up, that we get the kind of positions that we want and we don't get those positions that we are -- where DEEP BLUE will have problems.
Mr. Seirawan It seems to me like computer programming is a constant tweeking, a constant self-evaluation and until you get it better and better and better and better hopefully you don't do damage while you're making it better.
Mr. Hoane No free lunch. You do something and maybe it makes something a lot better or something a lot worse.
Mr. Benjamin You have to do a lot of testing.
Mr. Seirawan Lots of tradeoffs.
One last question for Joel.
(Audience commentary.)
Mr. Hoane It depends on what it is. I mean in the old machine -- you're asking him or me?
Audience Member Either one.
Mr. Benjamin If you're talking in terms of reprogramming, if you're just talking about opening variations, that's very simple. That's -- well --
Mr. Seirawan Inputing.
Mr. Benjamin That's just inputing. That's just like entering a game in chess base. That's much more simple. Much more complicated is if you want to, you know, adjust the evaluation of various situations. That's much more complicated. It's much more difficult to effect, and effect and it has much more serious repercussion that need to be tested.
Mr. Hoane I have a good example from Deep Thought two. I spent a month of my life on bishop mobility, just making it really understand bishop mobility and how to use its bishops. And after a month of good work it was fairly deadly with its bishops, with one bishop or two bishops. But it's a lot of work.
Mr. Seirawan I'd like you to join with me now to thank Joe for coming up and giving us a little bit of insight into the wiring. 22. f5.
I'd like to invite Joel to stay because we have a new move on the board. In response to DEEP BLUE's move B5, attacking white's pawn on C4, Garry has just played the move F5, attacking the bishop and it looks to me, Joel, like we're going to see a forced sequence of moves. If we can go for it let's see what the resulting position will be.
Bishop takes F5, bishop takes F5, queen takes F5. So now the move B5 having weakened the C6 pawn, it looks like Garry will happily slurp. Knight captures C6, leaving opening this possibility of knight E7 check. It looks like a move like either rook takes E3 or maybe queen C5 immediately.
Mr. Benjamin My first instinct is that that is not what he's going for. I think he's got something rather more forceful than that because it seems like his pressure is worth more than that kind of liquidation. So --
Mr. Seirawan What would Garry's trick be if he doesn't play knight to C6?
Mr. Benjamin Perhaps we can go back one move. Maybe there's something --
Mr. Seirawan Okay. We now have the position at hand with Garry just having played the move F5.
Mr. Benjamin I think black will take on F5.
Mr. Seirawan Okay.
Mr. Benjamin I think this is a big problem, black can run into big problems if he doesn't do that.
Mr. Seirawan Indeed. So after bishop takes F5, if there's not the followup bishop F5 the only possibility is like the rook might slip over and make some damage on the F file or is there some trick associated with knight F7 that I don't quite understand?
Mr. Benjamin I don't see it yet either because there's various types of tricks involving a move like knight to G4. But here I think black can just take that with the bishop.
(Audience commentary.)
Mr. Benjamin Knight takes C6 right away. That looks more dangerous because black can be overloaded. But then it will just take the knight. It will apply queen takes F6.
Mr. Seirawan If F5, bishop takes F5, knight takes C6, then we think --
Mr. Benjamin Just take a knight.
Mr. Seirawan Queen takes C6, bishop takes F5, queen takes C4 and black's happy. He just munched a pawn and so forth. So we'll go back a moment and we're again at this position and we think that bishop takes F5 is the most consequence move. There is also the possibility of bishop takes C4, but again bishop takes F5 being the most consequence. Let's continue with this line.
And again let's go with the flow here. Bishop takes F5 Mr. Seirawan Queen takes F5, knight takes E6. We can say we might -- Garry certainly at least has the possibility of getting into this position by force. There seems to be this move queen C5 that really puts a damper of enthusiasm on the position. The move queen C5 gets out of the threat of knight E7 check. It pins -- pardon me. It pins the rook to White's queen -- White's king.
Mr. Benjamin Well, I think perhaps White can just take the pawn on B5.
Mr. Seirawan B takes C5, winning a pawn. Because obviously queen takes B5 has some shot associated with knight E7 check, perhaps.
Mr. Benjamin Or the A5 pawn.
Mr. Seirawan Oh, just simply trading all the rooks, like rook takes E8, rook takes E8, rook takes E8, knight takes E8, knight takes A5. Certainly Garry would be happy about being a pawn up. Okay. We'll go back. So maybe rook takes E3 immediately. And now after rook takes E3, there is still this threat of knight E7 check, inducing the queen to move. We have queen B1 check or again queen C5. Perhaps queen C5 here.
Mr. Benjamin It still seems to me that this is --
Mr. Seirawan A cop-out. Why does it let something slip away?
Mr. Benjamin It seems that Garry is perhaps playing more to the situation than he should, that he's -- you know, he's a bit nervous about the awesome computational power, and he wants to simplify the position. I have no doubt that White has the advantage, but perhaps it's an indication that Black's position really was fairly solid and that the aggressive posture of the White pieces didn't necessarily mean that there was any big attack brewing. I think also with this move H3, that was kind of an indication that he was moving into a more positional mode, just making sure that his king had an escape square and cover the G4 square. It wasn't that much of an aggressive move
Mr. Seirawan Invite comments from the public.
(Audience commentary.)
Mr. Seirawan We have a buzz going on. Currently the programmers are all around the DEEP BLUE screen. Murray Campbell busily typing away. From time to time I know that they do have problems with the telephone hookups, and yesterday we had that confusing problem, Joel. I think Feng had accidentally typed in the wrong move and caused a fit of discomfort --
Mr. Benjamin Okay. I have --
Mr. Seirawan -- to DEEP BLUE.
Mr. Benjamin I have a more accurate description of what happened. Apparently -- this is according to Feng -- he said that he typed in Black's move, Garry Kasparov's move, knight F6 properly, except that there was some kind of residue on the screen. There was some kind of character that was somehow before the move, the move that he typed in. So the computer read in something -- some character knight F6, which it took to be an illegal move, and the consequence of that illegal move was to mess everything up. So that was what happened.
Mr. Seirawan What, there was a smiley face?
Mr. Benjamin Something like that, some mystery character.
Mr. Seirawan Well, bereft of no new knowledge, we don't know exactly what has gone on. Obviously this is a totally new model with all of these chips, and you've got telephone lines and modems and all kinds of things that could potentially go wrong for the computer. It would really be a tragedy if the computer had to resign at this point.
(Laughter.)
Mr. Seirawan We'll take it. But I suspect that we've got some very capable people. They'll probably fix the problem. In the meantime, any other questions that you might have.
(Audience commentary.)
Mr. Benjamin Yes. And I don't think it will take long because I think that move loses for Black. I'm afraid that the move --
Mr. Seirawan The gentleman's question was is Garry trying to force an end game quickly believing that the computer is playing so well in the opening and in the middle game that maybe he's going to try to grind him down in the end game as he did in game 2. And indeed he did win an end game that Garry thought was drawn. So sure, I'm sure he'd like to test it some more. But in case of bishop takes C4 --
Mr. Benjamin I believe knight G4 is very dangerous.
Mr. Seirawan The move knight G4.
Garry seems a little bit annoyed and questioning.
Mr. Benjamin It is running. Yes, it is running. I can see the time ticking off on the digital clock. It is running.
Mr. Seirawan Garry has spent an hour and 18 minutes on his -- what is it, 22, 23 moves, and DEEP BLUE has only spent 57 minutes, but the clock is ticking and obviously there's a problem with the board, but now it seems to have resolved itself, as only 1 person is at the screen.
Problem at the board.
After bishop takes C4, the idea was knight G4, keeping in mind there's this terrible threat on the E line, this discovered attack. So that in case of rook takes E3, the ensuing knight takes G6 check, and maybe just simply queen takes E3. This looks very strong for White.
Mr. Benjamin You know it's possible that that position might actually be defensible because the 1 thing that Black has going for him in this case is that the White bishop on C2 is finally blocked out of the attack.
Mr. Seirawan Because of the pawn on F5.
Mr. Benjamin Because of the pawn on F5. But this is the kind of thing that DEEP BLUE can actually probably calculate out, whether it's going to be an immediate win for White in the next few moves from this position and if he sees it, he'll avoid it. But actually now it's possible that this is -- that this line might actually be playable.
Mr. Seirawan And that's the interesting thing. Again we have the situation where, you know, somebody like Joel or I would say well we'll never capture that pawn on C4 because we would be frightened to death.
By the way, we just had Mike Valvo walk in. Mike didn't want to talk to us. Mike Valvo, the international arbiter looking for Ken Thompson. I haven't seen Ken. I have not seen Ken.
What I was saying is that grandmaster Joel Benjamin nor myself would be favorably disposed to this position on the middle of the screen, but it's something that DEEP BLUE might be able to defend. And again it seems we're getting mixed signals here. We don't know if they've resolved the problem or there's still a lot of problems.
(Audience commentary.)
Mr. Seirawan Don't know the answer to the second question.
Mr. Benjamin Well, I can answer the question about Thompson's role. He's very knowledgeable about computer chess, and he's somebody that both sides agree is both very knowledgeable and completely reputable and honest and fair. So just in the case that something like this, whatever it may be, some kind of problem would come up, he might be able to arbitrate the situation 22. ... Bxc4.
Mr. Benjamin We're back in business. We have a move.
Mr. Seirawan We do have a move. And again this was -- you described it so well. The 30 foot shot that you go no no no no.
Mr. Benjamin Exactly. That's what -- I have to say one thing, that going into this match I figured that there would be a number of occasions when I wished that I could jump in and take over for the computer. I don't think that's happened. You know, not even that second game, that it miss played the end game. I wouldn't have wanted to be tortured like that myself.
I'd have to say that in every situation I was happy that it was the computer and not me. It was the kind of thing, you know, where it was a position that could be very intimidating and daunting to have to handle and better it than me.
Mr. Seirawan We have a gentleman.
(Audience commentary.)
Mr. Seirawan Maybe I can answer that question. The gentleman suggests something that is quite standard in international play, what is called an autosensory board. The idea is that there's a coil in each and every piece, and underneath the squares you have this basically magnet radio signal so that whenever a piece is played, it's automatically projected on the monitor. The problem is that I've noted in those autosensory boards, the pieces are not properly weighted. They're quite light. You can't have metal inside the pieces because that would interfere with the radio signals. So I believe that the choice here was actually Garry's. Garry didn't want an autosensory board and he was accommodated. I like personally a beautiful wooden set.
Mr. Benjamin I don't know anything about that. Certainly from the point of the DEEP BLUE team we were happy to play with whatever equipment he wanted to use because it makes no difference to DEEP BLUE. And he should certainly have whatever -- Garry should have certainly whatever he's comfortable with. I don't know if there's any technical problem with hooking up an autosensory board to DEEP BLUE. It might not be as simple as you want to think. Tech stuff is not my field, so I don't know the answer to that.
Mr. Seirawan I'm just looking at the position at hand. Again, we're looking at a possibility of a pawn sacrifice for Garry Kasparov. Again, this would involve a direct attack against DEEP BLUE's king, something we saw rebound horribly in game 1 and something that Joel and I would not want to face as Black. But again we have seen the incredible resourcefulness of the computer. The idea is knight G4, rook takes E3. The Swiss ensuing knight takes F6 check, forcing G takes F6, opening up Black's pawn shield near his king. Then after a response, for example, like queen takes E3 or perhaps even rook takes E3, I'm having a hard time wondering what would be a better move, but Fritz likes rook takes E3 so we'll go with it for the time being.
