We could probably pin point a specific date, the day and time the battle was lost :-)
11.5.1997
I bet Tal would be able to beat even a very strong chess comp ... well i have proof :) Game 1 of the book "Sacrifice and Initiative in chess" by New In Chess.. its a mind blowing game .... and after seeing that game , i have stopped believing in chess computers .... even the stockfish in lichess is not trustworthy ... :(
Prog's do not make mistakes. They win by wearing the opponent down. Eventually, with longer time controls and more moves, the human will make a few inaccuracies and lose. In short time controls, the human simply can not calculate as fast.
Prog's are not that far superior in evaluating best moves. They just do it faster. The evaluations are based on brute force. +/- Evals before 20 moves in book openings are to be taken with a grain of salt. Humans thankfully don't have to play prog's, who are cold blooded, ruthless .... (no swearing permitted).
Prog's are not that far superior in evaluating best moves. They just do it faster. The evaluations are based on brute force. +/- Evals before 20 moves in book openings are to be taken with a grain of salt. Humans thankfully don't have to play prog's, who are cold blooded, ruthless .... (no swearing permitted).
mdinnerspace said:
Prog's do not make mistakes.
This is simply wrong: computers do not *play* chess well, in fact they "do not understand" the game at all. What they are very good at is brute-forcce calculation. Programmers of chess programs are reluctant to put any more chess-wisdom into their programs because the better (read: more complicated) the evaluation function becomes the more the program is slowed down. Better to evaluate 100k positions with a bad evaluation function than to evaluate only 10k with a good one.
But there is another reason why computers win against humans: they cheat!
A human is not allowed to bring his library of opening notes to the table, but computers are allowed to rely on their opening database. Humans are not allowed to consult endgame libraries to look up if a certain position is a draw or a win - computers use tablebases all the time. Note that Nalimovs tablebases and opening books are not the computers work OTB - it is prefabricated data the computer simply uses.
A friend of mine who happens to be a GM added another point when we had exactly this discussion: during a game humans play with varying strength: they tend to get tired, lose concentration for a moment and so on. Computers always play with the same strength. Alas, chess is a game where one weak moment is enough, even if you dominated the whole game - unlike, i.e. tennis, where bad 5 minutes will make you lose an unnessecary game or two, then the match continues and if you play 5 sets these weak moments won't matter.
He said, that while he has no chance against a computer OTB he is convinced to be able to beat a chess program 8:2 in a time control like correspondence, where momentary tiredness or loss of concentration won't matter and where he can consult his library the same way the computer does.
krasnaya
Prog's do not make mistakes.
This is simply wrong: computers do not *play* chess well, in fact they "do not understand" the game at all. What they are very good at is brute-forcce calculation. Programmers of chess programs are reluctant to put any more chess-wisdom into their programs because the better (read: more complicated) the evaluation function becomes the more the program is slowed down. Better to evaluate 100k positions with a bad evaluation function than to evaluate only 10k with a good one.
But there is another reason why computers win against humans: they cheat!
A human is not allowed to bring his library of opening notes to the table, but computers are allowed to rely on their opening database. Humans are not allowed to consult endgame libraries to look up if a certain position is a draw or a win - computers use tablebases all the time. Note that Nalimovs tablebases and opening books are not the computers work OTB - it is prefabricated data the computer simply uses.
A friend of mine who happens to be a GM added another point when we had exactly this discussion: during a game humans play with varying strength: they tend to get tired, lose concentration for a moment and so on. Computers always play with the same strength. Alas, chess is a game where one weak moment is enough, even if you dominated the whole game - unlike, i.e. tennis, where bad 5 minutes will make you lose an unnessecary game or two, then the match continues and if you play 5 sets these weak moments won't matter.
He said, that while he has no chance against a computer OTB he is convinced to be able to beat a chess program 8:2 in a time control like correspondence, where momentary tiredness or loss of concentration won't matter and where he can consult his library the same way the computer does.
krasnaya
#15 Unless he uses engine assistance in the correspondence match there is no way he is winning a correspondence match against one of the strongest engines.
@krasnaya
I stand corrected. Certainly programs make "mistakes". My point was they don't blunder, never tire as you pointed out. Their mistakes are not serious enough to lose on the spot, where as a human's mistake, a single moment of a lapse in concentration, are often enough to result in a loss as you rightly describe.
The present computer championship The TCEC, does not permit opening books nor tablebases. The opening 2 moves are preselected, all in an attempt to nullify those advantages you point out. The 3000+ progs are making small "mistakes" right and left, resulting in a higher win/loss percentage then previous years. They can't follow opening theory for 20 moves where equality is often reached. They are left to their own "devices" at move 3.
