In-depth analysis of game two

Nov. 15, 2003 – The second game of the match was a complicated strategic battle. The commentators were surprised at how well X3D Fritz played the early stages of this closed positional game. It had pressure on the queenside while Kasparov built-up aggressively on the kingside.

Then the machine released its pressure and it looked like Kasparov would have good chances of a breakthrough against X3D Fritz's king. However, he had slipped into time trouble and several hasty moves were followed by a horrible blunder. X3D Fritz wasn't going to miss such an opportunity and it immediately launched a crushing attack.

Well after the game ended Kasparov spoke about how much he had missed for both sides in the moves before his blunder. Getting into time pressure is a bad habit, but against a machine that sees everything it will nearly always be fatal.

[Replay the annotated game online and download PGN ]

X3D Fritz - Kasparov,G (2830) [C66]
X3D Man-Machine World Championship
New York USA (Game 2), 13.11.2003

1.e4 e5 Classical defense from Kasparov, who usually always meets 1.e4 with ...c5, his beloved Sicilian Defense (1.e4 c5). 2.Nf3 Nc6 3.Bb5 The Ruy Lopez, or Spanish Game. One of the oldest known openings. 3...Nf6 (D1)

Kramnik used this move against an earlier version of Fritz last year. 4.d3 According to Alex Kure, the opening book expert for Team Fritz, they wanted to keep the queens on the board and not allow Kasparov to follow the main line of the Berlin that Kramnik used to draw easily against them in Bahrain game one.

[4.0-0 Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 8.Qxd8+ Kxd8

The main line of the Berlin Defense of the Ruy Lopez. This line has been popular off and on going back a century. It's latest resurrection was caused by world number two Vladimir Kramnik, who used it to great effect against Kasparov in their 2000 world title match. The mighty queens come off the board and the game moves in slow motion.

Kramnik wanted to prevent Kasparov from displaying his famed tactical prowess and the same strategy makes perfect sense against a calculating monster like X3D Fritz. Kramnik played the Berlin in his first match game against Fritz 8 a year ago and, as he often did against Kasparov in their 2000 match, drew convincingly:

9.Nc3 h6 10.b3 Ke8 11.Bb2 Be7 12.Rad1 a5 13.a4 h5 14.Ne2 Be6 15.c4 Rd8 16.h3 b6 17.Nfd4 Nxd4 18.Nxd4 c5 19.Nxe6 fxe6 20.Rxd8+ Kxd8 21.Bc1 Kc8 22.Rd1 Rd8 23.Rxd8+ Kxd8 24.g4 g6 25.h4 hxg4 26.Bg5 Bxg5 27.hxg5 Ke8 28.Kg2 1/2-1/2 Deep Fritz-Kramnik,V/Manama 2002]

4...d6 5.c3 Played to prepare the push d3-d4, taking over the center. X3D Fritz is still making moves from its opening library, also called its "opening book." This is a massive database with millions of positions take from previously played games and prior analysis. An opening book often takes up close to a gigabyte of hard drive space!

This opening book is how the X3D Fritz team tells their baby what to play in the first moves of the game, the opening phase. Before each game they try to predict was Kasparov will play. Then they look at the possible reactions that will lead to positions that X3D Fritz will play well when it leaves the book and has to think on its own.

There are so many millions of possibilities even in the first 10 moves of a chess game that this is far easier said than done. Even if you suspect what general direction your opponent is going to take, you can't prepare everything, there just isn't time. When human Grandmasters play one another they are both limited by their memories. No matter how much they prepare and study they have trouble bringing it all to the board with them come game time.

A computer player like X3D Fritz doesn't have this problem. Its memory is limitless and will store everything the programming team has time to enter into its opening book. It will play its moves instantaneously as long as it is still in a book position (as will humans!). One of the tricks Grandmasters use against chess computers is to play tricky move orders to get the machines out of their opening books as early as possible. The opening phase is very sophisticated and poor decisions made early will have repercussions ten, twenty, or even thirty moves later, far too deep for a computer to calculate.

The best opening moves have developed over decades and are based on Grandmaster experience and praxis. Computers don't usually do very well in this phase of the game on their own, the search tree of moves is too broad and the strategy too subtle. Therefore they use their opening books to reach playable positions, sometimes not "thinking" at all until move 20 or even later.

The debate about whether or not the use of these books is an unfair advantage for the computers has raged since the beginning. Kasparov put it back on the table after this painful loss, although he did quite well in the opening phase.