White is pressing on the F6 pawn, as well as rook G3. We do have a different alternative, however. Garry didn't want to sacrifice. 23. Nxc4 bxc4
Mr. Benjamin I should say something, that not only did Garry play knight takes C4, but DEEP BLUE obviously expected it because it moved right away. So DEEP BLUE wasn't that worried about knight G4.
Mr. Seirawan So we have it. F5.We're getting there. I'm having difficulty. Pardon me. F5. We now see the move bishop takes C4 and I'm just -- how do I get that?
Mr. Benjamin Because you're in the wrong place there. 24. Rxe8 Nxe8
Mr. Seirawan Right, okay. Bishop takes C4. Now what Garry did, rather than go through that pawn sacrifice variation with a move knight G4 he played knight takes C4, opening up the E file. DEEP BLUE immediately responded by capturing the knight. He could not capture the rook on E3 because this would have resulted in the loss of a piece by knight takes E3. So immediately came the move B takes C4 and now Garry Kasparov has played rook captures E8 check and forcing the move knight captures E8 and now Garry Kasparov has the nice recapture -- the simple recapture queen takes C4, and you have to say that Garry has a good game due to the fact that Black has 2 pawn weaknesses, the pawn on C6 and the pawn on A5 give White a comfortable advantage. Maybe not too much of an advantage, but comfortable.
(Audience commentary.)
Mr. Benjamin It hasn't played knight takes C8 because it doesn't instantly respond just because a move seems to be obvious. It doesn't do that. It does its search. This is something that's a little bit of a negative in practical play. There are some situations where it will make an instant move, for instance, if it's the only legal move, it will play the move instantly. But it's a bit tricky to decide situations where you want to make it play an instant move because there are -- there's always the danger that it will play too quickly in a position when you need it to think.
Mr. Seirawan And we have seen the recapture knight takes E8 and Garry is now considering his next move, which we're expecting to be the simple recapture 25 queen takes E4. I'll just point out that the clock situation is as always, actually, in the favor of DEEP BLUE. DEEP BLUE has taken 1 hour and 2 minutes for all of its moves, whereas Garry has taken an hour and 23 minutes. And of course the time controls move till 40. Each side gets 2 hours for the first 40 moves and thereafter the game is not over after 40 moves, then each side will get an additional hour for the next 20 moves. So we have time controls at move 40, time control at move 60.
We're lucky to have back with us my colleague international master Maurice Ashley.
(Applause.)
Mr. Ashley I noticed, as you did, a bit of confusion that happened in the game, and I would just like to let you know actually what happened back there. It turns out that for a moment DEEP BLUE had crashed.
Mr. Seirawan Lost consciousness.
Mr. Ashley However you put it. It is something that doesn't happen too infrequently. But at the moment it did happen a lot of operators you saw came over the board and people were talking about it and trying to get it back together and everything back to position and that did happen and everything we would think would be okay. But Garry Kasparov was very disturbed by it.
In the middle of this attack that he was doing he felt he had to make a critical decision about 2 different ways to go, 1 a solid end game, another to continue with a daring attack. And he had actually already moved. It was the computer's turn to move. But he felt that he was so inclined in the position to be analyzing all these wild and crazy possibilities that at the moment all the operators came over he was really disturbed by it. He made a big stink about it back there. He was very upset and most times we've seen this happen so far, he's been cool about it. He's felt like okay no problem and let's just play chess. But this time he was really, really disturbed, as if the collection of the technical problems that have occurred has gotten to him. So he really made a bad stink about it, but then he calmed down and he's back at the chessboard.
Mr. Benjamin That's very understandable and we certainly apologize for the interruption in his concentration but all I can say is that, you know, that the team would really prefer if we didn't have these technical problems and, you know, we never had these crashes, but, you know, at this point we just can never be sure that they're not going to happen. And, you know, it's unfortunate but it's a little bit out of our control.
Mr. Ashley Well, Kasparov seems to have settled back into the game. I believe Yasser Seirawan, my colleague, will take a break at the moment.
(Applause.) Mr. Seirawan I'm going to be leaving you in 2 very capable hands. Treat them kindly. 25. Re4.
Mr. Ashley So this last move, after the last series of moves after knight takes E8 by DEEP BLUE Garry Kasparov has played a very intriguing move, Joel, not your normal walk in the park move. Rook to E4 seems very aggressively inclined as Garry Kasparov, with this last move, instead of playing just a simple recapture, what's happening with the nuances of this last move?
Mr. Benjamin It looks like an excellent move to me actually because with rook E4 he's preparing to, as I see, it simply recapture on C4 with the rook. And there he would -- he would bear down on the C6 pawn with both rook and queen. Also the queen would stay on C6 where it's well placed, guarding the White B2 pawn and also eying the sensitive G7 square. 25. ... Nf6
26. Rxc4
Mr. Ashley Well, DEEP BLUE has responded with the move knight to F6, attacking the rook. The possibility did exist of knight to D6, but as you said, rook to G4 would have been very strong, would have been very aggressive knight to F6 has been laid. DEEP BLUE with 5959 left. Garry Kasparov away from the board at the moment. He's coming back now to the position they have played only 25 moves. Garry Kasparov quickly, instantly respondeding with with rook takes C4 and now a very serious threat of rook takes C6 in the position. Fritz 4 is giving Garry Kasparov a point 22 advantage, which is very slight from its point of view, but it looks as if Garry Kasparov's pawn structure is superior to DEEP BLUE's in this position, Joel.
Mr. Seirawan Yeah.
Mr. Benjamin Yeah, it looks like Black is in real danger of losing that C6 pawn, and I don't know if DEEP BLUE would even try to defend it in this position with rook A6. It looks like in the long run that's not going to work. Perhaps we will find some way to jetison the pawn to get some kind of counterplay. It will do that sometimes. People have this notion that computers will just grimly hold on to their material because they never like to give up a pawn. But computers can often surprise you because since they have such excellent search capabilities, they can often find rather hidden prospects for counterplay by sacrificing material. So let's see if DEEP BLUE -- I think this is an important point in the game because White seems to be winning a pawn and Black has a moment. He has a move and then White's move is rook takes C6, and then he gets another move. So he's got to find something to do in that span. 26. ... Nd5.
Mr. Ashley Looking at this position, DEEP BLUE has responded with a very aggressive response indeed, knight to D5, after rook takes C4, knight to D5 attacking the White queen, and it doesn't seem as if DEEP BLUE is ready to give away that C6 pawn just yet by centralizing the knight and attacking White's queen. White's queen has to move off that aggressive C3 square and the C6 pawn will be safe at least for the moment.
Mr. Benjamin Yeah, that's a very logical move. The queen is forced away from its attack of the C6 pawn and the good thing is that the knight on D5, it's still in a position to scoot back to F6 for defensive purposes.
For instance, if White were to play queen to D3, he's not really threatened to push the pawn to F6 and set up a mate threat because Black could simply come back with a knight and take it and defend the F7 square. So the knight is pretty solidly placed on D5 for both attack and defense, and I think that DEEP BLUE has perhaps improved its chances over what it had in the middle game.
Mr. Ashley Well, certainly this knight on D5 is a very strong looking piece with many opportunities to control key squares. In addition to which, Joel, I'm wondering about DEEP BLUE's chances of counterplay. The White king seems to be safe for now, although that diagonal has been opened up towards it, but also what about that B2 pawn? It looks like maybe there might be some tactical opportunities for Black to try to make White pay attention, just a couple of ideas that might happen in the game.
Mr. Benjamin Yeah, you've hit the nail on the head of where Black's possibilities for counterplay come from. 1 is the B2 pawn, which is on a half open file for Black. So Black could bring his rook or queen to bear on that pawn. And also, perhaps more importantly, is the fact that White's king is a bit airy. It's on an open diagonal. And while it's not vulnerable to a mating attack, it does offer Black some chances for counterplay, perhaps a perpetual check at some point or simply a timely check with which helps Black get his counterplay going. 27. Qe5.
Mr. Ashley Well, Garry Kasparov has attempted to deal with at least 1 of those problems by playing his queen to the centralizing square E5. Garry Kasparov has posted his queen in the middle of the board, but right now 1 is not sure exactly where Garry Kasparov is going to aim his big guns. The queen is well centralized, but it may be vulnerable to a Black rook landing on the E8 square chasing the queen away and Black may indeed get some counterplay going and looking at Fritz 4's evaluation, it thinks that it's Black who has a slight advantage after all those exchanges. And that seems like a surprising evaluation.
Mr. Benjamin Well, Maurice, you made a very good point about this position. White's pieces have a great deal of flexibility. The rook on C4 bears down on the C6 pawn. It can also come to G4 to threaten G7. The bishop on C2 can go to A4 and attack the weak C6 pawn. But the point is that if White plays these moves he loses something. If the rook moves off the C file, then the pressure is taken off. Or if the bishop goes to A4, then White loses possibilities along the diagonal that the bishop is on now. And also the F5 pawn might become unprotected. So the point is that it's not so easy for White to press his position without giving Black something in return. The most obvious move to me at this point for Black is queen to D7, preparing to take the E file with rook to E8. 27. ... Qd7.
Mr. Ashley Well, Garry Kasparov has gotten up from the board again. I wanted to make a point about the time situation. Garry Kasparov has played 27 moves so far. That means he has 13 moves left. And DEEP BLUE concurs with you, Joel, it has responded with the move queen to . So we know you've been with your baby a lot. So you're predicting some of its moves there.
Mr. Benjamin I'm still not very good at it.
Mr. Ashley Queen D7 has been played and the rook on A8 ready to swing into the game moved for the first time after 27 moves to the square E8 and chase the white queen out of position. I've got to wonder if Garry Kasparov can get his emotions in check if he is indeed -- if he was indeed upset by that little disturbance which he full well knew could happen. It happened in prior games, whether or not he's going to play too quickly and end up throwing away the better part of his advantage and maybe even losing the game.
Mr. Benjamin I think he's definitely uncomfortable now because he has to feel that he had built up a very promising attacking position and he just either didn't see how to follow through with it or he didn't want to risk getting to time pressure by looking too deeply and using too much time
Mr. Benjamin At this point I think he feels a little pressure. He might still have the advantage, but he's down almost to 32 minutes. He's 19 minutes behind on the clock. Actually, even more than that. And against a human opponent he probably wouldn't worry about that. But he knows that if he's unlucky enough to get low on time and to play moves that the computer is expecting, it can shoot right back with an instance calculation and it could cause great problems for him. So I think if he had an hour on his clock he probably would be very comfortable but perhaps not here. 27. Rg4.
Mr. Ashley Well, he has quickly responded with the move rook to G4. And Kasparov as well as anyone knows that the timing -- that time can play a huge factor in these games and just the amount of time you get for each decision especially in a complicated position could have a bearing on how good, how well you play. And how good your moves are.
So Kasparov quickly playing rook to G4. We talk about the average of 3 minutes a move because it's 40 moves in two hours. So 3 minutes a move. And I always pay attention to that when I'm playing a chess game. If I have ten moves left, I like to have an hour on my clock. If I have 20 minutes I can handle that too, but it's always nice when you have that extra half hour.
