![[us]](/mwf/v12/flags/us.png "United States")
I am writing this in response to a topic posted at Talkchess concerning a new era for computer chess.Maybe we are approaching a different era for computer chess.The TOP chess engines are indeed stronger than humans and can even defeat Anand maybe with ease.The question should now be how can computer chess enhance mankind in any given field?Can we apply the computational aspect of computer chess to real life to advance man in science and medicine?When chess is solved by computers maybe by 2050AD we must ask ourselves how can we apply this knowledge to help mankind.Maybe by then an algorithm used to solve chess can be applied to the sciences or medicine--who nows?Yes we must ask ourselves what is the point now as the computer chess playing entities (thinking of Rybka in particular) are basically unbeatable by humans we must take the back seat and learn from what we have created,although there is still much for them to learn.
Chess is my life but my life is not only chess
FWCC
Chess is my life but my life is not only chess
FWCC
![[mx]](/mwf/v12/flags/mx.png "Mexico")
For computers being able to help in science and medicine, you need Artificial Intelligence. The chess engines are just looking at tables and calculating stuff, like actual calculators. (Like a machine that plays perfect Tic Tac Toe can't help in anything, with chess it's the same, it's just that chess is a lot more complex).
Maybe when we have Artificial Intelligence it will help computer chess, but I really doubt that it will happen the other way around.
Maybe when we have Artificial Intelligence it will help computer chess, but I really doubt that it will happen the other way around.
Your sentiments are either very naive or very forward-thinking, I'm not sure which. A case could be made either way.
Chess falls into the realm of mathematics--many different aspects of mathematics. How can chess help the world? Better question: how have and how will mathematicians help the world?
Solving chess won't do anything for mankind. However the kind of mathematical/computational power and insight required to solve the game, assuming it can be done (many would argue not in this century), is amoral. It could be used for good or ill. And from that point of departure we enter the realm of philosophy...
Chess falls into the realm of mathematics--many different aspects of mathematics. How can chess help the world? Better question: how have and how will mathematicians help the world?
Solving chess won't do anything for mankind. However the kind of mathematical/computational power and insight required to solve the game, assuming it can be done (many would argue not in this century), is amoral. It could be used for good or ill. And from that point of departure we enter the realm of philosophy...
![[de]](/mwf/v12/flags/de.png "Germany")
Better question: how have and how will mathematicians help the world?
This question is answered by history! No mathematics, ergo no science. Ok, there will be some mystics claiming that science don´t help the world (humans).
Last: Chess has nothing to do with mathematics. There is no proof about correlation of chess skill and math skill. The best German chess player in the past was Robert Hübner, a classicist.
Very last: A very strange thread.
This question is answered by history! No mathematics, ergo no science. Ok, there will be some mystics claiming that science don´t help the world (humans).
Last: Chess has nothing to do with mathematics. There is no proof about correlation of chess skill and math skill. The best German chess player in the past was Robert Hübner, a classicist.
Very last: A very strange thread.
I was thinking in terms of mathematics,the forthcoming computational ideas may help in the field of theoretical physics or quantum mechanics which deals with the uncertainty principle much like the complexity of chess.The ideas that might be derived as we approach the solving of chess might be useful in the subject of stellar phenomenon or the like.
I agree. The mathematical aspect of chess is not important, or so simple that it is not worth mentioning.
Something which can be generalised perhaps is though, how intuitive or non-exact human experience-based knowledge can be converted, to be used by computer algorithms successfully. All the positional evaluation factors chess programs use, are based on human judgements, which are based on human (master's) experience and the explanations/advice they have given. There is no mathematical formula which leads to the result that a knight outpost is good. - But as we have learned in computer chess, those factors, their number and their weights each, need to be different in computer chess than they are in human chess, due to the much different nature of a chess engine's method to process the information. Humans operate much more with patterns (and with more patterns if they are strong players). For comps - at least for most problems in chess and within a certain horizon - the "fast number crunching" approach, although refined by very intelligently improved search techniques, is far superior. Nevertheless, search without evaluation is nothing.
An example of how strong a chess program with a very minimalistic evaulation function can play if it achieves 10+ plies of search depth, is the engine microMax 4.8 by H.G.Muller. It's source code is smaller than 2,000 non-blank characters, in other worde less than 2 KB.
http://home.hccnet.nl/h.g.muller/progress.html
http://home.hccnet.nl/h.g.muller/dwnldpage.html
By looking at the code as someone who is no programmer, I am surprised that it can contain all the chess rules even, let alone all the other parts in a chess engine :-D I guess even writing down the chess rules in plain language needs more characters. But it plays good! I would estimate, 2000+ Elo. In the CCRL tests, it is last but it did score 56 wins (and 59 draws) in 253 against an oppo average of ~2140. That illustrates the entirely different balance between knowledge and speed. A human player of the same strength surely knows MUCH more, but also he is of course much slower.
