Mr. Seirawan A little better showing than game one, but we thought you had opportunity today. Drawing was a little disappointing maybe?

Mr. Kasparov Yes. I also saw that I had some opportunities, but ... you know ... I played very well. I was missing one tempo, just to get my knight on rook 5 and rook 8 to prevent C4. I could not figure out how to do it. We went to the rook ending, which looked good for me, but it is an excellent defense. It is a draw.

Mr. Seirawan We had an opportunity to speak with Joel Benjamin, who told us that he has been playing some training games with Deep Blue, and it sometimes was like a basketball game, where one of your players takes a 30-foot shot and the coach is screaming, "No, no, no, don't do it!" until the shot goes in and everybody cheers and says, "I knew it was going in!" So Joel says that, in his games with the program, it made some crazy moves that ended up working out, and we thought that rook C4 was one of those 30 foot shots, a training move that we did not anticipate.

Mr. Kasparov I did not expect that. Any human player would play C4, but Black is still fine. And the pawn ... you know ... I think that is the problem with the computer. It sees the calculation in its calculation, the danger for its pawns and it does not want to create new weaknesses because C4 is a natural move. It is working its pawns, the rook at C4, and it also takes an open file. From computer's point of view it is very logical, but I found a nice maneuver that is probably not easy for the computer to find because the moves are not so intuitive. But I was missing one tempo, my pawn A6. I would win because I would just create this wide square for blockade.

Mr. Ashley Are you impressed by the computer's defensive technique?

MR. KASPAROV No.
(Laughter.)
There are only a few moves that any human player would pick. That was the only move; otherwise, white is lost. A human player would not even consider taking on C4 because the end game is just lost if Black puts knight on D5 - the end game is just lost. But, you know, a few of those moves, they just proved that my attempts to win had failed, and again, it's probably for the machine that calculates some moves per second. It was not difficult to find the best defensive line because, unfortunately, it was quite clear what white had to do to bring King on center. The best moves were the most natural moves, and that was the unpleasant part of being in that position.

Mr. Seirawan It must have been frustrating for you when you played the move rook D8 [move 29].

Mr. Kasparov The problem is that, if the computer does not take on F6, it is losing something by force. Unfortunately I had to create so much stress that it was forced to find the only line of defense. But what can you do? This is definitely a computer thing ... I don't know who was preparing the machine for this game, but the free check was definitely missed, either by machine or by someone who was training the machine in this position, because it is quite surprising that the machine doesn't repeat the line of the first game and does not play H3.

Mr. Ashley I remember Joel Benjamin saying he saw this position. He saw that ... well here he is to explain it himself!

Mr. Benjamin What happened was we prepared it up to the position of the queen D3 [move 17], and I thought the next move is C4 or bishop E5 and it is okay. I didn't see the tactical ramifications of that, and we didn't have time to test it. Otherwise, DEEP BLUE would have told us.

Mr. Kasparov It took a lot of time, and I checked probably three more times than I do normally.

Mr. Benjamin Fortunately, we did not force it to play that.

Mr. Kasparov It is a very unusual idea, so the computer do fall easily into this trap. The first game, you know, I believe that the computer found -- queen D5, which forces a draw by repetition, but in the first game I was too optimistic.

Mr. Seirawan This is important. You said at the opening press conference that the first two games were very important for you, because it would allow you to understand how the computer plays offensively when White and defensively when Black. Now the first three games are over, and I must say that I thinktoday's game was very impressive on your part, because you neutralized the computer almost from the start. Almost right from the opening it was you that was sitting there, a little bit optimistic, seeing some chances, some more chances - that was a very nice strategy that you employed in today's game.

Mr. Kasparov I know. It was a bit scary, too. It is the Main Line I play, and I was very proud of my move bishop B4 [move 9], which I found. I think it neutralizes because it creates pressure, but in the game, I did not play a move that I would play at a tournament, queen D5, and computer must go on C3 - it is quite simple force. Today, I was just curious to find out if it was prepared. I don't know knight D5. It's my intuition. Black had to be okay. The position is not that dangerous.

