|
|
Match vs Rybka - Mexico 2007 |
|---|
In September 2007 my program Zappa played a 10 game match with Rybka, the reigning World Computer Chess Champion, and defeated it by the score of 5.5-4.5, winning a prize fund of $10000. This came as quite a surprise to the computer chess world, as most spectators had predicted that Rybka would win by a large margin. Writing these reports is always more fun after a victory, and it's especially sweet whan no one gave you a chance to win in the first place. I spent about a week in Mexico, and aside from nervously cheering on my program I also did some sightseeing, met some GMs, and generally enjoyed myself.
 |
Mexico City really reminded me of a flatter Harlem NY. There were some really nice places, but most things were just a bit run-down, like New York. Still, aside from Spanish advertisements and widespread graffiti I didn't feel that out of place. Many of the billboards hawk American products, and both Americans and Mexicans drive Japanese cars. Mexicans, however, are very passionate; it was tough to walk more than a few blocks without seeing a couple playing tonsil hockey. I offered to help our waitress practice but she demurred. We also went to see the Sun and Moon pyramids. Somehow they had a slightly malevolent air, not something I usually associate with geometric shapes. The faces are mostly smooth but have large numbers of jagged rocks protruding a foot or two. Erdo and I speculated that these were added to bash in the skulls of unenthusiastic victims who tried to slide down the pyramid to freedom.
I had a long discussion with Frederic Friedel (of Chessbase) and Vasik Rajlich (the author of Rybka) about Libertarianism and human sacrifice. I tend to convert the people I meet to Libertarianism. It is easy after you have read The Vision of the Anointed and The Fatal Conceit. The basic idea is that some cultural traits work better than others. For example, human sacrifice not only kills your own people who, even if they are slaves, are the greatest source of a nation's power, but also tends to annoy the neighbors, which is why Hernando Cortes was able to enlist thousands of Indians help his few hundred Spaniards against the Aztecs. Although slightly less obvious, Socialism works just as poorly; history has shown that civilizations that have extensive government control over their economies usually do worse than those that let their citizens determine their own allocation of resources. While most socialist policies lead to obvious benefits, the drawbacks are less obvious but more potent.
We were quartered with the rest of the participants in the Sheraton Centro Historico, an excellent hotel except for the loud aerobics class two floors down that awakened me every morning at 6:30. Unfortunately I only met a few GMs, but I was pleasantly surprised that they were not antisocial maniacs. In fact, except for extraordinary chess skill, they seemed pretty normal. Erdo amused himself by analyzing their games with Zappa and then asking them at the press conferences whether they had seen this or that complicated line. I was sort of surprised, though, that aside from this harassment the only one who really seemed to care about the computer match was Anand. The future of human chess is somewhat uncertain; even if many people are still interested when computers play vastly superior chess there is always the issue of cheating. The usual analogy is with running; we still have the 100 meter dash at the Olympics 150 years after the debut of the automobile. But running is a much more fundamental activity than playing chess. I would be sad but not surprised if there were no big chess events at all in another 20 years. I guess I don't blame the rest of the GMs for not wanting to stare into the void.
 |
Nevertheless we are still quite a ways from perfect chess, and both Rybka and Zappa have gotten a fair amount of criticism for some of the weaker moves in the match. Pretty much everyone involved with the match made extravagant claims about how these machines play at 3000+ elo - and they do - and yet in the game analysis you will see that I have annotated a lot of blunders. 5 decided games, and two more very close draws, almost require a large number of mistakes. But I think that some bad play was simply inevitable: the stronger your opponent, the worse you are going to look. Against a weaker opponent, Zappa (or Rybka) would get a good position very quickly and not have many chances to look stupid. But when the game involves a lot of positional maneuvering in equal positions you have more chances to look bad. However, these blunders - and some very complex games - resulted in a very spectator friendly match with a lot of action. I hope that this match served to increase public interest in computer chess a little. While there is less of a human element, in my opinion computer games are just as fun to watch as human games.
I also thought the match did a very good job of showcasing the differences in playing style. Zappa is more dynamic, while Rybka is more positional, sort of like Kasparov and Karpov. To me it is a little surprising that the personalities and styles of the authors are reflected in the play of their programs, that abstract concepts like style can be transmitted into concrete lines of computer code.
Game 1: A Ruy Lopez. Zappa was White and launched what appeared to be a dangerous attack on the kingside. Rybka defended precisely and reached an endgame where Zappa had the bishop pair and connected passed pawns while Rybka had a more advanced extra pawn and more centralized pieces. I felt Zappa was a bit better and had the chance to give Rybka some problems, but it never really succeeded and the game was drawn.
Game 2: Another Ruy Lopez. Rybka achieved a strong position out of the opening, and Zappa gave a pawn for counterplay that never really developed. Eventually Zappa was just a clear pawn down, and after it missed a draw with 57... Bxe5, when Black has a strange kind of fortress, it went down in flames.
Game 3: Yet Another Ruy Lopez. Rybka blundered with its first two moves out of book and Zappa was able to establish a dominating position, when Black's passed pawn on the queenside was actually weak due to White's d5-e4 structure in the center. After finding a few strong moves in the nick of time, Zappa managed to win this pawn, but its exposed king still gave Rybka reasonable drawing chances. Fortunately Zappa played very well and was able to take home the full point. 38. f5! imprisoning Black's bishop basically gave White an extra piece, and eventually this superior force won Rybka's d6 pawn and the game. The final position is actually quite amusing. You have to love Black's bishop on G8, a relic of of the 80s when amateurs could still beat supercomputers.
