Kramnik vs. Deep Fritz

Vladimir Kramnik will receive a lengthy section in any book devoted to history's greatest chess players. He's been a top grandmaster for close to twnety years. He defeated the seemingly invincible Gary Kasparov in a straight-up match. He has successfully defended his title twice, both times coming from behind.

So try to imagine the sting that comes from knowing the following position will forever be placed just below his name:




This is the penultimate position from the second game of Kramnik's ongoing match against the top computer chess-playing program Deep Fritz. The dust has settled from a difficult middlegame, and Kramnik was doubtless wondering if his strong queen-side pawns and active pieces could be put to good use. He played the seemingly plausible 34. ... Qe3.

No matter how big a patzer you imagine yourself to be, you might enjoy trying to determine, before continuing, why this is such a wretched move.

Did you notice that white can now play 35. Qh7 mate? If you did, you're one up on Kramnik.

That's right. For what might be the first time in the history of the game a World Chess Champion overlooked a mate in one. As a lowly Class A player, I find this vaguely inspiring. I mean, I make horrible, game-ending blunders on a regular basis. It's nice to know the GM's are capable of such things as well.

The present match is the latest in a long line of human vs. computer matches. Still the question lingers: Do computers play chess better than the top human players?

Up until 1997, the answer was an unambiguous no. Then IBM's computer Deep Blue defeated Kasparov in a six-game match. This vicotry notwithstanding,most people were still inclined to give the benefit of the doubt to the humans. Kasparov resigned a drawn position in game two, and he lost game six only because of an elementary blunder most amatuers would have avoided. Kasparov's loss was chalked up to the fact that humans can crack under pressure. Computer's can't.

Kasparov has played two subsequent matches against computers, both drawn. And Kramnik himself drew an eight-game match against a computer in 2002. Looking at the games, however, tells a slightly different story.

In Kasparov's most recent match, both players won one game and the other two were drawn. Looking at the two decisive games shows that Kasparov won a positional masterpiece. He was able to steer the game into a closed position in which long-term, subtle maneuvering to set up a breakthrough was essential. These are precisely the sorts of positions computers have difficulty playing. Kasparov nursed his advantage and crashed through convincingly. But the computer won as the result of a terrible blunder on Kasparov's part. Prior to that blunder, Kasparov held a slightly better position

Likewise for Kramnik's previous match. Kramnik won two games, the computer won two. Kramnik won by steady and convincing play. But one of the computer's wins came when Kramnik blundered away a simple draw in a fashion almost as spectacular as that given above. The other win came when Kramnik played into a speculative, sacrificial attack. A sane human being would flee the building before having to defend against a Kramnik attack. But the computer took it with its usual stoicism, cranked out a series of brilliant and counter-intuitive defensive moves, and won the game. An impressive performance, but Kramnik had played right into the computer's strong suit. You could still make a case that the humans were the better chess players.

Then came the 2005 match between British GM Michael Adams and the computer program Hydra. There were six games. Hydra won five of them. Adams had to struggle to draw the sixth.

Adams was a top grandmaster, and he got demolished. But even after this most people were inclined to give the humans the benefit of the doubt. Adams was plainly unprepared, and didn't treat the match with sufficient seriousness.

Which brings us to the present. Kramnik was dominant in the first two games. He was apparently winning in game one, and held a small advantage through much of game two. He drew game three easily. Yet he is down a point.

So what's the verdict? My view is this: Humans understand chess better, but computers play better.

There is such a thing as perfect chess. In any given position there is a unique best move, or a few moves that can be said to be equally good and better than any alternatives. Give five top grandmasters and five top computer programs that position, and the humans are more likely to find the correct move. Computers still have too many weaknesses born of their rigid evaluation functions and limited horizon.

But when it comes time to sit down and play, the computer's lack of nerves and emotions makes it very difficult to beat. The computers typically play super-solid, ultra-conservative chess. They create no weaknesses, even entering into slightly inferior and passive positions. But they are ready to pounce on the slightest error. Yes, Kramnik only lost due to an incredible blunder. But avoiding blunders is simply part of the game.

Playing defensive but solid chess while waiting for your opponent to crack is a perfectly honorable approach to the game. The second world champion Emmanuel Lasker was famous for it, as were the more recent champions Tigran Petrosian and Anatoly Karpov. That the computer wins so often by this approach does not indicate that it is getting away with something. It indicates simply that it is better at playing chess than the top human players.

In this and any future matches, the computer has to be regarded as the favorite. But when it comes time to get at the truth of a position, human grandmasters have more to contirbute.

Tags

More like this

I honestly think it takes something out of chess to know that within the century, a computer will probably calculate out all possible games of chess and determine whether white or black will always win given two perfect players. As complicated as chess is, it is in principle possible to solve in a game-theory sense. It would be an interesting project for a game designer to try to develop a game involving many of the same skills as chess but without that feature.

It would be an interesting project for a game designer to try to develop a game involving many of the same skills as chess but without that feature.

In general any game involving causative heuristics like chess is perfectly mathematically predictable, and most chess AIs can already practically do what you describe. It's actually pretty cheap to compare chess AIs against human players, since humans operate purely on pattern recognition rather than the combinatorial game analysis used to program chess AIs. A real challenge would be to develope and AI that operates on pattern rocognition a la a real human player.

