Kramnik v Deep Fritz

Match starts today, described here. Kramnik gets a cool mill if he wins, $500,000 if he loses.

Place yer bets!

And also: insert your own witty caption for this shot of Kramnik and his lady friend:

These 6 game matches are a joke, really. The match will probably end in a draw, and, with Kramnik’s rather dry game, it may end in 6 draws.

Kramnik to win by one match, all others are draws.

After victory he picks up the mouse and slaps the monitor repeatedly shouting “Look you are hitting yourself, you are hitting yourself.” He wrenches the mouse out of its socket throws it across the room then high kicks the monitor off the table screaming “Who’s the monkey now? Plug sucker?” before storming off to the stunned silence of the crowd.

Thinking “If I win, She lets me sleep in the same bed as her tonight!”

Why are there evil red tendrils shooting out of that dude’s glass and about to embed themselves in his face? Is this some sort of hentai thing? Or are those just some wicked cocktail straws?

Also, did anyone else read this thread title as “Kramer vs. Deep Fritz”? I was imagining Michael Richards shouting the N-word at Deep Fritz.


Have you ever reviewed a chess game Tom?

Because if you did, you would totally know all about the red face impalers.

“Hmmm, I think she’s starting to suspect I’m not really the guitar player from Oasis. Got. To. Think. Of. Something.”

Thinking: “At least half a meellion dollars and a chess groupie? Win or lose, I’m still a winner.”

“God, I hope I don’t lose and have her leave me for the computer.”

Yes, at the 47th, 82nd and 94th minute you may take one of the sticks and shove it into your opponent’s eye. That’s an en passant and a totally legit chess move. If you miss or your opponent is still able to play on, one of your pieces in play does not untap on the next turn.

Not an exciting first game. Fritz kept trying to start something but Kramnik just kept exchanging down material. You can replay the game from this link:

Kramnik somewhat wins if he can draw every game.

So are these Chess computers actually playign with real strategy? Or do they just brute force a best move at every point in the game?

Brute forcing the best move at every point in the game IS a real strategy. It quite possibly is the best strategy.

The only chance the human player has is to try open the board up to the point where the brute force approach takes too much computational time to look ahead enough moves to become undefeatable.

What do you mean by “real strategy”? They are certainly not brute forcing the game tree; they are deeply searching parts of the tree according to both conventional heuristics (i.e. “real strategy”) and board evaluation rules, and some fancy search algorithms that know when to probe deeply into a tree and when to look more shallowly. They do have large scale objectives like pawn structure and development, and pressure on opponent positions and so on. These programs are written by programmers who happen to be chess masters with the assistance of consulting grandmasters, so it’s not as if they just bake breadth-first-search with a piece count and the basic rules into some ASICs and just throw hardware at the problem – though they do use custom hardware arrays.

But it’s an old truism in AI that if you can understand what the program is doing, it’s not intelligent; and what a chess computer does is fully understood, whereas what a grandmaster or even a patzer does is not very well understood at all.

Chess programs use a myriad of traditional and non-traditional strategies, brute force, heuristic algorithms and other forms an analysis; when they do one-on-one matchups against specific opponents, they deep analysis of the particular opponent to influence decision making (they did this with Kasparov a lot).

— Alan

Well, and even the best computers can’t look more than 2-3 steps down an open chess board. There was some estimation (maybe Wikipedia) saying there was 10^120 possible moves in a chess game. By comparsion, there’s estimated to be around 10^90 protons in the universe. So brute-forcing chess games isn’t ever going to work due to pure memory. But, of course, “advanced look-ahead” will always be a major advantage to computers. Still, as previous posters have said, there’s a lot of work that goes into maximizing the (very large) amount of memory they have to work with.

Apparently, Kramnik fell victim to either thinking about the chick or the red face impalers.

In Game 2, he made what might be the biggest brain fart of all time in chess at this level, losing to a mate-in-1 that I could have avoided.

Game replay:

This is true though somewhat misleading. Human players often look much more than 3 moves ahead as they disregard bad moves to begin with using their advanced pattern-matching skills, so the computer had better be able to do at least as well or it can’t even come to the table.

It’s true the computer doesn’t look at the full tree for more than a few moves, but it has no reason to do so since most of the possible moves are “obviously” inferior according to the evaluation function. So if you do a full tree search for say 2 full turns in front of an “obviously inferior” move and there are no surprises, you can throw away that tree. Of course you can concoct complicated exchange and discovery sequences that cause “surprises” far down the line, but I presume the computer looks for oddities like that while it is discarding bad subtrees.

Furthermore, the above 10^120 calculation is only meaningful if search is being used for the whole game, when in fact for let’s say 15+ moves of opening and probably 10+ moves of end-game (in the unlikely event the game gets into an end-game) it doesn’t have to be used as much or at all, so really only 20 or so moves of mid-game are where search is most needed – the board is still complex and the opening book is over and done with. However midgames tend to involve a lot of pieces stopped up and locked together, so the quantity of moves within them may not be that high unless the computer is really kicking ass already anyway.

During a typical mid-game, I doubt there are more than 10 “possibly good” moves per ply worth deeply considering, and no more than 20 “probably inferior” moves worth shallowly considering, so if the computer looks at say (just pulling these numbers out of the air) 20^4 positions shallowly and say 10^6 positions more deeply it really hasn’t done all that much work to look as much ahead as is necessary for 6 plies or so. Maybe that takes a few seconds out of the many minutes it can budget, and so after those million boards are sifted through, it has plenty of time for the “deep probes” into the tree that we are told form the bulk of the cleverness of the champion-caliber systems. Of course when an opponent plays into one of the computed lines that is expected of him, that makes things that much easier as several whole plies are already there, so the computer can just forge ahead down the plausible paths some more.

That is pretty bad. Let’s take guesses on what his excuse for that blunder’s going to be?

My guess:

“I really had to pee!”