5...g6 6.0-0 Bg7 7.Nbd2 0-0 8.Re1 Both sides have made typical developing moves. Both kings have castled to safety and both sides put pressure on the critical central squares.

8...Re8 This move by Kasparov had never been played before in this exact position. It is hardly a shocking move and one that would likely be played soon regardless of the order. By playing it here Kasparov hopes (and succeeds) to get X3D Fritz out of its opening book. The machine is now calculating its moves on its own and still in a very early stage of the game.

9.d4 A Grandmaster likely would have waited a while to play this inevitable push in the center. White's pieces are a little tangled up. X3D Fritz sees no reason to delay; it wants central space. There is an immediate threat, the push d4-d5 attacking the c6 knight, which is now pinned by the bishop against the rook on e8.

9...Bd7 Breaking the pin, but guaranteeing that White can force the exchange of light-squared bishops. ..Nd7 was worth considering, but Kasparov doesn't mind exchanges that will decrease the computer's attacking potential. 10.d5 Ne7 11.Bxd7 (D2)

We like to say that exchanges help the defender, especially in positions in which he has less space. Black's pieces don't have much room to maneuver and that is exacerbated with more pieces bumping into each other.

But this capture is a good decision by X3D Fritz, although it was unaware of the real reason for this! In the coming position Black will attack on the kingside and White will advance on the queenside to take advantage of its central pawn wedge. In Black's kingside attack his light-squared bishop is a very important member of the army, but now it's back in the virtual reality box.

That attack is 20 moves down the line so there is no way X3D Fritz had even the slightest thought about that theory. It exchanged the bishops because it didn't like to keep a bishop that would be trapped behind its own pawns on d5 and e4.

11...Nxd7 "Black has already equalized here." – GM Seirawan. By a curious route we have reached a position that is basically a King's Indian Defense, a common opening against 1.d4. Kasparov doesn't play it anymore but for many years he used the King's Indian Defense (KID) with devastating effect.

Of course X3D Fritz has no idea that it's in a King's Indian, or that it's in New York or that millions of people around the world are watching. The KID is a strategically rich defense with very clear lines of play. Black attacks on the kingside and White expands on the queenside. Every human master knows this by heart.

X3D Fritz has no idea what it is supposed to do, it simply plays the move with the best evaluation on each move. Sometimes this can simulate a strategic plan, sometimes it can lead to inconsistent play by the machine. What would we see here?

12.a4 A good sign that X3D Fritz "understands" the position. It will expand on the queenside! This isn't an attacking move, there really aren't any direct targets to attack. But gaining space is a good thing on its own account. When your pawns control more area your pieces have more room to maneuver. Your opponent will be cramped and have trouble defending and guarding against your potential breakthroughs.

[12.Nf1 h6 13.Ng3 Rf8 14.a4 a5 15.c4 1/2-1/2 Langeweg,K-Spraggett,K/Zaragoza 1992/TD (63)]

12...h6 Kasparov is in no hurry. He wants to see if X3D Fritz will make a mistake in this subtle position, perhaps waste time. Meanwhile this little move protects the g5 square and prepares an eventual ..g5 pawn push and a kingside attack.

13.a5 a6 Black can't allow the pawn to continue all the way to a6. 14.b4 f5 15.c4 (D3)

X3D Fritz is playing disturbingly good strategic pawn moves! Pawn play, with its long-term weaknesses and lack of direct tactics, is far from a computer strong suit. They like to use their pieces in direct and open combat so their calculating power is at its best effect.

With all 16 pawns still on the board this brute strength is muted. But X3D Fritz is playing just like a human Grandmaster would in this position. It advances its pawns as far as possible before making any decisions with its pieces because it's still not clear where they will be best placed.

At around this point Australian Grandmaster Ian Rogers declared that X3D Fritz was passing the chess Turing test. You couldn't distinguish its play from a human's!

15...Nf6 White has its bind on the queenside with possible breakthroughs on b5 and c5. Black builds up forces on the kingside and will break with either ..g5 or ..f4. 16.Bb2?! The first sign that X3D Fritz is slipping in this subtle strategic war. This is a committal move when it's not yet clear where the best square is for this bishop. Bringing the queen out to b3 was a good alternative. [16.Bb2 f4 17.c5 g5 18.h3 Rf8 19.Rc1 Rc8]

16...Qd7 17.Rb1 Playing this rook to c1 to prepare c5 was what everyone was expecting. Is X3D Fritz wandering? It might have been looking at playing Bc3 and pushing b5, but it never does this. One of the problems computers have in strategic positions like this one is that they can't stick with a plan the way a human can. What looks good at one point might not make it into the program's search tree on the next move.