Mr. Seirawan I felt the same way. Mr. Ashley Before we let the spectators in on the questions, you spoke about intuition. What is your intuitive feeling about the computer's strength, its actual strength? Has this computer broken the 2600, 2700 barrier maybe?

Mr. Kasparov I don't know. We have to play more games, the match is still in progress. It is really too early to make any long-term predictions, but you have to find an average because in game 1 the computer played with the rating of the best players in the world. Now, in game 2, it played on a very, very low level. It made mistakes that most players would avoid. How to find an average? The machine's strength is varying from top level - an ELO rating of 3,000 - to maybe 2,300. Where is the average? I don't know. It depends very much on the number of positions that computer could play well. I would say today it is in the 2,700 range. It presents such a danger for the human being in simple practical terms. But if you create the right position, you will enjoy its deep misunderstanding of chess.

Mr. Seirawan Hans Berliner, a question.

Dr. Berliner Did you consider the move king to e7 at move 23?

Mr. Kasparov I did consider king e7 at move 23. But rook comes on C6 and then C4. I don't know ... again, I would probably consider this move in a normal game, but I saw the computer's maneuver, which I found in the game. I saw this wide square blockade and I thought - fine. It is very difficult for me to see ten moves ahead. That is, one little nuance like H7 would destroy a beautiful position or plan. It is difficult to move this rook.

Mr. Seirawan We will be taking two more questions.

Audience question Is there any particular aspect of the game that you found that the computer plays better?

Mr. Seirawan Say no, Garry. Come on.

Mr. Kasparov Any position where the pieces are hanging and when you have to calculate, you have a forced line, you know, you take, he takes and you take. Here the computer just, you know, plays cool, and you don't have any real chance to stand against that.

(Audience commentary.)

MR. KASPAROV End game also could be a problem. End game like today, you know, it probably does not understand the end game, but it sees too deep that when the number of pieces are limited and the pawns could be eliminated quickly. That creates an opportunity for the computer to see to the very end.

Mr. Seirawan I will ask a quick question at the end. I can understand your frustration a bit because it seemed to me that the game was a draw for quite a few moves, and I believe you had offered a draw and there was some question ... ?

Mr. Kasparov No, no, no. When I took on D4, the computer was going and I spoke to Murray and I told him, "This position is a draw even without the h pawn."

Mr. Seirawan What I said to the audience.

Mr. Kasparov And he said, "Yes, of course I see, but computer is thinking." and I said okay. So it was just a matter of time because the computer was calculating something and he called, and then we just discussed the idea of the situation if I offer draw, in a more complicated position, and I was curious what the computer's reaction was, and Murray told me that computer is programmed to decline any draw offer unless it is told otherwise.

Mr. Seirawan So the protocol is contemptuous of you. An "I will win every game" mentality.

Mr. Kasparov In fact, it is quite funny. I understand that with other problems they could accept a draw if the evaluation of the position from the computer's point of view is down. It is quite strange that machine declines a draw for any circumstances because position could be quite bad, but now I realize that I can withhold my draw.

Mr. Seirawan One last question, about the pace.

Mr. Kasparov The pace, that is something I have to get used to. It was a bit frustrating in game one and now I understand. The good moves are natural moves. It would play, you know, very fast because I could play also quite fast, but the problem is as a human being, I'm hesitant, you know, I'm trying to calculate and calculate again and again. The computer saves this for calculations. I still spend a minute or two just checking. And the computer just sees the good move and it plays that. That's why itsaves its time that otherwise would be spent for some doubts, reasonable doubts.

Mr. Seirawan That will be it for today. Thank you very much, Garry Kasparov.

(The proceedings ended at 630.)

[ The Game | The Program | Commentary | Grandstand | Deep Blue ]

[ IBM home page | (C) | (TM) | IBM Small Planet Pavilion | Internet 1996 World Expo ]