Right now Kasparov has 31 minutes and 41 seconds to play his remaining 12 moves.
Mr. Ashley So, okay. Garry Kasparov with that last move, Joel, rook to G4, has initiated an old cheapo threat of mate on G7. I would suspect that DEEP BLUE saw that in less time than it takes for me to blink. DEEP BLUE is now wondering what to do about this threat, and it looks as if the only move is F6.
Mr. Benjamin Well, DEEP BLUE knows what to do, but unfortunately it's thinking about it anyway. It will play F6. It's the only move.
But I have to say I'm not really that impressed with rook G4. I don't think that White has a kingside attack in this position. I think there's not enough pieces to crash through with it. And after Black plays F6, that guards against its threat.
And now, this is just what I said before. The C6 pawn is no longer attacked, so Black can at least breathe easier there that he's not going to lose a pawn in the near future. And I think that Black is still defending this position. 28. ... f6
Mr. Ashley And I beg to differ with the gentleman, 1 of our spectators, who felt that -- I overheard say that he thinks Garry Kasparov looks relaxed. From my experience with Garry Kasparov, the way he was shaking a bit in the chair before he made his last move, it seems to me that he's a bit more anxious. 29. Qd4
F6 has in fact been played by DEEP BLUE, but it seems to me that Garry Kasparov is a little anxious here, You know. And quickly he's played queen to D4. Garry Kasparov not wanting to think too long about his moves, just trusting in his intuition, what he feels walking into the position.
But 1 thing, 1 thing, I would say now -- Garry Kasparov even getting up from the chessboard -- in this volatile position many things could happen.
But 1 thing I should say. You know, you talked about that. Garry Kasparov has played so many marvelous games, textbook games, games that will live in chess history for all time. The guy has recorded so many novelties, so many dashing attacks, crushing positions, that we can read over his games and learn so much from them. And we'll do that for years to come. He will be known as 1 of the greatest chess players in the history of the game, if not the greatest.
But 1 thing you find about Garry Kasparov's games, the great games he plays, is the sort of inevitabilities about his win. He starts playing and playing, and first he gets the good opening. He throws out the great novelty. Right? So he's played this wonderful new move. His opponent gets rocked trying to figure out what to do. The person tries to defend. Garry Kasparov builds up a big position. The person tries to defend some more. Garry Kasparov sacrifices a piece. The kingside is shattered. Here comes Garry. Might sacrifice another piece. Check, check check. A few checks, and finally a beautiful mate to crown the whole thing off.
And we say, wow, Garry Kasparov. You know, typical Garry. That has been like his calling card for so many years. And if not that way, then you might go into some end game where he's completely lost and he beats you trivially. This position, it seems as if he was starting to do that. I'm sure against a human we'd be thinking you're in deep trouble, man, you've got to be suffering. But now the whole game has changed complexion.
And we've seen this happen to Garry Kasparov before, in games with Karpov, where he got a big position, it looked like he was doing great, and Karpov would make some ingenious defensive moves, and suddenly there was no real serious attack and, then Garry had to correct himself and replay the chess position. It seems like that's happened here in a really serious way.
And what are your thoughts on Garry being able to adjust himself and play good chess moves?
Mr. Benjamin I think you made a good point to start that. Garry's games often do have this aura of inevitability to them. And I think actually that he doesn't feel like that here, that during the games he actually has doubts as to whether he's going to bring home the attack. And I think that's probably an unusual feeling for him. I think that against almost any human opponent he feels that he's going to bring the point home when he builds up a strong forceful position. 29. ... Kh7
Mr. Ashley Well, DEEP BLUE has played a move that is that, to my eye, seems a bit surprising. I was anticipating, as you said, bringing a rook into the game, maybe to E8. It looked like a strong open file. But instead it has played this move king to H7 in response to Garry Kasparov's queen to D4. What's this move about? This move seems a bit strange.
Mr. Benjamin Well, it's not so strange because the king is clearly not well placed on G8, for 2 reasons. 1, the rook is on G4, putting pressure there and creating a pin. So that the F6 pawn could be taken if the Black knight on D5 should be neutralized or should move away. Also, White had a potential pin on the A2 G8 diagonal with bishop to B3. And there again the Black knight would be neutralized. So by getting the king to H7, Black frees up his position a little bit. He still has the option of what square he wants to bring the rook to.
And I have to say that I don't see the big danger in this position for Black. I think -- you know, I'm a little uncomfortable with the pawn structure on the queenside, but I think there's enough going on that it's not easy for White to exploit that and stamp out the Black counterplay. 30. Re4 .
Mr. Ashley And I think that Garry Kasparov himself realizes it because the way the pieces have sort of lost touch with each other, you remember the position where the 2 rooks were doubled on the E line. The knight stood on E5, queen on C3, bishop on C2. They were all harmonious. The picture of harmony, in fact.
And over the last 7 or 8 moves suddenly a lot of the pieces went off, the rook that was so wonderful on the E line is now over on the G file wondering could we get some mating attack going. The queen is on D4 and the bishop still remains on C2, but they're not as connected. And that bishop on C2 has to be so upset at that pawn on F5 that White pawn on F5 standing in its way because it's going to restrict that bishop somewhat and that pawn might even become a target in some positions.
But right now Garry Kasparov's pieces, not the same kind of logical flow as 1 would have expected from the prior position. That is not to say that Garry Kasparov has a bad position, but it's certainly not the same kind of position he had before.
He has now responded with the move rook to E4 and that has taken over the E line, which was a very important factor in the position. And that looks like a strong move to me, Joel.
There is a little trick, a very little trick. The pawn on F5, which seems to be hanging at the moment, indeed is not because the queen and king would be on the same diagonal. The Black queen and the Black king would be on the same diagonal. And after the very simple move rook to E2, the rook would defend the bishop and the queen would be gone.
So that's a little trick in the position. In the meantime it seems as if Garry Kasparov will be able to post his rook on a great square E6, and that could create some potential problems for Black.
Mr. Benjamin Yeah. That points up the 1 negative of Black's last move, is that it put its king on the same diagonal as the White bishop. So if Black were to oppose rooks with rook to E8, then White would simply play rook to E6. And the rook could not be taken because when White captures F5, takes E6, hitting the queen, it's also check.
Mr. Ashley Let's take a look at that because that is a very revealing variation. Rook to E8, the most natural move, White could post its rook on a very solid square, E6, because rook takes E6 for Black would be a huge blunder because of F takes E6 check. The bishop on C2 suddenly awakens and the Black queen is history because that tiny pawn on E6 is going to be a voracious monster after taking the queen. So this would be quite devastating.
In fact, Fritz is so happy about this position it's already calculated mate in 6. So we would not anticipate DEEP BLUE allowing this to happen. But that move king to F7, which looked a bit bizarre, has proven to have 1 major flaw in this variation. It looked -- you know, it looked a bit weird. It did have its points, but it also had its minuses. And this particular minus has occurred right now.
Mr. Benjamin Yeah. But before we get too excited, the E file is not the only open file on the board and DEEP BLUE simply felt that it wasn't that necessary to occupy the E file. And perhaps it doesn't want to exchange rooks and that's why it didn't bring its rook to the E file in the first place because it didn't want to exchange rooks. 30. ... Rd8.
Mr. Ashley DEEP BLUE has played rook to D8 instead of rook to E8. And as you mentioned the D file is also a strong file.
Mr. Benjamin The D file is a good file as well. And rook to B8 would have made sense as well because Black could have followed up with queen B7 and tried to hammer on the B2 pawn. So it still seems to be the Black position is fairly solid. I don't see anything concrete for White just yet.
Mr. Ashley Moving for the first time on move 30 is that rook and it seems as if it has connected all the Black pieces. We talk about harmony a lot in chess. Pieces need to coordinate with 1 another. They need to come together and play together.
Chess is really a quintessential team sport. The pieces have to get together and have concrete task as units together. No fighting on wrong sides of the board. Okay. I send my rook over there and my other rook over there and my other bishop in another corner and hope that they're going to work together. That's not how it works. You have to bring everybody focused on 1 particular point. It's usually the best thing. You have 3 pieces aimed at a particular point. The opponent usually cracks under that kind of heavy firepower.
So right now you see the rook on D8 for Black, the queen on D7 and the knight on D5, and even the tiny pawn on C6 supporting that knight on D5. Everything about Black's position says there's some harmony. Of course, there are pawn weaknesses. Pawn on the A5 could become a factor. But Black is slowly but surely improving its position.
Mr. Ashley Garry Kasparov now considering his next move. With 26 minutes left on his clock to make ten moves, he is thinking about the position.
And Joel maybe you can talk to the human feeling that I usually get with a situation like this when I had a great position. It was looking like oh, man this guy is out and suddenly things don't work out as planned. What is the feeling you have as a professional grandmaster when that happens and how you try to continue to play the game
Mr. Benjamin Well, actually it just makes me absolutely crazy. That's the big problem with chess, is that it's extremely nerve-racking for most human beings, and that's why I'm very happy to see DEEP BLUE playing the moves instead of me, because I think with either side I would have all kinds of thoughts running through my head. You know, maybe I missed something. You know, sometimes as a chess player, you even start to calculate something that has already happened. It's silly during the game, of course, but sometimes you can't help it. Did I just miss a win? Should I have played that? Sometimes you cannot avoid that. It is very difficult.
Mr. Ashley Kasparov will have to rein in his emotions in this position and try to do his best. He still does not have a bad position. He looks pretty good. He does not have what he really wanted.
Thank you, grandmaster Joel Benjamin. We will see him no doubt as this match continues in later games. He has to get back to the control room and hang out with his team. Probably the operators in that room are thinking where is Joel, where is Joel? Are we okay or what? They know you have been doing the commentary up here.
I would very much like to welcome back my colleague and friend, Yasser Seirawan. What is your feeling here? What is your feeling about this position?
Mr. Seirawan I definitely like Garry's position. What he has done in game 2 convinced him that what he should be trying to do is get into nice end games where there are a lot of things going on and try to play with the computer, try to take advantage of it rather -- I don't know, its propensity to accept weak pawns.
What we see specifically that Garry has done, which I really love, because as you very well know, Maurice, I like to take advantage of my opponent's weak pawn structure, and DEEP BLUE's pawns make ideal targets. The only weaknesses in White's position are the pawn on B2 and the pawn on F5. Both of these pawns are very nicely covered by the queen on D4 and the bishop on C2.
So the natural move for Garry in this position -- and I'm sure he is licking his chops with the idea of rook E6. I think this is probably the move that Fritz would also suggest. I don't like queen F2, so I would select rook E6. That is very simple.
What we are going to do is gang up on that pawn on C6. And if I can get one more move in, in Garry's words, one more tempo, queen C-5, that would really be the ideal because then the queen would be attacking both the C6 and the A5 pawn. So rook E6 is a devilish little move, putting a little pressure, more pressure on Black's position, and I think that's the way for Garry to go.
Mr. Ashley Basketball players certainly understand the importance of big man posted up low, and that would start looking like Shaquille O'Neill aggressing that pawn on C6, as you said. Deep penetration is one of the things that we like in a chess game. When high pieces get down to the other side, it is a sure sign that you are starting to get control of the position. We have to consider possibilities here for DEEP BLUE. B2 pawn seems like one thing that DEEP BLUE is going to have to try to take advantage of. The knight temporarilyy at least on D5 very secure, so maybe a move like queen to B7 would have to be considered in this position. I like queen B7, but I like the stable advantages you are talking about. The stable advantage of the strong pawn, better pawn indeed than the Black ones.