Something which can be generalised perhaps is though, how intuitive or non-exact human experience-based knowledge can be converted, to be used by computer algorithms successfully. All the positional evaluation factors chess programs use, are based on human judgements, which are based on human (master's) experience and the explanations/advice they have given. There is no mathematical formula which leads to the result that a knight outpost is good. - But as we have learned in computer chess, those factors, their number and their weights each, need to be different in computer chess than they are in human chess, due to the much different nature of a chess engine's method to process the information. Humans operate much more with patterns (and with more patterns if they are strong players). For comps - at least for most problems in chess and within a certain horizon - the "fast number crunching" approach, although refined by very intelligently improved search techniques, is far superior. Nevertheless, search without evaluation is nothing.
An example of how strong a chess program with a very minimalistic evaulation function can play if it achieves 10+ plies of search depth, is the engine microMax 4.8 by H.G.Muller. It's source code is smaller than 2,000 non-blank characters, in other worde less than 2 KB.
http://home.hccnet.nl/h.g.muller/progress.html
http://home.hccnet.nl/h.g.muller/dwnldpage.html
By looking at the code as someone who is no programmer, I am surprised that it can contain all the chess rules even, let alone all the other parts in a chess engine :-D I guess even writing down the chess rules in plain language needs more characters. But it plays good! I would estimate, 2000+ Elo. In the CCRL tests, it is last but it did score 56 wins (and 59 draws) in 253 against an oppo average of ~2140. That illustrates the entirely different balance between knowledge and speed. A human player of the same strength surely knows MUCH more, but also he is of course much slower.
Many thanks for this post. Let us have a look to computer chess and here to the borders :-). Rybka is top! microMax is lowest but great! We can see clearly the winning margin of computer chess. The human player begins with less than 1400 Elo and the "crude" chess engine begins with more than 2000 Elo. Just because of searching and hardware today. And what we see at top level? No +600 Elo :-). And in what we trust in computer chess? Better searching and better hardware (2, 4, 8, 16, 32 ... 128 cores with 2, 3, 3.5, 4 ... 8 GHz) and better OS (32-, 64- or 128 bit?) :-)!?? Isn´t it silly? In ten (or five?) years microMax 4.8 mp is better than the best human chess player and the source code is on 2 or 3 papers.
PS. H.G.Muller is ingenious.
PS. H.G.Muller is ingenious.
> Rybka is top! microMax is lowest but great!
No, microMax is the smallest engine, not the weakest.
There are other programs that are huge in comparison, but weaker, so microMax isn't the lowest.
I think that the mathematical aspect is more pervasive in chess than one might give credit, especially in such realms as graph theory. However, it is difficult to say whether that challenges your statement that they're "unimportant"--it may be that even though the mathematical theory in such realms exists, it may not be as useful as the knowledge we already have about the game.
![[hu]](/mwf/v12/flags/hu.png "Hungary")
There isn't really a very big difference between human knowledge and computer knowledge. Most logical evaluation terms do work for Rybka. It was quite interesting for me when Larry first joined our team - he of course immediately proposed a bunch of eval terms, many of which didn't make a lot of sense to me, but they did work and now they make sense :)
Maybe there are some unnatural 'computer-only' terms which would work as well, but I don't really see how to go about inventing them.
Anyway, the difference between humans and computers is that a computer on modern hardware searches a lot faster, and therefore plays a lot better than a human with the same knowledge.
Vas
Maybe there are some unnatural 'computer-only' terms which would work as well, but I don't really see how to go about inventing them.
Anyway, the difference between humans and computers is that a computer on modern hardware searches a lot faster, and therefore plays a lot better than a human with the same knowledge.
Vas
Is it no such a difference ?
a) computer: generates possible moves(2-40), calculates them and evaluates the reached position, selects the move
b) human: analyses the position, selects candidate moves on the base of the position(2-4) and calculates them, selects the move
Regards
Hetman
a) computer: generates possible moves(2-40), calculates them and evaluates the reached position, selects the move
b) human: analyses the position, selects candidate moves on the base of the position(2-4) and calculates them, selects the move
Regards
Hetman
I was talking about the eval.
Still, if a human was going to look at several million positions, his search tree might not be too different from that of an engine.
Vas
Still, if a human was going to look at several million positions, his search tree might not be too different from that of an engine.
Vas
It is obvious that human will not go to search so many positions :-).
Is it possible to separate eval from searching ?
The search tree is adjusted by eval, is not it ? The 'bad' variations are eliminated.
I think that human is going from differrent side to the chess. Decision from the position (soul or request) to the move, computer from evaluating the variations to the choosed move.
Rgds
Hetman
Is it possible to separate eval from searching ?
