Yermo vs. the Machines

By GM Alex Yermolinsky

GM Alex Yermolinsky utilizes ICC as part of his professional training regime. His public training sessions against ICC computers are well known, and extremely popular among the "man vs. machine" fans. In this issue, Yermo gives us his perspective on chess computers and their impact on professional chess.

How we match against computers is a burning question of our times.

In the eyes of the general public chess as the game has been "solved" when World Number One, Garry Kasparov, was defeated by Deep Blue in the notorious 6-game match. One story stands out as a striking example of the damage suffered by our beloved game.

A Japanese player was invited to participate in the World Knockout Championship in Groningen at the end of 1997. He came to his boss asking for a leave of absence. The boss, who had read about the GK-DB match, turned him down. He said that the Company now considers chess as a useless hobby for its employees, and suggested switching to Go or Chinese Checkers, any game where man can still be competitive versus the machine. Apparently it has never occurred to the enlightened leader of Japanese business community that the very reason for Garry’s failure was the sheer complexity of chess. Unlike other table games where common sense and a little practice can make one virtually unbeatable, chess is just too deep and complicated. It’s so easy to go wrong in chess, and one mistake is all it takes to lose. This sad truth holds even for the best.

Without a profound analysis of the games from that match it’s hard to make any conclusions, but one thing seems clear: Garry made a costly mistake of abandoning his game in favor of ill-conceived anti-computer strategy. The roots lay deep in the history of Man vs. Machine chess encounters.

In the old times ( some 10 years ago ) humans could easily get an upper hand by adopting a certain sit-and-wait policy. With no opening book to guide them the computers would self-destruct by committing horrendous positional errors, leaving the outcome of the game as a matter of simple technique. In such situations a human GM could avoid any tactical confrontations due to the superiority of his position. Later, when the computers caught up in the opening department, it became more even. Many middlegame positions with open center and developed pieces for both sides turned out tough to handle against the computers, who operate with "hard science" of calculation, the sheer power of which allows them to make up for the lack of other aspects of chess mastery.

Naturally, the humans smarted up. Keeping the position closed became a way to go, and for 3-4 years a b3/g3 kind of openings won the recognition of "winning strategy". Some relatively weak GM’s became accomplished "computer-bashers", up until the hardware grew in power. Poor Garry didn’t sense the moment and came out against Deep Blue armed with yesterday’s knowledge of computers potential. He voluntarily transformed himself into an average 2500 GM, with no contemporary opening repertoire and a "game plan" of avoiding any confrontations, hoping to win with no risk taken.

I say, these days it’s unrealistic to expect success against the computers if you just sit on your position doing nothing. No big mistakes will come, and little ones will go unpunished if the human plays scared. He has to go after the monster to put himself in a position to capitalize on its mistakes before computer resourcefulness will come to the rescue. A Grandmaster has to play like a Grandmaster with all the risks he usually takes, bringing out the best stuff he’s got to just have a shot of beating the computer.

This especially holds true in shorter time control games. The probability of making a time-trouble mistake grows with a number of moves made, so the best policy seems inviting the crisis while you still have time to handle the tactics. The computers tend to underestimate the danger to their king until it’s too late, and they also have a great deal of love for meaningless material gains. All these can be exploited by principled and energetic play.

For the past two years I was playing a lot of blitz here on ICC, taking on all Crafty clones and such. The results are self-evident: while they are still up there with 2800+ ratings, "Sweere" has been fluctuating between 2500 and 2600. That’s OK by me, as I only value these games as training. For that purpose my style of anti-computer play serves me good: I always get tactical skirmishes on the board to sharpen up my calculating abilities. Hopefully, it will help me to achieve my goals in the "real" chessworld.

(1) Topolino (2675) - Sweere (2564) [B66]
ICC 6/5/99, 5 0 ICC 6/5/99, 5 0, 1999
[Yermo]

This browser is not Java-enabled.