You explained before about pawn islands yesterday. Maybe you want to talk about that today, the importance of having fewer pawn islands than your opponent.
Mr. Seirawan This again is the classical knowledge that generations and generations of grandmasters have handed down to their preceding generations. Keep in mind that chess is thousands of years old. It was played in ancient Egypt and in ancient Greece and it evolved to the game of shattaroonga (phonetic spelling) today. So the game of chess that we know today is many years old. Capablanca, one of the greatest players of the game from from Cuba, explained pawn islands and he said it very succinctly. The more pawn islands that you have, the worse your end game is going to be. Pawn islands are groups of pawns that stand by one another or stand alone. Height has two pawn islands. On the King side he has his F5 pawn, the G2 pawn and the H3 pawn. This is a single group of pawn islands because they are on adjacent files. On the queen side he has A2 and B2 pawns that compose his second pawn island. On blockade of the board however he has three. The pawn on C6 is what we would call an Isolani, an Isolani pawn. And the pawn on A5 likewise an Isolani or a split pawn.
So basically black has three pawn islands and white has two. This is very, very crucial as you get into end game strategy that those pawn islands will become weaker and weaker. And I know Dr. Hans Berliner had his hand up. 31. Kh1.
MR. RIGHT Before he does so. Kasparov has employed a somewhat strange move. He has moved his King off the G1 square and has put it on H1. I call it strange not because of the positional factors because I can see the point of moving the queen off the long diagonal where the -- the King, I'm sorry -- off the long diagonal where White's queen guards from any funny checks, but it seems strange for the timing of it. It did not seem that it was quite necessity moment to play, but Kasparov has seen some variation where that King move did have significance.
Mr. Berliner make your point, please.
(Audience commentary. ) 31. ... Qc7
32. Qf2.
Mr. Ashley Interesting point, Hans Berliner. In the meantime we do have a move played by the computer and Kasparov. Queen C7 getting off the D pawn, sorry off the F pawn. Pawn on F5. Queen C7 moving the queen off the D line and potential threat of moving the knight to some square. It was not clearly aware, but Kasparov quickly responded with king to F2, queen dropping back to F2 and staying on that long diagonal. Keeping many of its pluses in that position.
Does this seem strange to you, Yas?
Mr. Seirawan Yes, indeed. I like that posting of the big player down low, and I like the move rook E6 and I especially liked it in combination with the move queen C5. Again hitting the pawn weaknesses directly.
Garry's last two moves, queen H1 and queen F2 I think have been affected by his understanding of how to play against the computer. In fact, since game one he has not tackled the computer directly. He is trying to mix it up, if you will. The last two moves, queen H1, a highly refined defensive move, very unusual for Garry. I don't quite understand. I don't grasp the point. I'm sure if Garry will be kind enough to share his thoughts about that when he comes out. And also this last move, queen F2, it was clear that white's queen needed to get off the B4 square, but again a move like queen C5 might have been more aggressive. One point that we have not quite explored is that G6 square. Again, as pawns move forward, it can never move back and as pawns move forward they leave weak squares behind. The pawn on F6 and the pawn on H6 no longer protect that nice little square on G6, and if Garry could ever land his queen with queen G6 check, it would be a real thud. That would be a real knock out punch.
So it is something that he may want to have explored with the move queen F2, aiming in some long range plan of trying to bring his queen to the G. 32. ... Qb8
Mr. Ashley Now he has played the move queen to B8. I was just checking out because it seems so strange that Kasparov should be so much on the defensive. When the more aggressive possibilities seem to present themselves.
Let's go back to the moment when he played the move queen to H1. And let's take a look at the possibility of rook to E6. Now, I don't know if this is the most representative variation but it seems to me that possible is queen C7 and maybe if queen C5 in this given position, queen C5 in this given position, Kasparov had anticipated the move knight to F4, and this creates some dynamic possibilities for black. I'm thinking maybe a move like knight takes H3. What if rook takes C6 in this position. I was thinking maybe knight takes H3 check was possible.
Mr. Ashley Maybe that's why he made those two safe moves where he did anything aggressive.
Mr. Seirawan A very good point, Maurice. And there indeed you may have the answer as to why Garry chose this refined defensive move because of some variation exactly as you suggested, the knight sacrificing itself. It is kind of interesting if we were to reverse roles for a moment and put DEEP BLUE in Kasparov's shoes and see Kasparov sacrificing, we would know he is doing it again as he did in game 1. 33. Ba4 c5.
Mr. Ashley Dramatic changes are occurring on the chess board. Kasparov has played the move bishop to A4, attacking the C6 pawn and that C6 pawn seemed to be quite happy on the square C6 as it was defending the black knight, giving good support to that knight, but DEEP BLUE has instantly responded with the charge C6 to C5, and this seemed for a moment like maybe it was a sacrifice of a pawn, but apparently DEEP BLUE has contemplated it all down to the end.
Queen takes C5 seems like the most obvious move. But I think queen takes C2 and black will be doing fine. The active black queen, I'm not sure if black is doing fine. It would take some analysis because it is the pawn on A5 hanging, but this is the type of thing that as Grandmaster I would not want to do if I was white. I would not want to exchange my good B2 pawn for that isolated split C5 pawn. But what I would like to do, and again that pawn on C --
Mr. Ashley Had one good thing going for DEEP BLUE. It supported the knight on D5, so that knight on D5 has now been a little bit weakened and what I would like to do is play bishop to C6, just preparing to capture on C5. One idea is if black rise to chase the bishop away with queen D6, which I grant you would be a bad move, there is a possibility of playing bishop to E8 in order to try to exploit that G6 square once again.
We could bring the bishop back to the G6 square and kind of snuggle up to that black king and say look who is coming to dinner. But I do like the move bishop C6 for Kasparov at this moment.
And I would also like to point out the time.
Mr. Ashley Very significant in fact. The time left Kasparov with only 17 minutes and 22 seconds and counting down for his remaining moves and he is aware that the game has changed complex, at least a shift in the wind. I want to say dramatically, but it is at least a significant shift here because now the position is getting sharp.
Now it is getting a little wild. If you take my pawn, I have some tactical shot. Do you want to play safe and continue to play safe? But I'm making my moves instantaneously so I know these moves are okay. These kinds of thoughts have to be going through Kasparov's mind. Am I all right? Is this a good position? Have I blown it?
Mr. Ashley I should make sure I'm drawing at least. There should be some variation, Kasparov here shaking his head looking as if he was mumbling to himself which is something that he does when he is not happy with the way the game has changed direction. I have very rarely seen him get off late from start to finish, but when he has a good position and it turns to a bad position or a position that he thought he should have gotten because he should have done something better, the head starts shaking and he starts muttering a bit to himself. Human reaction of course. Very disappointed with the way he has handled this game.
(Audience commentary.)
Mr. Seirawan The question is, first of all, in this type of position, the immediate thing to look at is the most direct, so let's take a look at what the gentleman suggested.
Queen captures C5. Well, there is a pawn hanging in the position. What are the ramifications. Let's take it and take a look. After the queen takes C5, the suggestion was the move rook to C8. Do keep in mind at that Black's queen on B8 controls an extremely important square. Look at the diagonal all the way down to the H2 square. So what that allows is this rook to come down to C1 with check, but let's take a look at what happened. Let's continue. Queen takes D5, and I say thank you very much. You go rook C1 check, but it's not checkmate. I have the defensive play bishop to D1 and so I have one apiece, and that will win the game. That will bag the game for me.
So after queen takes C5, the move rook C8 would lose, but the problem that we think that Kasparov is annoyed with himself at, if we go back to the position at hand, after queen C5, we think that DEEP BLUE will pounce on that B2 pawn. Just queen takes B2, and we have a problem because the Black queen has become extremely active.
MR. RIGHT I was looking at this variation, one might think that the move queen takes E5 is possible in this position, and I'm not sure whether or not this move is right, but I'm going to venture a guess that maybe not at 4 in response to this move is actually playable. Is that a winning move here attacking the G2 square? And if the rook were to take that knight and queen C1 check picks off that rook with check and that's going to be an awesome attack, probably just completely winning.
So Kasparov cannot even nibble on some pawns because his head would be taken off immediately, so that has got to be something concerning the world champion, now down to only 13 minutes and 27 seconds on his clock to make his next moves. In a rich, sharp complicated position like this, Kasparov cannot be happy with having such limited time to think. He has to be looking for some variation that calming Black down. Obviously DEEP BLUE is starting to get busy. C5 was certainly unexpected, but it is showing his prowess. We can go back and forth, but I worked everything out in detail and Kasparov being very agitate at the moment. 1250 left on his clock and ticking down. And that time situation, DEEP BLUE moves so quickly. It has 38 minutes and 55 seconds left. No problem. Very annoying.
Mr. Seirawan I will turn to the audience and we invite them for comment.
(Audience commentary.)
Mr. Seirawan Many time champion of Pennsylvania, mike Shahady (phonetic spelling) making a suggestion. Mike's suggestion is what we have seen before, trying to exploit the G6 square, bishop to E8. This sets up a well known mating pattern, just so that we can discover the mating pattern. Mike discovered a reputation to his own idea. The problem in the position is that there is queen to B1 check. This is known as a double attack. The king in the corner on H1 and the rook on E4. Sorry Mike, you are going to have to do better than that.
Mr. Ashley Let us get back to the position, instead of this move, Hans Berliner has a comment.
(Audience commentary.)
Mr. Ashley He wants to play the move queen to E1 with the idea of playing bishop to E8.
Mr. Seirawan Let's take a look. Queen E1.
Mr. Ashley What happens if queen takes E2, bishop to E8 could be played. And this is a position of continuation that DEEP BLUE would have less respect for unless it was being mated, so one has to wonder. But this move, very nice tactical possibility, but maybe only leading to a draw. I don't think this move really works as well in this position because of maybe rook takes F4.
Mr. Seirawan The element with knight F4 is there always is this back rank check. The queen on B2 is very aggressive. So for example, now, if white was to play rook takes F4, there is the move rooks takes E8, capturing the bishop, queen captures E8, and for a moment white is a whole rook up, but then there is this cold shower, queen takes C1 check, and once again we have seen this kind of tactical motif where Black picks up his rook and gets the F pawn too.
Again, this is a type of thing that the computer does so very well. It completes all these tactical variations and I suspect has done so many moves ago.
Mr. Ashley So many brutal variations. We talk about it, if you look at the moves we are suggesting, these unusual moves, but more like forcing moves I should call them, take on C5, take on B2. What happens then? The computer excels at that. DEEP BLUE is a super Grandmaster in coming up in variations like that. So those concrete calculations Kasparov is well aware that the computer does a better job of it. Too many busy things are happening on this chessboard. Kasparov wants to make a move that will make simplified exchanges. He has to do that very quickly. However, he has still been thinking of this last move C5. If you had completed it before, Yas, it is the kind of move you like to play as Black because your opponent is in some deep time pressure. And the person has to calculate it to the end.