The search tree is adjusted by eval, is not it ? The 'bad' variations are eliminated.
I think that human is going from differrent side to the chess. Decision from the position (soul or request) to the move, computer from evaluating the variations to the choosed move.
Rgds
Hetman
Analysis on neural activity in people has shown that positional evaluation and calculation (search) are separate activities that are actually done in different parts of the brain. In people, positional evaluation is a pattern recognition skill that allows someone like Kasperov to look at a new position and almost instantaneously decide the game is won or lost before even doing the calculations to find the relevant variations.
Regards,
Alan
Regards,
Alan
This actually isn't too different from what an engine does. The main differences are:
1) Kasparov's evaluation is (somehow) done in parallel, while Rybka's is sequential (term A, then term B, etc)
2) Kasparov has (much) better heuristics
3) Kasparov is (much much much) slower
Vas
1) Kasparov's evaluation is (somehow) done in parallel, while Rybka's is sequential (term A, then term B, etc)
2) Kasparov has (much) better heuristics
3) Kasparov is (much much much) slower
Vas
You would be in deep trouble if you had to rely on the use of heuristics to make sense of incoming sensor data. As an example, when you walk into a room where you've never been before, you can immediately identify the sofa, the television, a potted plant, the dog, and millions of other objects without performing any conscious thought. You can do this, even if each of the objects is different than all of the objects of that type that you've seen before. This is all almost certainly done in the brain by pattern matching against numerous templates of each of the objects. Picasso made a living by painting the template matching characteristics of objects while distorting or eliminating the objects themselves.
I suspect that Kasparov's strength relative to other chess players is based more on his pattern recognition skills than on superior heuristics or the ability to calculate faster than his peers. If this is correct, it would be significantly different than the engine model, where the search is done many orders of magnitude faster, but the evaluation is based on simple heuristics (maybe less simple for Rybka than for other engines), and pattern recognition is not used at all.
Regards,
Alan
I suspect that Kasparov's strength relative to other chess players is based more on his pattern recognition skills than on superior heuristics or the ability to calculate faster than his peers. If this is correct, it would be significantly different than the engine model, where the search is done many orders of magnitude faster, but the evaluation is based on simple heuristics (maybe less simple for Rybka than for other engines), and pattern recognition is not used at all.
Regards,
Alan
This same view on importance of pattern recognicion in (human) chess is supported by the television program "My Brilliant Brain (2)" which is about Susan Polgar. In the documentation, they talk almost entirely about patterns, while the role of calculation is almost neglected. In my opinion, chess calculations (vision) as being fundamental for concrete tactical variations, should have been mentioned more, or at all, as I am not even sure if it appears at least once. Except for that, it is a great documentation which also contains some very nice video effects to symbolise chess thinking. Also, you can see how she plays a 1+0 bullet game with real board and clock(!), in a park.
http://susanpolgar.blogspot.com/2007/11/national-geographic-chess-documentary.html
http://susanpolgar.blogspot.com/2007/11/national-geographic-chess-documentary.html
Thanks Mike. I heard about this documentary, but haven't seen it. I'll definitely watch it when I have the requisite 47 minutes.
Regards,
Alan
Regards,
Alan
Ok, I use the word 'heuristic' in a more general way, as anything which gives an imperfect answer.
http://en.wikipedia.org/wiki/Heuristic
Of course, in a computer program, heuristics are formally defined (in a computer program, everything is formally defined). In a human brain, heuristics may or may not be formally defined.
In both cases, they work quite similarly - you recognize a pattern and evaluate it. The patterns themselves may not be that different. Sure, Kasparov's patterns are going to be more complex and powerful, but the patterns of a 2000 player should be fairly similar to Rybka's.
Vas
http://en.wikipedia.org/wiki/Heuristic
Of course, in a computer program, heuristics are formally defined (in a computer program, everything is formally defined). In a human brain, heuristics may or may not be formally defined.
In both cases, they work quite similarly - you recognize a pattern and evaluate it. The patterns themselves may not be that different. Sure, Kasparov's patterns are going to be more complex and powerful, but the patterns of a 2000 player should be fairly similar to Rybka's.
Vas
It is working so but not always.
Sometimes the estimation of the position depends on that who is moving and in that case pattern recognition has lesser meaning.
Rgds
Hetman
Sometimes the estimation of the position depends on that who is moving and in that case pattern recognition has lesser meaning.
Rgds
Hetman
It took me a minute to understand your comment. You are correct that side to move will, of course, affect the evaluation of the position and this is not obvious when just looking at the chess board, so it would require conscious thought to add in this variable (or one could possibly have two sets of pattern recognition going to give both answers and just pick the one that is appropriate).
