#1161
Yes, originally I thought:
10 men table base
quantum computer
this century
Sveshnikov:
7 men table base
conventional computer
5 years
True or false? Chess will never be solved! why?
Sort:
Sep 19, 2021
0
#1163
Ubik42 wrote:
Capablanca solved chess when he was on his undefeated run, and wrote the solution in the margins of his manuscript for “Chess Fundamentals”.
The editor removed it because he “didn’t think it was of general interest to the beginning chess player”.
The editor removed it because he “didn’t think it was of general interest to the beginning chess player”.
This claim was made by a woman with whom Capablanca was cheating during his second marriage. As Chess Fundamentals was completed before he met his first wife, it should not be trusted. The woman obviously never saw the original manuscript, which Capablanca left with the publisher in summer 1920.
His first wife was one of the "fans" who traveled to Habana to observe the WCC in March and April 1921.
tygxc wrote:
#1161
Yes, originally I thought:
10 men table base
quantum computer
this century
Sveshnikov:
7 men table base
conventional computer
5 years
Sveshnikov is like a grand prix racer that tries to tell his mechanics how to design his engine. His expertise is not directly applicable, only indirectly. Great game players != great game designers != expert software developers.
There is overlap in that Venn diagram, but I see no evidence that Sveshnikov had any particular expertise in chess engines, databases, or computers in general. Since he became a GM in 1977, it's highly unlikely. Especially since he argued that chess games and positions reached by two human players should be copyright protected, and not be allowed to be stored in databases at all 
. One thing is certain, he liked to toss his opinions around. His claim would only be marginally more credible than yours, regardless.
tygxc wrote:
#1157
Neither you nor any other has refuted anything, you have only ridiculed and made nonsense comparisons to pies and cherries and whatnot. Laughing is no argument, it is lack of argument. You have never posted any credible numbers. Schaeffer has solved checkers. Chess is more complex: more squares, more men, more complex rules, but the same method is applicable.
Checkers has 10^20 possible positions, and this was reduced to 10^14 to solve checkers. That's 6 orders of magnitude.
Chess has at least 10^40+ positions (we'll discard your somewhat arbitrary decision to cull it to 10^37 and 26 piece positions only), and that number is already culled quite a bit (10^40 already represents 6 orders of magnitude of culling, in fact, much like checkers). But that's not the whole story...checkers is a much simpler game with half the board space and extremely limited move options in a given position, so the calculations required are much easier.
You keep shifting your position. First, you claimed that chess positions would be culled to 10^20, and that a 10 man tablebase would suffice. Now you say it can be done with 10^37 and only an 8 man tablebase.
There will be ~38,176,306,877,748,245 positions in the finished 8 man tablebase. That's 10^16.58. You are still 20 orders of magnitude away even by your vastly overreaching numbers. It's more like 26 orders of magnitude, a quadrillion times larger than the solution for checkers required. So, by your own petard, lets hoist you...if the "same method is applicable" (though it's not actually the same, as I pointed out above), then it would take how long to solve chess compared to checkers?
Those are not the most accurate numbers, but even by your own methodology, you are way off the mark. I have played chess since I was 6, started programming at 12 (and retired after a career in software development management and systems/databse design), and have played chess against computers since the very first horrifically bad programs came out...there's nobody who would like to see computers solve chess more than I...but it's nowhere near happening. I was where you are now 8 years ago...trying to make it work. If a breakthrough occurs that changes the equation, then it can be revisited, but for now, solving chess is soft science fiction/fantasy, not hard science fiction/imminent.
#1164
"Great game players != great game designers != expert software developers."
Sveshnikov was a player and a theoretician. To solve chess requires no software development, but application of software.
"I see no evidence that Sveshnikov had any particular expertise in chess engines, databases, or computers in general." Sveshnikov won the world senior championship 65+ in 2017. If he had not already been GM in 1977, then this win would have earned him his GM title in 2017. In his theoretical work on his Sveshnikov Variation, the Alapin Variation, and the French Advance Variation he surely used engine assistance.
"His claim would only be marginally more credible than yours, regardless."
His claim > my claim > your claim
tygxc wrote:
#1164
"Great game players != great game designers != expert software developers."
Sveshnikov was a player and a theoretician. To solve chess requires no software development, but application of software.