I agree with your points. Prog's reach these 3400 ratings because of them. Not because they are so much better in chess skills, but because of the brute force it's processors have in short time controls. They play each other, inflation slowly ups their rating.
I'm with ya. I think a 2700 in a correspondence match, without assistance of another engine, stands a far better chance than most anticipate. Unfortunately, there is no upside for a human and a match is unlikely.
What takes the 2700 hours to find takes the engine minutes. The result of a draw would be a very reasonable result if the 2700 so wished to spend hours a day on a single move. But I'm sure they have far better ways to use their training, besides all the live games they are playing nowadays.
I stand corrected. Certainly programs make "mistakes". My point was they don't blunder, never tire as you pointed out. Their mistakes are not serious enough to lose on the spot, where as a human's mistake, a single moment of a lapse in concentration, are often enough to result in a loss as you rightly describe.
The present computer championship The TCEC, does not permit opening books nor tablebases. The opening 2 moves are preselected, all in an attempt to nullify those advantages you point out. The 3000+ progs are making small "mistakes" right and left, resulting in a higher win/loss percentage then previous years. They can't follow opening theory for 20 moves where equality is often reached. They are left to their own "devices" at move 3.
I agree with your points. Prog's reach these 3400 ratings because of them. Not because they are so much better in chess skills, but because of the brute force it's processors have in short time controls. They play each other, inflation slowly ups their rating.
I'm with ya. I think a 2700 in a correspondence match, without assistance of another engine, stands a far better chance than most anticipate. Unfortunately, there is no upside for a human and a match is unlikely.
What takes the 2700 hours to find takes the engine minutes. The result of a draw would be a very reasonable result if the 2700 so wished to spend hours a day on a single move. But I'm sure they have far better ways to use their training, besides all the live games they are playing nowadays.
When I was a baby GM's looked at computer chess, and laughed.
When I was a child GM's looked at computer chess, and watched...
When I became an adult GM's looked at computer chess, and learned...
Computers evolve faster than humans.
Back in the day a GM could be compared to a skilled Ninja with a Katana, and a computer to a stupid little monkey... Monkey tried to use brute force, Ninja used skill and foresight, and intelligence... Monkey stood no chance...
That monkey started small, and was weak. Then it grew and grew and was the size of a Gorilla... Still a Ninja could usually outsmart it, but if the Gorilla got in a lucky hit it could win easily with brute force...
Now though... It's like that Gorilla has become King Kong... GM's can still get lucky... In a few more years, maybe a decade or two and the time will come and it will be like King Kong, but with Ninja Skills , and a Katana the size of a bus and King Kong shoots lasers out of it's eyes... No chance.
In short computer strength grows exponentially where human strength does not.
When I was a child GM's looked at computer chess, and watched...
When I became an adult GM's looked at computer chess, and learned...
Computers evolve faster than humans.
Back in the day a GM could be compared to a skilled Ninja with a Katana, and a computer to a stupid little monkey... Monkey tried to use brute force, Ninja used skill and foresight, and intelligence... Monkey stood no chance...
That monkey started small, and was weak. Then it grew and grew and was the size of a Gorilla... Still a Ninja could usually outsmart it, but if the Gorilla got in a lucky hit it could win easily with brute force...
Now though... It's like that Gorilla has become King Kong... GM's can still get lucky... In a few more years, maybe a decade or two and the time will come and it will be like King Kong, but with Ninja Skills , and a Katana the size of a bus and King Kong shoots lasers out of it's eyes... No chance.
In short computer strength grows exponentially where human strength does not.
Just how is a computers strength growing exponentially, except of course in processing speed? (I have said this before, their rating is increasing because of inflation and they only play each other, much like the elite 2700 human players). They are not AI. Their programming only gets better as humans develop their skills. Faster means just that. Along with faster comes the ability to calculate more plies in 60 minute time controls. It has nothing to do with getting "smarter". Not until AI evolves and a comp can truly learn from it's mistakes and create new ideas.
The program AlphaGo has a big head start on the chess programmers.
The program AlphaGo has a big head start on the chess programmers.
Wow thanks guys this is some cool stuff from y'all. I agree that it will be interesting to see a correspondence match between an engine and a top grandmaster though. Limit usage of all the opening and endgame tablebases and perhaps GMs still stand a chance at drawing. But I still have a question, how do computers evaluate positions? How did humans teach them what's inferior or superior even when we can't tell for ourselves many times?
This topic has been archived and can no longer be replied to.