1.e4 c5 2.Nf3 d6 3.d4 cxd4 4.Nxd4 Nf6 5.Nc3 Nc6 6.Bg5 e6 7.Qd2 a6 8.0-0-0 h6 9.Be3 Be7 Main line Richter-Rauser. In general, I am happy with opposite side castling against a machine, as it accelerates the tempo of play. 10.Be2 More usual are 10.f3 or 10.f4, but on the next move we transposed to a theoretical line anyway. Nevertheless, I view the computer's 10.Be2 as an encouraging sign. The depth of its opening knowledge turns out very shallow. No surprise there - computers don't play the opening in a GM sense; instead they can play some opening moves without extending their plans into the middlegame. Good enough to handle a 2.c3 Sicilian, but hardly sufficient in sharp lines. 10...Bd7 [10...Nxd4 11.Bxd4 b5 is another way of developing Black's play. The threat of ...b4 may induce some passive reaction, 12.f3 or 12.Bf3, typical of computers' superficial approach to the opening play. Our experience in the Sicilian speaks in favor of the pawn sacrifice 12.f4!? b4 13. Na4 that indeed has been played in numerous tournament games. I didn't think Topolino would do that, and I was more concerned with 12.Qe3. In any case this plan would be interesting to check out in my future battles against ICC computers.] 11.f4 b5 12.Bf3 Rc8 This position was all rage some 5-7 years ago. Kamsky used 13.Kb1, a typical prophylactic move, to beat Atalik in Tilburg 1993. 13.Nb3 Na5! Much better than 13...Qc7 that I played against Dvoirys in 1988. Black needs to open the c-file for his counterplay and renew the ...b4 threat now, as the Bd7 controls the a4-square. The only problem is White's e4-e5, and Topolino is on top of things. 14.e5 Nc4 [14...b4!? also deserves some attention here.] 15.Qd4 dxe5 16.fxe5 Nxe3! Now we begin to appreciate Kamsky's Kb1. The tactic along the c1-h6 diagonal bails Black out of a serious mess. 17.Qxe3 [17.exf6? Bxf6 18.Qxe3 Bg5-+] 17...Ng4! Same tactical theme. Black needs to avoid passive moves like ...Nh7 at any cost. A human player operates from experience and understanding of the essential needs of the position that just happens to be a little too much for the computer to handle correctly. If 14.e5 fails to produce its primary goal, which is driving Nf6 to a passive square, then White is left with nothing but a glaring weakness on e5. Surrendering the bishop pair in the process only aggravates the situation. 18.Qe2 b4! A logical continuation of Black's previous play. The white knight has to squares to go. 19.Ne4 costs the e5-pawn, and 19.Nb1 leaves the king in jeopardy - 19...Bg5. [18...Ne3 19.Rd3 Nc4 20.Bb7 would give White powerful initiative.] 19.Bxg4? [19.Bb7!? would be interesting. Nevertheless, the position after 19...Rc7 20.Ne4 Nxe5 21.Bxa6 0-0 has to be good for Black.; 19.Nd5! is what any of my GM colleagues would play without much thinking. White needs to change the pawn structure to get new targets for his pieces. 19...Bg5+ 20.Kb1 Nxh2 21.Rxh2 exd5 22.Bxd5 Qe7] 19...bxc3 20.Qxa6 Tell you a little secret: only with this move we branched out from the game we played earlier same day. [20.bxc3 was played by Topolino in our first game. White's extra pawn didn't scare me much, but the opening subtleties took a large chunk of my 5 minutes, so I rushed to trade the queens. 20...Qc7 21.c4 Qxc4 (21...0-0 followed by ...a5 should give Black an overwhelming attack.) 22.Qxc4 Rxc4 23.Rd4 Rxd4 24.Nxd4 Bg5+ 25.Kd1 Bf4 26.Be2 (26.Nf3 would defend the pawn, but misplaces the pieces. 26...Bc6 27.Ke2 h5 28.Bh3 Ke7) 26...Bxe5 27.Nf3 Bc3 28.Bxa6 Ke7 Here I only had about 1.5 minutes left. Black has a large advantage of course, but winning this ending with that kind of time on the clock surprised me. I guess I was quicker than usual. Nevertheless, I'm going to skip the rest of the game.] 20...Qc7 Topolino's improvement, Qxa6 came in the wrong moment. I only took seconds to repeat the opening moves and was poised for a quick kill. 21.Qd3 cxb2+ 22.Kxb2 [22.Kb1 0-0 and Black quickly switches to the a-file.] 22...0-0 23.c3? [23.Rhe1 was better. I think Topolino considered the e5-pawn untouchable.] 23...Qxe5 24.Rhe1 Qc7 25.Rd2 Shifting gears to the defensive, but it's way too late for that. [25.Qxd7 Qxc3+ 26.Kb1 Qc2+ 27.Ka1 Bf6+ 28.Rd4 Ra8 Five moves into Black's "sacrifice" and White's position is perfectly resignable. Can't Topolino calculate that deep? If so, how do you explain its 23rd move that left the e5-pawn en prise?] 25...Bf6 26.Rc2 Bc6 27.Qh3 Bd5 28.Rd1 Ra8 29.Ra1 Rfb8 I'll spare you the next sequence of moves that began with 30.Be6!? Don't worry, I had plenty of time to collect the booty and bring the point home.