"I see no evidence that Sveshnikov had any particular expertise in chess engines, databases, or computers in general." Sveshnikov won the world senior championship 65+ in 2017. If he had not already been GM in 1977, then this win would have earned him his GM title in 2017. In his theoretical work on his Sveshnikov Variation, the Alapin Variation, and the French Advance Variation he surely used engine assistance.
"His claim would only be marginally more credible than yours, regardless."
His claim > my claim > your claim
Nothing you just said has any real bearing on this topic. Player and theoretician, senior GM title, a few opening variations...all largely meaningless to the issue at hand.
Some people should know
Sep 19, 2021
0
#1169
Casual_Joe wrote:
Moses2792796 wrote:
Casual_Joe wrote:
They said that a computer would never be small enough to fit into a single room. Now we have 100000 times the computing power on our smart phone as they had in the original supercomputers. Chess will certainly be solved, it's just a question of when. And the argument about the number of atoms in the universe doesn't hold -- quantum computers are based on subatomic particles, and there are plenty of those to go around!
This is what I thought, although it still seems somewhat implausible that we could develop technology that could store that amount of data.
Electricity, internal combustion engines, and modern medicine all seemed implausible to the ancient romans as well! I'm not saying it'll be solved soon, but eventually, unless the Second Coming occurs first!
True chess is a tactical game and you can’t predict the future of chess but the way humans are evolving people will be able to predict chess
#1165
"Checkers has 10^20 possible positions, and this was reduced to 10^14 to solve checkers. That's 6 orders of magnitude."
++ That is 6 orders of magnitude, but you cannot interpret it like that. If somebody were to apply the same method as used for 8*8 checkers on 10*10 draughts then it will be more than 6 orders of magnitude. For chess it will be even more orders of magnitude.
"we'll discard your somewhat arbitrary decision to cull it to 10^37 and 26 piece positions only" ++ That is not arbitrary, it is the result of John Tromp research that 28 men hold most chess positions, hence it is an advantage to start at 26. It is also what Sveshnikov implicitly says when he asks for good assistants and starting from the opening: the human assistants should guide the search starting at 26 men, then the computer e.g. Sesse should calculate until it hits the table base.
"that number is already culled quite a bit (10^40 already represents 6 orders of magnitude of culling, in fact, much like checkers)."
++ No, not at all. The 10^46 is an obsolete and incorrect number. I quoted the exact numbers from the John Tromp research. The number of possible positions contains illegal positions, and non sensible positions like 5 light square bishops.
The 10^6 culling for checkers is about relevant positions: positions which are legal and sensible, but which need not be assessed for the proof. E.g. for chess after 1 e4 all positions with a white pawn at e2 are no longer relevant. Or after 6 Bxc6 dxc6 all positions with a white light square bishop or a black pawn at d7 are no longer relevant. So the reduction to about the square root of the number of possible, legal, and sensible positions stems from the logic of the proof.
Example: the Four Color Theorem was proved by letting a computer color 1482 relevant maps, not by coloring all possible maps.
"checkers is a much simpler game with half the board space and extremely limited move options in a given position, so the calculations required are much easier."
++ On that all agree.
Checkers: 32 squares, 24 men, 2 types of pieces, compulsory capture.
Chess: 64 squares, 32 men, 6 types of pieces, optional capture.
"First, you claimed that chess positions would be culled to 10^20, and that a 10 man tablebase would suffice. Now you say it can be done with 10^37 and only an 8 man tablebase."
++ The opinion of Sveshnikov made me rethink. I was too pessimistic.
"There will be ~38,176,306,877,748,245 positions in the finished 8 man tablebase."
++ That is irrelevant: Sesse or other takes care to calculate from the humanly prepared 26 men opening tabiya to the table base. A table base with more men shortens the calculation time but the method stays the same.
"then it would take how long to solve chess compared to checkers?"
++ 5 years per Sveshnikov, 5.5 years for 1 ECO code per my calculation above
"there's nobody who would like to see computers solve chess more than I...but it's nowhere near happening."
++ Interesting:
you want it to happen but you think it will not happen.
I do not want it to happen, but I think it will happen.
"If a breakthrough occurs that changes the equation, then it can be revisited"
++ The John Tromp research was a new input.
The Sveshnikov opinion was a new input.
The arrival of 8 men table base will be a new input.
New versions of Sesse or other with improved performance will be new inputs.