17...g5?! This is not objectively a bad move by Kasparov, the problem with it is that now X3D Fritz will have a clearer understanding of the position. In order to confuse the machine it's best to leave as many pawns on the board as possible and keep lines closed. That allows the human to make long-term plans and build up behind the lines, something computers aren't particularly good at.

After this move X3D Fritz has much clearer targets to work with, especially the e4 square. On the other hand, the typical attack Black plays for with ..f4 and ..g5-g4 isn't so strong in the game because Black no longer has his light-squared bishop. That piece is usually a critical attacker against the white king in the King's Indian, but it was exchanged off early on here.

[17...f4 The "move of principle" in this position. 18.c5 g5 19.Nc4 Rad8 20.h3 Ng6 21.cxd6 cxd6 22.Nb6 Qf7 23.Rc1 Nh7 24.Re2 h5 25.Rec2 g4 A standard King's Indian game with Black crashing through on the kingside and White doing the same on the queenside.

Two problems for Black here: the absence of his light-squared bishop and the fact that his opponent is a super-strong computer program! X3D Fritz just isn't going to get checkmated. Sharp counterattacking positions are not recommended against machines. 26.Rc7 Re7 27.Rc8 Forcing exchanges will remove most of the danger around the white king and increase the significance of White's positional advantage on the queenside.]

18.exf5 Or else the Black pawns will roll down the board. Now White has the nice e4 square for his pieces. 18...Qxf5?! The Black queen joins the battle, but without much scope in this blocked position.

[18...g4 Kasparov 19.Nh4 Nxf5 20.Nxf5 Qxf5 21.Nf1 h5 Similar to the game but gaining time over the game line.]

19.Nf1! Very nice. X3D Fritz redirects its knight to be able to control the e4 square.

19...Qh7 Maybe too subtle from Kasparov. He wants to keep the queen on the kingside to support his attack on that flank. He doesn't want to be forced back to d7 after ..Ng6 Ng3, for example. Kasparov also wants to keep an eye on the critical e4 square. [19...Ng6; 19...g4 20.Ng3 Qg6 21.Nd2 h5 22.Nge4; 19...Qg6? 20.Ng3 Nf5 21.Qc2]

20.N3d2 Focusing on the e4 square. 20...Nf5 [20...Ng6 21.Ng3 Nf4 22.Nde4 A powerful knight on e4 makes this position very pleasant for White.] 21.Ne4 Nxe4 22.Rxe4 h5 23.Qd3 Rf8 (D4)

Kasparov continues to build up his forces on the kingside. He will double his rooks on the f-file.

24.Rbe1?! Not a blunder, but X3D Fritz is getting distracted from its own attack. This is where human planning and experience are superior to calculation. A Grandmaster simply knows that in these positions you have to play where you are stronger. You must attack, not defend. 24.c5 was the normal move. [24.c5 Nd4!? (24...g4 25.Ng3)]

24...Rf7 25.R1e2 Defending f2 in advance. Again continuing his queenside initiative with c5 was better. 25...g4 26.Qb3 [26.c5 Raf8]

26...Raf8 27.c5 Finally this required advance, but why now when the queen appeared to come over to support b5 instead? Diffident strategic play from X3D Fritz after a fine start.

[27.b5 Ne7 Threatening ..Rxf2 because of the attack by the queen on the Re4. 28.b6 cxb6 29.Qxb6 Rxf2 30.Rxf2 Rxf2]

27...Qg6 (D5)

28.cxd6? Voluntarily giving up all the pressure built up over the last dozen moves. A terrible strategic decision, at least from an objective analysis standpoint.

From a computer-player perspective it makes some sense to liquidate and open lines for its pieces to create tactical chances. Indeed this pays off almost immediately when Kasparov blunders in mild time pressure on move 32.

[28.c6? bxc6 29.dxc6 Ne7!; 28.Nd2 Ne7=; 28.Qc4 Maintaining the pressure.]

28...cxd6 Recapturing with the pawn is the solid option. You don't like to have loose pawns against a computer. The worst is over for Black, but time pressure and perhaps that feeling work now against Kasparov.