There are a few other conditions that are not always obvious from looking at the current state:
1) Castling rights, and
2) En Passant
Regards,
Alan
There are a few other conditions that are not always obvious from looking at the current state:
1) Castling rights, and
2) En Passant
Regards,
Alan
Naturally, all three of these things are important in figuring out who has the advantage. However, I believe that someone of Garry Kasparov's caliber can, from the look of the position and thinking in terms of similar types of positions that he's seen before, know all three of these the majority of the time.
There is no need for pattern recognition to be strictly visual.
Vas
Vas
No, but the brain's pattern recognition mechanisms for sensor information have been highly developed during the normal process of evolution, and since these processes work very quickly (on a human scale anyway) and have many orders of magnitude more bandwidth than a calculation process, it makes sense to use them to the greatest possible extent. Even an intermediate player can use these pattern recognition skills to quickly give a judgment on the merits of the position. Better players presumably would give both a more accurate, and a more precise evaluation.
Going back to the example, anyone, with very little training, could almost instantly perceive a complex event (e.g. Dagh barfing on the analysis board), while detecting this same event in a process defined by a flow chart would be almost impossible.
Alan
Going back to the example, anyone, with very little training, could almost instantly perceive a complex event (e.g. Dagh barfing on the analysis board), while detecting this same event in a process defined by a flow chart would be almost impossible.
Alan
Alan (white) plays 1.b3-->Dagh (black) goes to openings book after playing his favorite response-->Dagh then sees nothing in the openings book-->Dagh realizes that the last time he analyzed the line, he accidentally clicked "Remove Tree" instead of "Allow Move Adding"-->Dagh becomes very sick in his stomach-->Dagh barfs on analysis board
Ok, this is true.
In fact, systems which help you memorize things typically exploit this. You set things up in your mind in a graphical way.
Vas
In fact, systems which help you memorize things typically exploit this. You set things up in your mind in a graphical way.
Vas
Thank You for the understanding. I have meant not only that.
There are 'wild tactical positions' :-) where everything depends on the calculation and here the pattern recognition does not solve the problem. E. Lasker would classified that as 'dynamically balanced position'. I have been reading his 'Chess manual' n-times and I am delighted with his way of understanding the chess.
Rgds
Hetman
There are 'wild tactical positions' :-) where everything depends on the calculation and here the pattern recognition does not solve the problem. E. Lasker would classified that as 'dynamically balanced position'. I have been reading his 'Chess manual' n-times and I am delighted with his way of understanding the chess.
Rgds
Hetman
I certainly agree with you that noone has developed their pattern recognition skills to the point where they no longer need to do calculation to solve many chess problems.
On the other hand, consider a F1 race. The drivers must react to many events too quickly to allow time for calculation (e.g. a crash has occurred in front of them, what action should be taken?). This is an example of pattern recognition being used to attempt to solve a dynamic problem.
Regards,
Alan
On the other hand, consider a F1 race. The drivers must react to many events too quickly to allow time for calculation (e.g. a crash has occurred in front of them, what action should be taken?). This is an example of pattern recognition being used to attempt to solve a dynamic problem.
Regards,
Alan
I agree in that case. Many matters are decided 'automaticly' in the chess, either. It is done on the base of the training and experience.
I think that it is very important in otb play. It is not enough time to calculate and think through everything what is happening on the board.
Rgds
Hetman
I think that it is very important in otb play. It is not enough time to calculate and think through everything what is happening on the board.
Rgds
Hetman
![[lu]](/mwf/v12/flags/lu.png "Luxembourg")
Hi,
I think there would be substantial differences because in the process of searching these millions of positions, the human would obviously be doing concept-level learning about the position. I think that this is probably also among the main reasons why people appear to gain more in playing strength from a doubling of thinking time than computers (the other main reason being that with more time, the human can eliminate many short-term tactical blunders and hence alleviate his main weakness).
Regards,
Neuromancer
> Still, if a human was going to look at several million positions, his search tree might not be too different from that of an engine.
I think there would be substantial differences because in the process of searching these millions of positions, the human would obviously be doing concept-level learning about the position. I think that this is probably also among the main reasons why people appear to gain more in playing strength from a doubling of thinking time than computers (the other main reason being that with more time, the human can eliminate many short-term tactical blunders and hence alleviate his main weakness).
Regards,
Neuromancer
Rybka does this type of learning. When she searches branch A, she learns things which are used in branch B, and I'm not just talking about transpositions.
Vas
Vas
E. Lasker was mathematician. S. Tarrasch medicine doctor. A. Alechin lawyer. Botwinnik electrical engineer.
M. Euwe mathematician and computer engineer.
Chess are logical-strategical game. The logical mind shall help in the chess. Logic is the part of math.
Rgds
Hetman
M. Euwe mathematician and computer engineer.
Chess are logical-strategical game. The logical mind shall help in the chess. Logic is the part of math.
Rgds
Hetman
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill