08 Jan 07
With Kramnik recently losing his match against Fritz and previously Michael Adams being crushed by Hydra have we seen the last competitive human vs engine chess match at classical time controls?
Other than a spectacular blunder by Kramnik it is pretty clear that he played well during each game and even then he barely had a sniff of a win.
I'm sure in correspondence chess human vs engine matches have a long way to run (so blunders like kramniks can be eliminated) but it seems that at classical time controls the engine has triumphed.
Comments?
08 Jan 07
1 edit
On another forum people were talking about a similar subject. They said humans can still utilize anti-chesscomputer tricks(certain fortress formations, tricking the computer with sacs that destroy it's positional and long term play, and other things that are a bit too complex to my beginner's mind).
I'm wondering though, if such tricks exist, why weren't kasparov and kramnik able to utilize them to defeat their ai opponents? Did they refuse to or are computers really that good that such tricks no longer work on them..?
edit: actually, Kasparov did try, but after game 2 of his match he suspected all of Deep BLue's moves weren;t being entirely by a computer.
08 Jan 07
Originally posted by Dyce WilloughbyI think Kramnik knew he was going to get beaten before they started. Computers are going to get better, a humans development is finite. We can learn from experience, which is our advantage at the moment, but it's only a matter of time before computers will do that too. Imagine a deep fritz strength engine that can learn from experience. Instead of getting old and 'rusty' like us mortals, it just gets better and better and better.... 🙄
With Kramnik recently losing his match against Fritz and previously Michael Adams being crushed by Hydra have we seen the last competitive human vs engine chess match at classical time controls?
Other than a spectacular blunder by Kramnik it is pretty clear that he played well during each game and even then he barely had a sniff of a win.
I'm s ...[text shortened]... eliminated) but it seems that at classical time controls the engine has triumphed.
Comments?
08 Jan 07
1 edit
Originally posted by MarinkatombChess computers don't need to learn. They aren't smart. They are brute calculating machines. The only intelligence comes in their ability to prune out bad lines instantly, so they don't waste time calculating them.
I think Kramnik knew he was going to get beaten before they started. Computers are going to get better, a humans development is finite. We can learn from experience, which is our advantage at the moment, but it's only a matter of time before computers will do that too. Imagine a deep fritz strength engine that can learn from experience. Instead of getting old and 'rusty' like us mortals, it just gets better and better and better.... 🙄
The thing we really need to worry about is chess being solved. Finding the perfect lines that cannot be refuted. And then ultimately the perfect game, any deviation from which leads to a loss. This may seem unlikely or impossible, but I don't think it is.
Edit- I guess there could be a learning engine, one that constantly updates its databases of games and positions so that it one day it will possess all known variations of a chess board and then cross-reference these during game play.
08 Jan 07
Originally posted by Dies IraeI played against an AI chess engine online the other day actually. It uses the Fritz engine to calculate, but learns from it's losses. Fritz is limited by it's horizon but this engine develops familiarity with the various different lines and will over rule the fritz first choice if it's previous experience has taught it a better continuation.
Chess computers don't need to learn. They aren't smart. They are brute calculating machines. The only intelligence comes in their ability to prune out bad lines instantly, so they don't waste time calculating them.
The thing we really need to worry about is chess being solved. Finding the perfect lines that cannot be refuted. And then ultimately ...[text shortened]... ssess all known variations of a chess board and then cross-reference these during game play.
I drew with it on moderate strength but it destroyed me on full strength. I have a link at home somewhere, i post it [/i]if i remember[/i] (...shows his human fallibility 😉)
08 Jan 07
Originally posted by MarinkatombThe thing is is that the computer isn't really learning, in the broadest sense of the term. The computer is just filling in the gaps in its calculations. The difference is is that the computer will never have an epiphany and realize some new strategic concept which will be applied to games. The computer just sees further into the outcome of a certain calculation and then develops around this.
I played against an AI chess engine online the other day actually. It uses the Fritz engine to calculate, but learns from it's losses. Fritz is limited by it's horizon but this engine develops familiarity with the various different lines and will over rule the fritz first choice if it's previous experience has taught it a better continuation.
I drew ...[text shortened]... a link at home somewhere, i post it [/i]if i remember[/i] (...shows his human fallibility 😉)
Its a subtle distinction to make, and I guess one could argue that when learning chess concepts, you are just learning calculated outcomes in a sense. But chess computers are mostly stupid and their ability to crank out lines is what makes them powerful adversaries.
08 Jan 07
Originally posted by Dies IraeBut why?
The thing we really need to worry about is chess being solved.
If a hypothetical machine told us that 1.e4 is a forced win for White, and 1.d4 is win for Black, will it really make much of a difference? If a GM plays 1.d4 against us, our chances of winning aren't going to change.
08 Jan 07
Originally posted by Dies IraeI cannot explain how it works i'm afraid. I read a bit of blurb but it was a while ago now and i can't remember the details. I know what your saying, computers can't conjure up deep strategic ideas like humans can, baced on subtle observations rather than brute calculation, but they can play deep strategic chess by accident!
The thing is is that the computer isn't really learning, in the broadest sense of the term. The computer is just filling in the gaps in its calculations. The difference is is that the computer will never have an epiphany and realize some new strategic concept which will be applied to games. The computer just sees further into the outcome of a certain ca ...[text shortened]... are mostly stupid and their ability to crank out lines is what makes them powerful adversaries.
If you look at Hydra's games with Adams, it displays an unbelievably positional style. Generally speaking computers don't 'think' like that. They apply the 'i go here, he goes there' approach to billions of different lines until they win material or mate. Hydra had such an immense amount of processing power behind it that it stumbled on some deep positional ideas that I personally haven't seen an engine do. I went to watch one of the games live and the commentators made a big deal out of this! How it comes to the conclusion is neither here nor there. Creating a machine that thinks like a human doesn't mean it will play the strongest chess possible. Perhaps the 'true' way to play this game is deeper than we realise...
08 Jan 07
Originally posted by VarenkaThere is no forced mate! That is a myth. If you look at ANY win, it is always the result of a mistake.
But why?
If a hypothetical machine told us that 1.e4 is a forced win for White, and 1.d4 is win for Black, will it really make much of a difference? If a GM plays 1.d4 against us, our chances of winning aren't going to change.
08 Jan 07
Originally posted by VarenkaThis is a vaild point. However, if computers solved chess, then chess theory would cease to evolve. And once that happens, you have a dead game on your hands that is good for shallow recreation but nothing else.
But why?
If a hypothetical machine told us that 1.e4 is a forced win for White, and 1.d4 is win for Black, will it really make much of a difference? If a GM plays 1.d4 against us, our chances of winning aren't going to change.
GM's would know the points to exploit for every opening, and there would be a very high degree of memorization. Innovation would cease. It would become a game of paint by numbers rather than a blank canvas for creativity. There would be no point in seriously pursuing chess because there is nothing to aspire to except in a scholarly sense. What I mean is that chess would fall to the wayside as an collection of databases, interesting only to people who like to grind through monotonous reference indexes.
Once something ceases to evolve, it dies, and becomes interest only to scholars, historians, and curators. Not to people who possess dynamic, innovative minds. Chess being solved would cause the stagnation, and then ultimately, the death of the game.
08 Jan 07
Originally posted by MarinkatombWell, the thing is is that positional ideas accomplish one thing- they maximize the possibilites for a win. They are broad concepts, from which, tactical lines are played out.
I cannot explain how it works i'm afraid. I read a bit of blurb but it was a while ago now and i can't remember the details. I know what your saying, computers can't conjure up deep strategic ideas like humans can, baced on subtle observations rather than brute calculation, but they can play deep strategic chess by accident!
If you look at Hydra's ga ...[text shortened]... ess possible. Perhaps the 'true' way to play this game is deeper than we realise...
But computers go about it from the other direction. They don't need large concepts if they can calculate everything out. They just use the outcomes to determine the moves. Computers know precisely what will happen. Humans are unable to do this. This is why we need strategic concepts.
Its kind of an inductive/deductive split. Humans deduce they will win because they have the better position. Computers induce the positional concepts because they know they will win. Generalizations mean nothing to a computer if it sees everything.