11 January 2007

Kramnik Had a Win

Sometimes your words come back to bite you. Writing for About.com on the 2006 Kramnik - Fritz match (The Last Man - Machine Match?), I said, 'At no point was Fritz in any real danger of losing a game.' A couple of blog posts -- Karsten Müller on the 1st match game Kramnik-Deep Fritz (www.chessvibes.com) and Counter-consensus on Kramnik vs. Deep Fritz (www.kenilworthchessclub.org) -- warned me that I was probably wrong.

The Chessvibes.com post had analysis by GM Karsten Mueller, while the Kenilworthchessclub.org post pointed to analysis by GM Yasser Seirawan in his ChessBase report, Seirawan on Kramnik vs Deep Fritz game one. The analysis starts with the diagrammed position. Kramnik played 30.a4 and followed up with 31.h3. According to the GM analysts, this was the wrong plan.

Bonn 2006, Match Game 1
Deep Fritz (Computer)

Kramnik, Vladimir
(After 29.Bb2-d4)
[FEN "8/5pkp/1p6/3Npp2/3b4/6P1/P3PPKP/8 w - - 0 30"]

Seirawan gave the straightforward variation 30.e3 Bc5 31.Kf3, threatening to win with Ke2, Kd3, Kc4, and then Kb5 or a4. His analysis continued 31...f6, the only move he considered, followed by 32.Ke2 Kf7 33.Kd3 Ke6 34.Kc4.

Mueller used a different move order, but arrived at the position 31...b5 32.Ke2 by transposition. Now he considered 32...e4 and 32...Kg6. The upshot of the analysis is that White wins the b-Pawn, while Black's attempt to find counterplay by raiding the Kingside falls short. The move 32...Kf8 loses to 33.Nc7.

I looked at another possibility: 31...Kf8, when 32.Nf6 doesn't work because of 32...Ke7 33.Nxh7 f6. White can complicate with 34.g4, but after 34...fxg4+ 35.Kxg4 Kf7, the Knight is still trapped. The straightforward 32.Ke2 also fails to 32...Ke8 33.Kd3 Kd7 34.Kc4 Kc6. Best is 32.g4; if 32...fxg4+ 33.Kxg4, Black's e-Pawn is in danger and White wins.

While I'm convinced that Kramnik had a win, I'm still not convinced that people have any chance against the best machines. Just to stay in the game they have to match the computer at playing near-perfect tactical chess, then have to outplay it with near-perfect positional chess. Kramnik, the best shot we have at beating a computer these days, tripped on both tactical and strategic lines. The next time that he plays Fritz, or that some other top GM plays another computer, the machines will have made a little more progress. There is no evidence that they have reached maximum strength.

To play through the complete game see...

Vladimir Kramnik vs Deep Fritz (Computer) 2006
http://www.chessgames.com/perl/chessgame?gid=1440787

...on Chessgames.com.

3 comments:

Wahrheit said...

To me, chess against a machine is just not very interesting, because emotion and fatigue and error and personality and the other human factors are what make it the greatest game.

You might compare it to motor racing--no doubt computers could be installed in the cars to drive beautifully, with nearly perfect reactions to events on the track, but would thousands turn out to watch? I think not.

Computers are tools, not chessplayers. They don't play the game. It might be interesting to see the result if the computer is not allowed millions of opening moves in memory, no endgame tablebases and a gradual reduction in strength after the third hour of play to simualte fatigue. That could be an intriguing experiment.

Mark Weeks said...

I'd like to see how the machine performs if you disable its opening book. Let it play its own moves starting from the first. The situation is even worse in the man-machine matches, where the machine is aided by a GM-level opening specialist who prepares an opening repertoire designed to cause maximum pain for the human opponent. It's not man vs. machine, it's man vs. machine's team.

kindredspiritks said...

Hello! I have 3 chess programs that I use mainly to play some fun games, check some of my annotations to games on my blog. I agree that playing a computer is not as satisfying as another human opponent. Not that I want to see them squirm, I just like people more than machines!
Don R. kindredspiritks.wordpress.com/2008