[28...Nxd6?! 29.R4e3 e4 a) 29...h4 30.Bxe5 Bxe5 31.Rxe5 Rxf2 32.Rxf2 Rxf2 33.Qe3 Rf5; b) 29...Rf3!? A remarkable shot, suggest by X3D Fritz (in its analysis! 30.Qc2 (b) 30.gxf3?? gxf3+ Discovered check. 31.Ng3 fxe2 32.Rxe2 h4) 30...Qxc2 31.Rxc2 Rxe3 32.Nxe3 Rf7; 30.Bxg7 Qxg7]

29.b5 axb5 30.Qxb5 Bh6?! Kasparov activates his passive bishop, but this is a very slow plan and there are still many dangers in the position. With the threats on the queenside finally gone Kasparov seems to have lost his sense of danger. You can never relax against the machine because it never, ever relaxes. [30...h4 31.Qc4 Nd4 32.Bxd4 exd4 33.Nd2 Be5]

31.Qb6 A single-minded pin tactic, threatening Bxe5 (or Rxe5) because now the black d-pawn is pinned. 31...Kh7 Protecting the queen but creating other weaknesses.

[31...Rf6!? A remarkable move, offering the b7 pawn based on a spectacular (tactical continuation. 32.Qc7 (32.Rc4; 32.Qxb7? Ne3! 33.R4xe3 Bxe3 34.Nxe3 Qd3 35.Re1 Qd2 36.Rf1 Rxf2-+) 32...R8f7 a) (32...Bg5 33.Bc3 (a) 33.Qxb7 Ne3 34.R4xe3 Bxe3 35.Nxe3 Qd3 36.Bxe5-+) 33...R6f7 34.Qb6 Rg7 35.Ng3 Nxg3 36.hxg3=; b) 32...Qf7 33.Qxf7+ R6xf7 34.Ng3 Nxg3 35.hxg3 Rc8 36(a3=; 33.Qc4 (33.Qd8+ Kh7) 33...Rg7 34.Bc3 h4 35.Bd2 Bxd2 36.Rxd2-/+; 31...h4?? 32.Bxe5 dxe5?? 33.Qxg6+

31...Rg7 Kasparov 32.Ng3 h4 33.Nxf5 Rxf5 34.Ba3 Rf6 "Black is clearly better" - Kasparov] 32.Qb4 (D6)

Preventing ..h4 now. According to X3D Fritz programmer Frans Morsch, the program had been increasingly distracted by Black's kingside attack, leading to curious play like this move.

32...Rg7?? A horrible blunder that leaves the f8 rook unprotected. Such a thing was very hard to see because the rook was protected twice and now is not protected at all! Kasparov had around 10 minutes on his clock at this point, with nine moves to go to the time control on move 40 (when another hour is added to each clock).

You would usually like to have at least three minutes per move until the time control. Kasparov was now down to around one minute per move. That leaves precious little time to deal with the unexpected and blunders become more likely, at least for the human. Computers of course, never blunder at all. They don't play perfectly but they are simply incapable of losing a game in one move as Kasparov does here.

[32...Ng7; 32...Rg8 Kasparov 33.Ng3 Nxg3 (34.hxg3 h4! 35.gxh4 g3 36.fxg3 (36.f3 Bf4) 36...Qxg3 37.Qc3 Rf3; 32...Rc7]

33.Rxe5! Crushing. White exploits the pin on the d-pawn. A coincidental effect of White's last queen move. White wins a pawn but more importantly all of his heavy pieces spring to life against the open black king. There is no hope of ever defending this position against X3D Fritz. [33.Bxe5 Also winning handily.]

33...dxe5 34.Qxf8 Nd4? Opening more lines for White's pieces, but it's already too late to do anything more than delay the end by a few moves. [34...Rd7 35.Qc8 Qf7+-; 34...Re7 35.Qc8 Bg7 36.Ba3]

35.Bxd4 [35.Rxe5?! Nf3+ 36.Kh1 Nxe5 37.Bxe5 Rg8 38.Qe7+ Bg7 39.Qxb7 Qf5+/-] 35...exd4 36.Re8 Rg8 37.Qe7+ Rg7 [37...Qg7 38.Qe4+ Qg6 39.Re7+ Kh8 40.Qxd4+ Bg7+- 41.Qb6] 38.Qd8 Rg8 39.Qd7+ 1-0 (D7)

Final position

[39...Rg7 40.Qc8 White continues to threaten checkmates on h8 to win more material. The white a and d-pawns will run down the board to make new queens.

40...Rg8 41.Rxg8 Qxg8 42.Qxb7+ Kg6 43.a6 d3 44.a7 d2 45.Nxd2 Bxd2 46.a8Q]


other news

Opening ceremony photos and transcript
Analysis of game three
Analysis of game one
About Katie Horn
Report on game two from New York City
Q&A with Garry Kasparov
US Chess Trust | ChessBase