Look at Kasparov now, his fingers on his forehead trying to figure out all the details of this move, and he he has been ahead of us certainly in calculating these variations, but he is noticing them on every variation he looks at, good for Black, good for black, good for black. He is concerned now about what move he should make. He is spending this time after this spectacular move C5.
And look at his time. We mentioned a second ago he has 12 minutes, but he has only 8 minutes and 24 seconds left. That is a dramatic change in the situation and the problem we spoke about on many occasions, Yas, when you have a little bit of time on the computer, your propensity to make errors increases and the less time you have it increases dramatically and we could see a horrible blunder by Kasparov and he loses immediately.
Mr. Seirawan That is the whole problem with the human condition. We get nervous and we get more and more nervous with -- we have seen two moves, so I will stop my thought there. 34. Bc6 c4.
Mr. Ashley Look at this. Kasparov after all of that thought, six or seven minutes, came up with bishop C6 which you had suggested before and instantly two seconds elapsed before the move C4 by Black was played so now Kasparov has to be a little concerned. He is in heavy, heavy time pressure with seven minutes left and he is pointing to the C-1 square. He put his finger on the square C1. We have seen the Black queen here on the C1 square on several occasions in our calculation and Kasparov did the unusual thing of actually point to C1. And look at this, Kasparov whipped off the C-4 pawn. This pawn move was a sacrifice by DEEP BLUE and Kasparov has accepted the sacrifice. He decided that he gets a pawn. He doesn't see why it is bad. He doesn't see a mistake in taking this pawn and looking at him myself, what is the problem?
Mr. Seirawan I'm not going to say anything because I think the computer does not, does not, I emphasize, blunder pawns. This was certainly prepared variation, and again let's just point out this that Black queen controls the H2 square so that a black rook could come down the file to D1. It would be very nice, so let's see the move knight E3. That is a very consequent move.
Mr. Ashley I would call it spectacular. It looks interesting, your point being that you want to play rook D1, and also the rook on C4 is hanging, so --
Mr. Seirawan The only move is rook C1. Now, we are likely to see -- there is the move knight D1 forking the queen on F2 and the pawn on B2 so that DEEP BLUE could win its pawn back. But DEEP BLUE might go for the attack itself and might try to centralize its queen with queen to E5, the intention being to pick up that F pawn as opposed to the D2 pawn.
In this position DEEP BLUE has pretty good coordination of its pieces, for example, rook to E1 and now a variation like rook D1. So that if rook captures D1, knight captures E1, black has very solid pieces, the queen on E5 is a giant in the position, very nicely centralized and this pawn on B2 is going to be grabbed next.
Mr. Ashley I'm impressed by Kasparov's audacity. And you know it sees 200 million positions a second and Kasparov looked and looked and it took him all of 30 second to say what about it? I'm ready to take your pawn. This is a dare. This is a bluff. I know you see all, so I hear. Show me. Well, let's see. The computer certainly has not responded to his move instantly, which is a good thing, but it instead played a very interesting move, the move knight to B4, another way of sacrificing this knight. The point, that the knight can move on several different squares because the move rook to D1 check, Black rook pounding on D1 square is threatened. That move is definitely threatened. And since that is a threat in the position, Kasparov does not have time to just pick off that knight on B4 and say I have it, no problem. So Kasparov has to deal with the fact that rook to D1 check would force him to lose his queen in the way of G1 and he would lose his queen. So dealing with that will be his first concern.
In the meantime the knight on B4 is attacking the bishop on C6. So that is another cause for concern for Kasparov, which means that rook to C1 now would be a huge blunder. Then this rook would be overworked trying to guard D1 and the C6 bishop.
Mr. Ashley That is not a move and we have talked about that move before, the move rook to C1 defending nicely it seemed, but now the rook finds itself tied down to its bishop. So a deep tactic by DEEP BLUE. Kasparov apparently overlooked this possibility or maybe had not studied it carefully enough.
He has five minutes and 43 seconds left to make these five moves, and he has to be careful because he knows full well that this thing sees some sharp possibilities that one mistake and the game ends. This is precisely the kind of position he has. Garry did not want to be in this position.
Mr. Seirawan Especially at this time of the game. The human sitting there, he comes to the game or she comes to the game full of verve and fighting trim and then after three or four hours of exhaustive analysis starts to see little stars under those bright lights on the stage. So the last thing that Kasparov wanted was a very sharp technical position as we are seeing later in the game as he may be tired and prone to miss a tactic.
As you also pointed out, Maurice, Garry is facing a couple of threats. The knight certainly hitting the bishop on C6.
One idea is to draw the bishop back. For example, bishop A4, however, runs into a problem. There is knight to D3. 36. Bf3
Mr. Ashley Kasparov played the other bishop move, bishop to F3, but this seems to allow, Yas, a knight to plant itself on the D3 square and that seems like a strong possibility in the position.
Mr. Seirawan Indeed. And that is most certainly what DEEP BLUE will do is bring his might to this beautiful post. It would force the queen on F2 as well as the pawn on B2 and allow for further penetration with the black queen, so Garry has definitely been thrown on the defensive in this last few sequences of moves.
Mr. Ashley Shaking his head is the world champion. He has to make sure there are no real dangers presented in this position. He is certainly concerned. He knows that the time pressure is on. He is waiting now as we are for move 36 by DEEP BLUE, but, you know, looking at that clock and DEEP BLUE has 35 minutes left, so brutal. We talked about how quickly it moves. And we just know that inside of two minutes DEEP BLUE is going to play a move and Garry knows it, so he has to be ready for all eventualities.
So the next question is can DEEP BLUE come up with anything spectacular. Knight to D3 looks very obvious and what will be Garry's best move? Can we anticipate what Garry will do in the event of this move, knight to D3.
One of the things that is very, very human is when you are under attack and you see yourself facing a host of threats, you like to counter as quickly as you can. And again we have spoken about this G6 square. It is not as weak as it has been, but it is the move queen H4, outlining an idea, for example, like queen takes B2, bishop H5 trying to grope for the bishop G6 possibility. I'm not very happy about it myself.
So we may see something a little more defensive minded. For example, like queen C2. I think this would be a little bit better. Again, there are just so many possibilities suddenly for DEEP BLUE. There is knight E5 and knight E1. There is even a possibility of queen G3 intending to play knight F2 check, so Garry is under enormous pressure here. He won't have an easy decision to make. Three more moves I believe? 36. ... Nd3.
37. Qh4
Mr. Ashley I think it is four more and finally knight D3 has come. If it were humans, Garry would have been in trouble because knight D3 would have come a long time ago. But DEEP BLUE has finally played this move and Kasparov has gone over to that side of the board with queen to H4. He had to prevent the possibility of king to G3, so that move as defensive as it is aggressive, but now DEEP BLUE with the opportunity to rip-off the pawn on B2 and equalize the material and now as we look at this position, Yas, there are a number of things about it that truly favor black.
Black has limited number of weak pawns, certainly those pawn identifies no longer exist. It seems temporarily at least that black's king is safer than white's. So a number of things have changed about this position. Garry has to be extremely concerned and unhappy with what he has done. And now will he be able to avoid the blunder? I say this again. I can stress this. It could easily happen that Garry makes a move that suddenly just loses to some weird tactical variation where Fritz starts going off the scale. This wins. You are over.
The time pressure is the most significant factor. He will not be able to check all the variations. I think it is interesting. We spoke about DEEP BLUE spending time over obvious moves. Garry is lucky he is not playing a human in some level. Because that would have been played by a human being instantly. And the time pressure would have been on. Right now Garry gets the time to take a look at queen B2 and says what is the best move I should make? There is also a double edged sword where that is concerned.
Sometimes when you are in a bad position and you have a lot of time pressure, the worst thing is when the guy stops to think for a while. Just move, just move. Just hurry up and move. And a great technique when you have somebody in time pressure in a tournament game is to not employ your obvious moves quickly. I have done this on so many occasions. I love it when this happens. I have a couple of moves, but even if I have an obvious move I will sit there and wait because the guy in time pressure is just sitting there.
Mr. Seirawan We have seen it.
Mr. Ashley So so now queen takes B2 has been played by DEEP BLUE. Kasparov's time at 415. There are different possibilities here. You are anticipating the move queen to G3, but the move knight 5 potentially could be a great response guarding the G6 square and forking some pieces, so it is not entirely clear what the best move is for Kasparov. And notice by the way, despite the fact that Kasparov was sitting there thinking as DEEP BLUE thought about the move, Kasparov still has not responded instantly. You think he has a move prepared, all that time that Kasparov gave him, but under this type of pressure you keep looking for better and better possibilities. Kasparov might have found a variation and was dissatisfied with it.
But now with only three minutes and 26 seconds left, Kasparov is in need of a good move and he knows that his pieces are not nearly as good as Black's. This is the kind of position -- and he has in fact played the move queen to G3. This is the kind of position that could easily be a draw, in fact. It could be much worse for white easily.
Mr. Seirawan I have noticed that if you take a look at the scale, the graph here, the very nice pretty graph, we see all of this green. This green all represents an advantage for Kasparov. Throughout the game, as you can see, Garry has been on top, but ever since he grabbed that pawn on C4, Fritz has turned and said, whoops, look at this. It is going right off the scale. In fact, after this move, queen G3, Fritz is giving DEEP BLUE a large advantage.
Mr. Ashley Fritz is giving it a 1.03 advantage in fact. To Fritz that means -- a point means I have calculated a concrete win of a pawn. That's what it means. It does not mean a positional advantage. It will not give you a point based on positional advantage. We can bet that Kasparov is certainly going to lose a pawn by force in this position, no doubt about it. Now, the question is, what pawn is it? We still do not move. The move -- excuse me. DEEP BLUE, instead of protecting the G6 square, has allowed Kasparov to cash in on what, to my eyes, is a magnificent square for the white queen. It has to make Kasparov feel a lot better that that is possible. 39. Rc7+
And instead, interesting though, instead of taking immediately -- going to G6 immediately, Kasparov has opted for the move rook to C7. I am very surprised that DEEP BLUE played this variation. If he takes A3, this looks like a poison pawn. It seems that DEEP BLUE pieces are going to be pushed back very quickly after this last move. That is a very surprising move.
Mr. Seirawan What Fritz hadexpected was knight E5. But first of all, the queen and rook, although they act in concert to create this nice threat against the G7 pawn, even if Fritz -- if DEEP BLUE goes on the defensive with the move queen F8 he has covered all the threats and his very next move is going to be knight E5. And well, he does remain a pawn up and that's exactly what is going on. DEEP BLUE has won a pawn. Do you feel that the active pieces of white though will compensation for the pawn deficit?
Mr. Seirawan Well, it better. It sure better because one thing is for certain, if they don't, Garry will lose the game because he is a full pawn down and he now has to rely upon the attack for giving him compensation.
Mr. Ashley Did we or did we not see this before? Is this or is this not shades of game 1? The computer took a pawn that we said was way too far, way too far from the main theater. Why were you going to give Kasparov, the great attacker, all of those great possibilities? And the computer said you can have those attacking possibilities and brought its pieces back near the king. And it won the game. It is the exact same scenario being set up here, taking a pawn that seems ludicrous to capture, miles away of the main theater of the action, and now maybe, maybe the computer actually seeing further than all of us. And I wouldn't doubt that it has, and said there is no problem here, I'm okay. As a chess player, as a guy who likes to sacrifice pawns, I want to feel that Kasparov has sufficient compensation in his aggressively placed pictures and that might still prove itself to happen in this game.