But if you want a cool game chess players can totally get into, I've been addicted to Pentago for a while. Google it.

By Tyler DiPietro (not verified) on 29 Nov 2006 #permalink

Thanks for posting this.

My feeling is, and I think this is what you are noting, there are two kinds of chess- human and machine- and they can't really be compared. Or, if the computers improve, they won't be comparable. I am content to enjoy grandmaster play even if a computer might play better.

You might want to try Japanese chess (Shogi) and it's many variants. Although there are computer programs that play Shogi & variants, Japanese chess is much more complicated since it involves "drops" (re-introducing captured pieces back into the game). Google Shogi variants for more info.

At the press conference, Kramnik claimed to have calculated 34.- Qe3, that is, the error was in failing to see the mate some moves in advance. Still, a gross blunder!

It?s a bit ironic, this match seems to be the opposite of the Topalov-Kramnik match earlier this fall, where Kramnik was up 2-0 after the first two games despite blundering into a quick mate in game 2 (a mate that Topalov didn?t see).

By Pastor Bentonit, FCD (not verified) on 30 Nov 2006 #permalink

Jason says:

This is the penultimate position from the second game of Kramnik's ongoing match against the top computer chess-playing program Deep Fritz.

*Nitpick Alert*

The position posted is actually the antepenultimate position. The penultimate position is the one next to the end, or after 34. ... Qe3??.

As to how this could happen, I've done similar (although obviously on not nearly so rarefied a level!). White makes a move which is forced as black just captured a rook. It sometimes doesn't really register that just because a move is forced it doesn't mean it can't be threatening too. This is doubly so when you know, and your opponant knows, that he's fighting for the draw. The idea he might suddenly win sometimes doesn't enter the picture till it's too late.

As discussed on the Chessbase site, if you're on Kramniks side you're focused on those two queenside pawns, knowing that if you exchange queens or even get your queen down there that you could probably score the touchdown on a1. Also, this particular mating pattern is not common. Now if the N had been on f6 or g5 (assume for the sake of argument the N wasn't hanging) Kramnik would have instantly seen the threat without a doubt. But outside certain problem positions, knights and queens rarely find themselves in that configuration.

Ouch! Thanks for posting that.

I've played through grandmaster games a few times with a move accompanied by a comment along the lines "X makes a gross blunder", which left me wondering whether I would have spotted the error even on a good day.

This one must indeed be unique. (It took me barely two seconds to spot the mate.)

Dave S.-

I stand corrected.

And you're right about this particular queen+knight pattern being rather unusual. Backward moves in general can be hard to spot.

It's a very common sort of blunder in chess to become so caught up in your own plans that you forget to look at what your opponent can do. One time I was playing a blitz game against former US speed chess champion James Sherwin. The game was played in the last round of a tournament at the chess club I attended at that time. Sherwin had perviously beaten everyone else in the tournament, and I was his last obstacle.

We sat down to play, and by some fluke I quickly got an advantageous position. My advantage grew and grew, and eventually I won a pawn. Sherwin's position got worse and worse. And then I saw it. Mate in three. Check, check and mate.

I calmed down. Examined the position again. I play check. He has only one way out. I check him again. Once more he has only one way out. Then it's mate. I couldn't quite believe it, but I was about to beat the great Jimmy Sherwin in a blitz game. I played the first move of my combination.

And Sherwin promptly captured my king. I hadn't noticed that I was in check, and in blitz chess you can simply capture the king and win the game when that happens.

That's why I don't play chess for a living!

I once had a similar experience against one of the top players in the country. Although I did see that I was in check the result was the same ... as I unfortunately chose to get out of check by castling, which is of course illegal. He took my king too, and game over.

Sometimes blindness works the othe way. Long ago when I was a C player I was up against a strong A in a team event. I played the Pirc and he put me under a tremendous attack, sacrificing a piece to open the h file. Evetually I got a breather by holding off all the immediate threats with N-h8 (how often does Black play that in the first twenty moves?) He made a developing move and on my next two turns I gained tempi by bringing out my queen and queen's bishop, each with check. Suddenly my teammates were patting me on the back and shaking my hand: that second check was mate, and all I had thought about was developing with gain of time.
I'll disagree a bit about Lasker. I don't think he ever played a Petrosian kind of waiting game. It is true that he often won when his opponants cracked, but that was because he took them into difficult postions (some slow manouvering, but also a lot of wild tactics like in his game with Napier) where he counted on seeing thngs the other player didn't. I think his true distinguishing feature, and what separates him from Kramnik, was his iron nerve. He almost never blundered. In very tight spots he just played on with incredible tenacity until he found a hidden resource the other player missed. Look at that 1924 game with Eduard Lasker. How many players would have gone into the endgame with knight against rook and pawn? Or the way he found the win against Alekhine with two rooks and pawn against rook, knight and paawn when the pawns weren't even passed. I think the most important element of his style was his willingness to run great risks in order to create winning chances. In this way he resembles Kaparov much more than Kramnik or Petrosian.
So excuse my rant. I am just a big fan of the immortal Emmanuel.