Game 4: Yet Another Ruy Lopez. By this time some spectators were actually complaining about the openings. Rybka came out of book with a completely won position after Zappa missed gxf5, and used its queen to tie down black's entire army of two Rooks and a Bishop. But just when it was on the brink of a brilliant win, it missed 52. Qe1! (expected by Zappa) which would have ended Black's resistance. Both sides shuffled and the Rybka team finally offered a draw when it was clear Zappa had established a fortress. Erdo, however, smelled a weakness and cleverly insisted on playing on. When Rybka reached the 50 move rule, it gave a pawn to avoid the immediate draw. And then another pawn. And another. And suddenly the Black rooks stormed back onto the board, won the remaining two white pawns, and eventually the game. An amazing comeback.
Game 5: The Rybka team finally decided to avoid the Ruy Lopez with 1... c5. Zappa played 3. Bb5, a line that had given Rybka some trouble at the 2007 World Championship, and after some maneuvering Rybka blinked first with f5?, which gives Black a permanent weakness on the e-file. Even so, it probably could have drawn but after 61... f4 it is curtains.
At this point the Rybka team had lost 3 games in a row after going over a year without a single loss, including seven tournament wins. They were also starting to sound a bit less confident and posts like "We're going to play with Rybka 2.3.2a just to give Zappa a chance to get on the scoreboard" were becoming less common at the Rybka forum. I arrived in Mexico for the second half of the match just minutes after the end of Game 5, in time to see Dagh shaking in disgust and Erdo shaking with glee. Zappa might have been feeling a bit nervous, because after my arrival it did not win another game.
Game 6: Rybka repeated the line from Game 4, but Erdo had discovered a hole (gxf5 instead of Kxf8) and was able to force a draw. Zappa played less than 20 moves and had a 0.00 score on every move.
Game 7: At this point the Rybka team was growing desperate because they needed two wins in four games just to draw the match. In this game they tried a Sicilian Dragon with 2... Nc6 in order to avoid a repeat of Game 5. Zappa made a few silly moves in the opening (a3, Qd3 when Bf3 or Qe1-h4 are typical) and Rybka was able to equalize, but later in the game chose a poor plan with a5-a4-axb3. This gave Rybka a passed pawn in the center, but Zappa connected passed pawns on the queenside. In an incredibly sharp position, Zappa was able to trade its connected passed pawns on the queenside for new connected passed pawns on the kingside, and obtain a winning advantage in the process. Sadly, just when it could have delivered a knockout blow with hxg7, grabbing a Rook and winning in 33 moves, it blundered with Ra3 and went into an optically superior but easily drawn ending. I actually tracked down the reason that it missed this move, and it is an evaluation error that occurs in approximately 0.000001% games.
Game 8: Yet Another Ruy Lopez. Rybka blew what little advantage it had when it played Nf5 (Zappa wants Rf1 and blood on the kingside). After b3! Zappa was probably better, and ended up with a Rook vs Bishop and passed pawn. After some shuffling Zappa finally managed to break the deadlock and establish some winning chances, but was too low on time and blundered twice (Qxb5 and Qd8) and went down in flames. The final position is amazing: Rybka is down a full exchange, but the Queen and a7 pawn paralyze Black's entire army and White can simply play g5, g6, Kg4, Kh5, Bxh6, Kxh6, and g7 mate!
Game 9: The Rybka team continued their attempts to mix things up and played a Caro-Kann, with the intent of allowing the Deep Blue/Kasparov sacrifice Nxe6. Erdo played a different line, and play continued into a fairly quiet middlegame. Unfortunately, Zappa had a small bug and gave itself a bonus for pinning its bishop to its Rook, and Rybka unleashed an intricate combination that won a pawn and promised considerable winning chances despite the presence of opposite colored bishops. Unfortunately, at move 71 Rybka played f4??, an absolutely horrible move which allowed White to establish a blockade. Mark Uniacke called it one of the worst blunders in modern computer chess. I had actually left earlier, sure that Rybka would win, to go exercise. I was shocked when Erdo came back to the hotel and said that we were still up a game.
Game 10: At this point Zappa was up by one game and needed a draw to win the match. Rybka played Nf3 and Zappa ended in a hedgehog defense, which was probably not too bad until Zappa played b4?? leaving itself with no counterplay. After several moves Zappa was panicking and every PV contained Kh1, Rg1, g5, and other such violence on the kingside. Fortunately, Rybka let Zappa off the hook with a3? and Zappa managed to draw the game, win the match, and collect the $10,000 in prize money, which, even after I gave half to Erdo, raises my hourly wage for Zappa work to slightly more than that of a Mexico City janitor. This is actually the first time the Zappa team has ever won prize money of any kind.
Jorge Saggiante for sponsoring and Paul German for organizing the event. They both did a fantastic job in a tough situation, and they both have excellent taste in suits. Paul did fail to get my foosball session, though.
Harvey Williamson for supplying the hardware. We had arranged for two 8 core computers to be shipped to Mexico City, but they were tied up in customs. Fortunately Harvey stepped in at the last minute with a big computer and some good fish recipes (Rybka means "little fish"). I cannot vouch for Harvey's taste in suits, but since he is a member of the BBC I suppose he has a good wardrobe.
Erdo, whose decisions to play on in games 4 and 9 probably gave Zappa a full extra point.
Whoever uses the handle "sidserious" on the Rybka forum. His torrent of deranged posts provided a critical distraction for the Rybka team during their moment of crisis. Also thanks go out to the user "turbojuice" who seems determined to educate the unwashed, knuckle-dragging masses of Rybka users that Zappa is a decently strong chess engine if you give it time, CPUs, and a reasonable opening book.