Mr. Seirawan As a guy who likes to take pawns, I'm ahead, and that's the way DEEP BLUE certainly feels. 39. ... Qf8.
Mr. Ashley Queen to F8 has in fact been played, defending the sensitive G7 square, and Kasparov with 253 remaining for his final move. This is the 40th move, and he needs to make a good decision here. 238 remaining on the clock, and he needs to make a good chess move. 40. Ra7 .
You know, the 40th move is a critical time. When you make that 40th move, you feel like I've done it. You get the sense of relief. Kasparov has put his hand up off the chessboard and he made the move rook to A7. He is getting behind that pawn on A5. And now again we wonder what is DEEP BLUE going to do about this aggressive posture? It seems to me that Garry might win his pawn back.
Mr. Seirawan Garry has a very good plan. He just got behind the A5 pawn and his intent is very simple, to pick up the pawn with rook takes A5 on the next move.
But this is one of the advantages of the knight. The knight over the bishop is one of the most complex relationships in chess because they both have the same value of three pawns. A knight is worth three and a bishop is worth three, but in certain positions the bishop dominates the knight and in other positions the knight dominates the bishop.
One of the things that is very strong for a knight is when in an end game all the pawns are on one side of the board. So after a variation, for example, like knight E5, rook takes A5, just winning back the pawn, this bishop is a little bit useless, and Black's pieces could coordinate very well in combining for an attack. It may not be enough to win the game, but in my view it will give Black an advantage.
For example, he could continue with the move rook D7, just preparing to bring his queen out to this square, queen out to B4, and sometimes just coming down on the back rank.
Mr. Ashley Rook to D3 looks like an interesting possibility to me. 40. ... Ne5.
Mr. Seirawan An excellent move in some cases to simply double the pawns. And we have seen knight E5.
Mr. Ashley Kasparov is away from the board.
Mr. Seirawan Probably recovering.
Mr. Ashley Yes.
Mr. Ashley He did get to the time control. He has to be feeling a sense of relief that he made it to time control but he has to be kicking himself.
Remember, if we go back 20 moves, in this game, around move 20, you remember Kasparov building up his position, beautiful pieces, it looks like Kasparov can pull the game through with strong attacking possibility, but the computer with great moves, moves that one might have suspected in a position with a computer, had seen several weaknesses. It simply dealt with those weaknesses and finally removed those queen side pawn weaknesses and has a good position. 41. Rxa5.
The bishop on N3 condemned to light squares. With pawns on one side of the board as you said the knight could become superior. In concert with the queen it could become very strong. Kasparov, rook takes A5. It seems that black just has a great game.
Mr. Seirawan Fortunately for Garry, he has reestablished material quality. In Grandmaster chess the advantage of a single pawn is certainly well enough for victory. So the very fact that he recaptured the pawn is a positive for him.
Mr. Ashley. He knows he is worse in this position.
Mr. Seirawan He is not offering any drawings that's for sure.
Mr. Ashley DEEP BLUE would not accept it anyway.
Mr. Seirawan What Garry has going for him is if he could reestablish his rook back on the 78th rank, the refreshment stall, he will have this nice attack on the move here.
I expect that DEEP BLUE might just play a move like queen E7 or even rook D7 or as you said, Maurice, rook D3.
But I think the important thing, first of all, before DEEP BLUE goes on the attack is to wipe out any chance of white getting an active rook. If we put queen E7 on the board for just a moment, what black can do to win this game is to create a dark square strategy going after white's king. And the knight on E5 can be very effective. 41. ... Qf7.
We have seen not queen E7 but the move queen F7. And again I think the emphasis here is in just preventing white from any attacking options. I overlooked one thing. When I suggested the move queen E7 as opposed to what was played now, I forgot there is always this bishop H5.
Mr. Ashley Neither rook.
Mr. Seirawan Indeed we will go back to prophylactics. Remember in game 2, the one that Garry won so brilliantly is what he mentioned to do is handcuff his opponent by attacking opportunities, and this is a marvelous example of that kind of strategy.
Mr. Ashley A bishop could have gone to D5. That queen F7 move has controlled many white squares and so must be admitted as a very nice move.
I think Kasparov has offered a draw. I'm not sure but there was some bit of conversation between Kasparov and Murray Campbell, the operator for DEEP BLUE, and then Murray Campbell with a phone right next to him spoke to his team somewhere off stage, and it would be very likely that Kasparov would offer a draw in a position like this. It is a kind of a position where he knows he is worse, he does, but he also knows that pawns are on one side of the board. I remember that in game 2 pawns were on only one side of the board and we had anticipated that this game should be a draw. DEEP BLUE did not handle the end game as best it could and Kasparov managed to win.
It seems now as if the shoe is on the other foot. Kasparov is going to have to defend the pawns on one side of the board in this position. It may not be a draw.
Mr. Seirawan Definitely the shoe is on the other foot and Garry wanted to get the shoe off the foot if in fact he offered a draw.
My take on the position is that if Garry was black, he would certainly decline a draw. The position objectively because of the extremely reduced A material should end up being a draw, but this is kind of like the sunny side of the torture. If you are going to have an advantage you might as well be the person with the advantage and enjoy yourself. And Fritz certainly thinks that black has an edge, and so if I was black, I would want to play the position on for some time.
Mr. Ashley Let's take an overview of this match and what this game 4, which I thought would be so critical to the match is saying. They have played -- they have reversed wins in the game 1 and 2, and now we are seeing -- we saw a draw in game 3 and possibly a draw in this game.
We don't expect DEEP BLUE to blunder and lose this game. The question is can DEEP BLUE win? If the game does end just in a draw, then after four games, this supercomputer will have played the world champion to a standstill. We played four games and what have you shown me is not so special.
You won one game but I attacked you one game and now we played two draws now that we both settled on and know what we are playing against.
Before this match all of the experts, or most of them, were saying that of course Garry is better than the computer. Are you trying to disrespect my man, Garry? Garry is the strongest player around.
Mr. Seirawan Most of those experts included myself too.
Mr. Ashley I thought Garry would win by a margin of 4 to 2, but forget that now. It is a match where anything could happen. This is a draw we have seen a single slip by Kasparov what it can do. It can change the nature of the position dramatically.
Not only that, not only in the play itself but you have to wonder in the spirit of the man, Garry Kasparov, very confident that he would win this match and now come in and realize that he is in a fight for his life.
If this comes down to a game match, anything could happen. One blunder and you are gone, match over. So it seems that the computer is really showing impressive skills.
Mr. Seirawan Totally agreed, and all kudos to the IBM team.
I want to emphasize the confidence slash arrogance that Garry had going into the match.
Of course the ACM, the sponsor of the match, offered Garry an opportunity to battle for a half million dollars. And what they said is Listen, Garry, we will give five-eighths to the winner and three-eighths to the loser. And Garry said, What are you crazy? I want a winner take all, a half million, please. And ACM said you have to be a little bit fair here. IBM and so forth and so on. And they compromised and it came out $400,000 to the winner and $100,000 to the loser.
So the point was going into the match Garry was so sure of his victory, he was thinking this is the easiest $400,000 he ever pocketed. And what a surprise. Mr. Ashley Garry was like, they are giving me $400,000. They are giving me $400,000 in response to that, man. So this guy felt that Garry was being paid $400,000 to show up. And I said, I don't think so. I said, I think he has to win the match.
But clearly Garry knew that he had to win the match to get the $400,000, but he was so sure that he was going to that he was telling his friend, they are giving me $400,000. So I think he is in for a rude awakening. It is clear right now that it is not that kind of simple battle.
Mr. Seirawan I would say what Garry has to do in this particular game now, what he should be trying to do is exchange off one or two pieces. The move bishop E2 plays successfully to the knight on E5.
What I mean by that is the knight on E5 attacks the square C4, D3, F3, and G4, and the bishop on E2 would cover all of those squares so that the two pieces would be in balance. The move bishop E2 drops back a square in order to prepare Kasparov's next offensive move, which would be queen on G3 to E3, the anticipation of that move being that then we could play rook A7.
So Garry has to go into a defensive shell. And I think the best way of setting that up is bishop E2 stopping the knight from moving anywhere. And also the bishop, it is very important to harness the energy of the rook on D8. That rook on D8 right now can slip to all kinds of squares, D3 most especially, as you pointed out, pinning the bishop and queen together.
So bishop E2 would block the rook from two key squares while keeping the knight under control. So I think we are going to see that move.
What Garry's aim should be at this point, if he could take off any pair of pieces, trade queens or rooks, they could agree to a draw comfortably. Because it would not be enough. For DEEP BLUE to have any winning chances, the computer needs to combine all of the pieces.
(Audience commentary.)
Mr. Ashley We have seen that on different occasions.
Mr. Seirawan This question oftentimes arises. Just to be clear, DEEP BLUE is programmed to continue actual play. So the idea is the operators are not going to accept draws on DEEP BLUE's behalf unless DEEP BLUE is absolutely exhausted all of the possibilities. So Garry understands full well that basically it is play to a finish, and only when they get into what is known as a book draw, a theoretical draw which generations and generations of players know that this is a draw, then they will agree to a draw.
(Audience commentary.)
Mr. Ashley The comment was that certainly we are putting a lot of accolades on DEEP BLUE for its play in this match, but a continuation of that comment was DEEP BLUE is not capable of making a mistake and Garry is as he did in game 1 and lost.
Mr. Seirawan Which I, of course, do not agree with. In fact, the computer makes all kinds of mistakes. In fact, positionally, strategically, it makes terrible mistakes.
We know when you look at chess from the very starting position the game is in complete balance or, more properly put, the word equilibrium. So if the game is perfectly played and there are no mistakes between either side, the game should be a draw. In other words, one army versus another army, equally matched, should be a draw, but the computer lost. Wait a minute. How can the computer lose if it never makes a mistake?
Clearly the computer had made a mistake and the mistake that the computer made is based upon judgment. The computer sees every move that Garry plays, but it judges -- its evaluation of that move is different from Garry's. And what Garry has shown, at least in game 2, is that in certain positions his judgment or his evaluation is better than the computer's.
(Audience commentary.)
Mr. Seirawan The point was that DEEP BLUE did not make the mistake, the programmers did. And you have to say that this is all the programmer's fault.
Mr. Ashley You might put it that way, but I think the problem is that they have a very tricky task, and that is to make the computer play a certain kind of game that we understand to be good, but it is very difficult to do, to teach positional intuition and understanding, experience.
When in game 2 the computer took a pawn on B4 and allowed his kingside to be opened up and Garry's bishop to be posted on a very nice square on E4, it took exhaustive calculation on the computer's part to realize the dangers only on the next move after it had already done it to itself.
How do you teach -- this is the problem these programmers have. How do you say to the computer, you know, positionally what you did is bad? That is not a trivial task. That is a monumental task. If they could do that, chess would be done.
(Audience commentary.)
Mr. Ashley That is why Garry is so confident. The problem is that they are able to give DEEP BLUE other things and that the most potent weapon is the ability to calculate all of these numerous variations. And that compensates somewhat for Garry's positional understanding.
And we saw it happen in game 1, and we are seeing it happening now in this game. The computer is able to compensate for Kasparov's superiority in positional understanding of the kinds of things that he wants to do that would make his position great 15 moves from now. A slight slip and it will change the game into a battle of calculation, and the computer says you are much worse than me at this.
I would not say that the programmers have not made any mistakes, but they have not cracked the problem of teaching positional intuition to a computer. That would come to almost teaching the computer to think, I mean, really. And that's -- I hope that doesn't happen. Then they are going to be really dangerous. They might figure out that they are better than we are.
(Audience commentary.) 42. Rxe5.
Mr. Ashley Before we get onto that question, Kasparov has played a move that I thought would not be played. And I wasn't about to suggest it, even. The move rook takes E5 was sacrificed. He sacrificed his rook on E5, getting rid of that extremely powerful knight, which creates for him a solid position where his bishop and pawn will compensate for the disadvantage he will have of a rook.
And that, Yas, has brought cheers from our audience here. And maybe some anticipated this possibility. It had just briefly flicked in my mind, and I said no, he would not want to do that. But now a dramatic move by Kasparov, changing the whole complexion of the game.
Mr. Seirawan I thought that he would do that.
But to answer the gentleman's question a little more succinctly, I think you are putting it in too stark a position of black and white. Kasparov is the fact of the world's best tactics in the human's world of thinking. So you cannot say that the computer is overwhelmingly better, but rather equally matched.
As we have seen, the computer is not outplayed strategically game after game after game. There are nuances and advantages and disadvantages. And I think that the human being is probably better than the computer tactically, when forced to do so, in a given position. So that the human player can look deep, deep, deep, deep, deep, as deep as the computer, and I think strategically we have seen that DEEP BLUE is quite a competent statistician.
Let me get back to the game. Rook takes E5 is a very nice sacrifice because again from the perspective of Garry, he knows that it is only through a direct attack that he can lose the game. Once he eliminates the knight, what he will be left with is three pawns versus two pawns on the same side, and bishop versus rook and pawn, bishop and pawn versus rook.
Even if an ending where it's bishop and three pawns versus rook and three pawns on the same side, it is a loss for the bishop. But it takes perfect play and it's -- there are some positions that even then are drawn. I think this was an excellent practical position by Garry, and I think that now it will be very difficult for DEEP BLUE to coordinate attack. 42. ... fxe5
43. Qxe5.
Mr. Ashley The moves F takes E5 by Black and queen takes E5 played instantaneously by Kasparov are now on the board. And the advantage of material is certainly Black's. He has a rook against a bishop and pawn.
But experience, as you said, experience -- I love when you started calling back on your knowledge and experience of these kinds of positions. Experience has taught us that positions like this tend to the draw. And no, Garry Kasparov has not calculated all the moves to the end. He couldn't. And he doesn't need to. He knows from experience, from studies of these kinds of situations, he knows from seeing many, many grandmasters' games played with this kind of team, that this tends to work out favorably for the side with the material deficit, in that the draw is still in hand.
And this is why Kasparov has decided on this decision in this situation. That is the difference in the material on the board right now. He has no winning chances. Kasparov is not attempting to win the game here. He is creating a situation where Black will not be able to bring his position down. And in chess terminology that is called a fortress. You set up a position where the guy cannot get in and your inferior forces still manage to balance for the superior forces for the other side.
Mr. Seirawan The practical consideration of the clock is a lot more difficult to defend when these are tons of attacking material and just this elimination was a very good practical decision. I would like to invite questions.
(Audience commentary.)
Mr. Ashley Would DEEP BLUE have played this move? I don't think it would have. I don't think it would is have played this move because in order to justify this move you have to calculate 15 moves from now.
First, you would have to figure out that it can set up a position that is actually winning. You would have to set up a position that is actually drawing for itself. I don't think DEEP BLUE -- I think there is a clear case where a human and a computer make totally different decisions. I think DEEP BLUE would have played this move only if it contemplated a ways that clearly won for black if this move were not played.
If it could not find such a variation I don't think the computer would have played this move. I think.
Mr. Seirawan I think the answer that Maurice is giving is accurate. It would not have captured. I don't know, again, if DEEP BLUE had seen something so clearly as leading to a victory for its opponent, it just might have played this rather unorthodox type of chess.
Mr. Ashley Kasparov has kept his queen on a very important diagonal. The B8 or H2 to B8 diagonal. The reason for that is potentially his king might be forced on that diagonal. We could see that easily happening. It would be on this long diagonal and then a black queen would love to be on this diagonal check. But with the white queen on that diagonal it is not going to happen.
So Kasparov is setting up a defensive position. Maybe this is his intention to draw the game, to set up a fortress. And it is well known from older military strategy that sometimes the interior forces can easily hold the superior forces if a nice fortification is set up. I know from being very interested in the Civil War. You want to force the guy to come back at you because you can strike better from those positions.
It is very difficult in chess to break down the pieces inferior in material. As a matter of fact, Yas, there are some positions where a queen can beat a rook and a pawn or a big queen cannot beat some smaller pieces. You think queen should be able to break this down but you can't because the forces are so perfect.
Mr. Seirawan I just wanted to say when we started this line with rook takes E5, Fritz had given the black player a large advantage, 81 hundreths of a pawn advantage, and since we have played just a couple of moves on the board, Fritz evaluates it as 00.
I would like to take you through this variation. This is the position on the board, queen F4, and in case of this idea of rook E1 check, king H2, queen to A7, in order to penetrate with queen to G1 check, after the move F6, queen to G1 check, king to G3, there is not really anything more that black can do. Black has flushed the white king out of the king side, but on the other hand it is perfectly safe. As Maurice was saying, a wonderful fortress has been built.
When I was starting chess, one of the things my early teacher did to me was slap my hand any time I wanted to make a check because his expression was patzer sees a check, patzer plays a check, patzer being a weak player, and I know that in passing the last move of the computer was queen F6. The computer did have a chance to give a check but it refused.
So we are not seeing a beginner here, are we?
Mr. Ashley No, we are seeing a very potent Grandmaster. The computer is playing excellent chess. This move is a wonderful one. It almost by force to trade queens if the computer so desires and thinks it is an advantage to do so. It may exchange queens by giving check on E1, and then when the king gets up to H2 by playing queen H5, that position I would imagine is not winning for black. 45. Bh5 .
But now, Kasparov has played a very aggressive move, the move bishop to H5, attacking the rook on E8, a definite tempo gainer and it seems to me that the rook will have to move. But in the meantime the bishop planning to go to G.
Mr. Ashley And give check. And now will be a very strong position for the bishop.
What about this end game we have been talking about? This brings about a forced exchange of queens. What is going on here, how is white going to draw this position?
Mr. Seirawan Dead draw. This is exactly what Garry had in mind when he sacrificed the exchange. He had this long term vision. He did not need to calculate each and every move, but he knows that after an exchange of queens, again, when it is a rook and three pawns versus a bishop and three pawns on the same side, it is a win for the rook, but it's really, really hard. You have to play virtually perfect chess, and I have seen that end game drawn a number of times.
When it is bishop and three pawns versus a rook and two, there is no problem whatsoever. The draw is readily apparent. So, for example, what we would see, let's just take this variation a couple of steps further. Queen takes E5 and rook takes E5, G4. The only way for black to win would be to walk his king around, king G8, king G3, king F8, H4, king E7, king F4, and there is absolutely no way that black will be able to prevent G5 and F6 just trading all of the pawns off the board.
Mr. Ashley Also, well known is the fact that a rook versus a bishop is also a draw in 95 percent of the cases, so this knowledge of chess history is what we as chess players thrive on, and because of that, we instantly can evaluate positions like this to say no, this position should be okay for the defending side. 45. ... Rf8.
But look at this, Yas, after bishop to H5 the rook on 8 was under attack and instead of keeping it on an active square, DEEP BLUE has played a somewhat unusual move, rook to F8. That looks bizarre, at least we would say. But what is this move about.
You know, wait a minute. I'm starting to stop and think about this for just a second. This move clearly the pawn on F5 is under attack, and a very obvious move for Kasparov is to put the bishop on G6 with check, but one thing that is interesting about this bishop is if it leaves, then the white king might be less protected.
I say might because it is not so clear that black has a strong attack. Maybe black is planning now to coach the bishop forward. I don't see how that is going to work. It is an interesting way of trying it, but it has to worry about checkmate on the back rank.
It seems to me that that is going to be a big factor in this game. I don't see how DEEP BLUE is going to carry that out.
Mr. Seirawan In fact, now I would say my own personal assessment is that all of Garry's problems are behind him. After the moves bishop G6 check, king H8, queen C7, the bishop, as you say, has been coaxed forward, there by meaning that are you no longer setting up a fortress.
Are you going for something else? And what Garry's new inventive setup would mean is that the bishop is so powerfully posted on G6 that black's rook could not get active due to this black mate problem.
I think it is a very simple draw now. Garry will put his bishop on G6 with check. Queen to C7.
Please note -- woops. I did something wrong. After king H8, please note that this is a threat of queen takes G6 capturing a bishop. So queen C7 immediately -- and then the point that all that white has to do is play king H2 and just move his queen back and forth on this diagonal and there is nothing that black will be able to do. I think before too long --
Fritz is still giving black a considerable advantage, but I think before too long the computers will both recognize there is no danger to Garry's position. Mr. Ashley The bishop and the pawn will compensate in large measure. This is the kind of position that most humans will have stopped already and said -- grandmasters, any way, will say I know. You have set up a good fortress and I will not be able to win.
Mr. Seirawan This is what makes chess such a maddeningly complex game because of the paradox, paradox of material. You will be in a situation where you are way ahead material. You have enough to win the game. But this particular position, you have a fortress or there is something else going on. And although you are materially ahead, as she just described, Black on the material count is a whole point ahead, but he can't do anything with that extra advantage, and I think the game is going to be shortly drawn.
(Audience commentary.)
Mr. Seirawan The question was, after rook takes E5, why did Garry not make a formal draw offer. It has to do with chess etiquette, actually. Chess etiquette says that even if the position is drawn, but you are behind in material, you allow the person who has more material to make the draw offer. So I mean, I don't know. Does the computer have good manners? Should he be worried about the computer's etiquette?
Mr. Ashley The thing is, he may have offered the draw before. I think he did offer a draw before and the computer evaluated it with an advantage for itself, and so the DEEP BLUE team decided it did not want to accept the draw offer.
But some moves have been played, and exactly as you said, DEEP BLUE responded with king to H8 and queen to H7, finding now Mr. Kasparov leaving the chessboard feeling very comfortable knowing that a draw is in hand. In this position, losing would be either miraculous or disastrous, however you wish to see that. It looks like there is no winning chances for Black, to our eye. It is just not enough.
Mr. Seirawan My guess is that in another ten minutes the watch is going to go back on.
Mr. Ashley I wonder that it has not gone on already in fact, that Kasparov believes that this position is a trivial draw and he does not have to watch out for any real dangers. Clearly queen and rook together are a potent force.
Mr. Seirawan They can complement each other very nicely because as major pieces vertical and horizontal movements, very quick strikes, and the rook is an excellent backup for the queen. The problem in this position is that the rook and queen do not have free access. They cannot coordinate unfettered because that Black king is in a box. So they can't go far.
Mr. Ashley The bishop on G6 is a very strong piece.
Arbiter Mike Valvo has come over to the board. He, himself being a very strong chess player and international master, knows that this position is a draw and he has stepped in. Kasparov no doubt has made his feelings clear to Mike, saying I know this is a draw. Why don't you do something about this business here?
And Murray Campbell, the operator, is on the phone. DEEP BLUE's operator and part of the team is on the phone, talking with the members backstage, and disagreeing, saying that the team wants to play on. It seems to me that that is what the shrug was. And I think Mike is looking with a bit of human disdain.
Mr. Seirawan The neutral arbiter is leaning towards the other side.
Mr. Ashley Come on, now. In the meantime Fritz 4 is busying itself with other more weird matters, as a comment it makes is -- it has been programmed with some comments built into it for different situations in the game and apparently it thinks that this position is so trivial that it is saying here don't you think Judge Ito lost control of the case? Maybe more interesting than the position we have in front of us is the debate about that, about the judge's skills, but --
Mr. Seirawan This is another point. There are several ways of looking at this particular position. The first point is that when people are beginning chess, although you and I both know it is a trivial draw, you often tell your students to squeeze the last drop out of the position. Just, you are not going to hurt yourself by playing on and figuring it out.
And the operators, to their credit -- this is a new machine. This is the first time they are playing with it. So they just perhaps want to give a little bit more feel to the computer and see how it will handle this position.
Mr. Ashley They certainly have to feel pretty good backstage. And I'm sure Joel Benjamin was consulted as to his opinion about their chances, and I'm sure they were saying to themselves, wait a minute now, I know it is a draw. Control is probably say I know it is a draw, but we can't lose, right? That's good.
So if there are some weird tricks to be discovered by this monstrous calculator, why not go ahead and do it. Why not just try it and see. Maybe it will discover something that we have never seen before, which is a valid point. I mean, play a few moves. You never won by just accepting a draw before. No one has every won by saying draw, let's leave. So why not make a few moves.
DEEP BLUE has in fact made a not centralizing move, queen to D4, and this move would look good if not for the fact that that rook on that H9 is not going to join the queen in this battle.
Mr. Seirawan Exactly. I think queen D4 is a wonderful centralizing move, appears to attack towards the white king. We have king H2, an expected move. One response might be rook to A8, and again white can just move his queen comfortably back and forth on the diagonal. Because whenever the rook gets too active, there is always this check on the back rank.
Mr. Ashley Kasparov, after that last move, has played the move king to H2, putting his king on the same diagonal as his queen. We mentioned before that that was one of the dangers in the position. So it still looks as if the position is going more towards a draw.
Grandmaster Joel Benjamin has joined us on stage and we would like to hear his opinion about what is going on in the DEEP BLUE team.
Mr. Benjamin Basically, as I see the position, it looks to me like it is a draw. I don't see how to win it. But I feel that there is no reason not to play on at least a few more moves, see if there is any possible progress, and see if DEEP BLUE can see that it is going somewhere.
It is the kind of thing that there is no risk involved. It can easily have a draw any time it wants, and it is not going to get nervous about having this position. So we don't intend any kind of disrespect to the world champion. And if in a few moves it becomes clear that there is really no chance that it is moving its pieces back and forth, it doesn't do anything, then we will, but it seems like it is best to go for every possibility.
Mr. Seirawan Joel, just a quick question Earlier did Garry offer a draw, do you know?
Mr. Benjamin I'm not exactly sure about that, but from my point of view, I felt that from the end of the time the Black had some advantage, certainly not a winning advantage but had the better of it and really no risk in the position.
And we know that DEEP BLUE is not going to make an elementary tactical error and it is not going to get nervous, so there is no reason not to play out the position. Besides that, we have people that might be interested in seeing how long the position is played out. And I think it is best in chess when any possibilities that are there are exhausted.
Mr. Ashley That is very true, Joel, but there are some of us that want to go to dinner soon. We've been on stage a while.
Mr. Seirawan Thank you, Joel.
(Applause.)
Mr. Seirawan I've been watching the screen. Garry is going through some of the permeations what would happen if he tried something else. This is what Garry had anticipated the main line to be. This was Garry's main line, and I guess he felt it would have led to an advantage, not much of an advantage, if I do say so, but this is what Garry thought was going to happen.
Mr. Ashley Those of you who are only going to be able to read the transcript of this, what is happening in the hall, Kasparov analyzing with the DEEP BLUE team going over many, many variations as he saw as possible and what was good for him and what was not so good, and he was wondering -- I think Murray Campbell is a very capable chess player. The computer operator is going to respond, I think, to some of what Kasparov was talking about. We have seen this before, Kasparov suggesting moves and Murray not saying anything.
Mr. Seirawan He is getting a good lesson.
Mr. Ashley And one has to wonder what Kasparov gets out of it except maybe some release as to what he thought was possible. He is quickly going through variations here, his hand moving as quickly as his brain does at light speed and showing the variations to Murray as to what he thought was really possible in the position. Murray seems to be agreeing, but since we cannot hear what they are doing, maybe telling yes, DEEP BLUE saw this variation, X, Y, Z moves were possible.
So Kasparov, just analyzing, not looking too upset it seems, but it seems like the DEEP BLUE team should be content, should be content. Their baby survived four games. They are even with Garry after four. What a dream.
Mr. Seirawan Absolutely. I think that coming into the match, everybody had high hopes and I'm sure the programmers for IBM had no clue really how strong or weak their baby was because, again, it was almost birthed just a few months ago, and even as C.J. admitted himself when they got here and they turned it on and it worked they were quite relieved. So they did not know what to expect, and obviously prior to the match they said, listen, after four you are going to be even, you are going to accept it. Yes, absolutely. Here again we are seeing a lot of the tactics.
Mr. Ashley Some of the tactics that we had seen here when we were analyzing, but Kasparov looks quite animated to me. He does not seem at all disturbed, probably very relieved that he found this idea rook takes E five which forced a draw and it seems to me that he felt as if this game was a fight, was a real fight between two top-level grandmasters and the fact that he did not manage to pull his advantage was a bit unlucky and it seems to me that he was impressed by the computer's skills, but he does not look upset. I would like to know his thoughts on the matter. He is like I played the game and it was a draw, so what.
Mr. Seirawan Setting him up for the weekend, indeed.
Garry is going through the game, but on the screen in the middle I have what I think is a crucial moment, the 25th move. On that move Garry played the move rook E4 and in my view this was a bit of an overplay. He wanted to follow up with the move rook on E four capture E4 putting pressure on the C6 pawn. I think the correct continuation was queen take C4 leaving the rook on E1 with the idea of going to better pastures with rook coming down perhaps to E7, and I think that was a crucial moment in the game, and when Garry comes out here we will ask him if he agrees that rook E4 did not work out exactly the way he had intended it to. So maybe just the direct queen takes C4 would have improved his chances.
Mr. Ashley In the meantime he is still entertaining the DEEP BLUE team with his ability to see the positions as well. And also suggesting this move, you know, at one point in this game and allowed the sacrifice of rook takes E5 and Kasparov is saying it seems he would not have had to worry about even this variation queen to E7 because of rook D five which seems to him to also hold the position, so the tendency of pawns on the same side only to draw is being shown very clearly here.
When you have pawns on one side of the board, it is so much easier to draw in the position where the material is indeed in balance.
(Audience commentary.)
Mr. Seirawan Interesting question regarding style. There is no question that Garry's great trumps are his attacking skills and he has had to neuter them because of the great defensive skills of the computer. Quite often it happens in chess where you have a match-up where player A always beats player B. Player A is the nemesis of player B and then player B always beats player C and player C always beats player A. So you get this kind of a loop or a circle and no one can understand why they are cursed to lose all the time to one particular player and it is a question of stylistic match-ups, and the gentleman suggested that there are styles, defensive, prophylactic styles of Karpov and others that might be more adept at beating the computer and it could be the case conceivably.
Mr. Ashley Kasparov has said that he may not be the best player in the world against computers. He has made mention of other players that are extremely adept at playing computers. Joel Benjamin is awesome against computers. When he plays in the Harvard Cup, which is the match-up between six top grandmasters and six top computers, he wins every single year. I think his record is 17 wins one loss against the best computers around. That is a true testament to his ability to play the kind of positions that are not favorable for the computer, so Kasparov may -- okay he is the best player in the world, but he may not be the best player against computers in the world. And now we are seeing maybe a match-up -- I would hate to say that a match-up between Joel and the computer would be a better one because I don't think so because Kasparov is such a wonderful chess player, but it is true that he has had to adjust his style dramatically in order to play against the computer.
Mr. Seirawan The future may hold that chess players engineering their styles to be computer specialists and Garry has obviously geared his style to beat all humans and maybe he is going to have to make adjustments as we all might in the near future.
Mr. Ashley It is possible that the adjustments may not matter because the computer may get so good that you may lose no matter what style you play.
(Audience commentary.)
Mr. Ashley It is still the question of time against computers. You say strength. When you say strength, what do you mean, raw calculative ability is still what DEEP BLUE does better than any other computer. But we have seen DEEP BLUE lose chess games. We saw it lose -- we have seen it lose to Kasparov already and how did it lose? Kasparov got it in the kind of situation where it did not play very well. Now, there are players, grandmasters that know how to do that. Kasparov did not play particularly spectacularly in that game. He made certain mistakes because he allowed certain things to happen where he could have one even quicker, but he got himself into a situation. So there are players around that no exactly how to do that. There are grandmasters around that feel like I will beat the computer with no problem. Put me against the computer because they feel that their style and strategy is so perfect that they can beat this computer with no problem. They would never let that happen that they would control the game.
I remember there are some events you played in, Yasser, where with grandmasters and computers -- is it AJON?
Mr. Seirawan AGON. Every year, very much like the ACM that puts on annual events. All around the world you are seeing computer match-ups and each year the Dutch Insurance Company, AGON, puts an event where they invite 32 computers and 32 humans of all varying degrees and strength from Grandmaster to club player and they say you guys go out and may the best mind win.
And for the first time in the history of the AGON computer challenge last year the computers outscored the humans overall, so we are seeing great progress being made in computers and it continues.
Mr. Ashley Who was the best human in that?
Mr. Seirawan Grandmaster John Vanderweil played a great tournament.
Mr. Ashley And he, I know for a fact, feels that he could handle these computers any day of the week. He gets them in the kind of positions where they are very weak, and the truth is Kasparov himself has said that you will see an age when these match-ups become more and more interesting to the public. He says you will see an age of chess grandmaster who specialize in playing against computers, and I think if you put it to him, he would say he is the best player, but he may not be the best player against computers. He may still have to learn a thing or two about how to control these calculating beasts, the fact that they can calculate so well and that he has to avoid these types of situations. Here we see a typical game.
I believe Kasparov has left the room. We would anticipate that the world champion would arrive shortly to discuss this game. 2, 2 though. Computer with white in the next game, anything really could happen.
Mr. Seirawan That is a scary thought because I really want to see Garry win this match and get out of here still intact as a human being. I think this would be a very shattering experience for Garry because his entire life has been built up from one success to the next and he has never in my view been in a match like this one where he realizes a single mistake may cost him, and I do believe we will be joined by him in a minute. Can we have a drum roll? Game ended 729 PM US EST Drawn by mutual agreement.
[ The Game | The Program | Commentary | Grandstand | Deep Blue ]
[ IBM home page | (C) | (TM) | IBM Small Planet Pavilion | Internet 1996 World Expo ]