New applications of quantum computers will be new inputs.
tygxc wrote:
#1165
"Checkers has 10^20 possible positions, and this was reduced to 10^14 to solve checkers. That's 6 orders of magnitude."
++ That is 6 orders of magnitude, but you cannot interpret it like that. If somebody were to apply the same method as used for 8*8 checkers on 10*10 draughts then it will be more than 6 orders of magnitude. For chess it will be even more orders of magnitude.
"we'll discard your somewhat arbitrary decision to cull it to 10^37 and 26 piece positions only" ++ That is not arbitrary, it is the result of John Tromp research that 28 men hold most chess positions, hence it is an advantage to start at 26. It is also what Sveshnikov implicitly says when he asks for good assistants and starting from the opening: the human assistants should guide the search starting at 26 men, then the computer e.g. Sesse should calculate until it hits the table base.
"that number is already culled quite a bit (10^40 already represents 6 orders of magnitude of culling, in fact, much like checkers)."
++ No, not at all. The 10^46 is an obsolete and incorrect number. I quoted the exact numbers from the John Tromp research. The number of possible positions contains illegal positions, and non sensible positions like 5 light square bishops.
The 10^6 culling for checkers is about relevant positions: positions which are legal and sensible, but which need not be assessed for the proof. E.g. for chess after 1 e4 all positions with a white pawn at e2 are no longer relevant. Or after 6 Bxc6 dxc6 all positions with a white light square bishop or a black pawn at d7 are no longer relevant. So the reduction to about the square root of the number of possible, legal, and sensible positions stems from the logic of the proof.
Example: the Four Color Theorem was proved by letting a computer color 1482 relevant maps, not by coloring all possible maps.
"checkers is a much simpler game with half the board space and extremely limited move options in a given position, so the calculations required are much easier."
++ On that all agree.
Checkers: 32 squares, 24 men, 2 types of pieces, compulsory capture.
Chess: 64 squares, 32 men, 6 types of pieces, optional capture.
"First, you claimed that chess positions would be culled to 10^20, and that a 10 man tablebase would suffice. Now you say it can be done with 10^37 and only an 8 man tablebase."
++ The opinion of Sveshnikov made me rethink. I was too pessimistic.
"There will be ~38,176,306,877,748,245 positions in the finished 8 man tablebase."
++ That is irrelevant: Sesse or other takes care to calculate from the humanly prepared 26 men opening tabiya to the table base. A table base with more men shortens the calculation time but the method stays the same.
"then it would take how long to solve chess compared to checkers?"
++ 5 years per Sveshnikov, 5.5 years for 1 ECO code per my calculation above
"there's nobody who would like to see computers solve chess more than I...but it's nowhere near happening."
++ Interesting:
you want it to happen but you think it will not happen.
I do not want it to happen, but I think it will happen.
"If a breakthrough occurs that changes the equation, then it can be revisited"
++ The John Tromp research was a new input.
The Sveshnikov opinion was a new input.
The arrival of 8 men table base will be a new input.
New versions of Sesse or other with improved performance will be new inputs.
New applications of quantum computers will be new inputs.
You realize that Sesse is Stockfish running on a supercomputer, right? Sesse's evaluations/nodes per second are not a calculation that can be used to solve chess. Those valuations are flawed (as are all engine valuations as engines continue to improve), and are not at all like tablebase calculations...so using Sesse nodes per second in your premise is just wrong from the start.
Quantum computers are not going to be applicable to the problem, as you have been informed at least a dozen times by many posters. This is another case of you applying resources and numbers where they don't fit or work.
Sveshnikov opinion is not a new input. It was a layman's opinion.
John Tromp's numbers are not your numbers. He's still at 10^44 or so. You knocked off another 7 orders of magnitude on your own with shoddy logic.
tygxc wrote:
#1165
"Checkers has 10^20 possible positions, and this was reduced to 10^14 to solve checkers. That's 6 orders of magnitude."
++ That is 6 orders of magnitude, but you cannot interpret it like that. If somebody were to apply the same method as used for 8*8 checkers on 10*10 draughts then it will be more than 6 orders of magnitude. For chess it will be even more orders of magnitude.
"we'll discard your somewhat arbitrary decision to cull it to 10^37 and 26 piece positions only" ++ That is not arbitrary, it is the result of John Tromp research that 28 men hold most chess positions, hence it is an advantage to start at 26. It is also what Sveshnikov implicitly says when he asks for good assistants and starting from the opening: the human assistants should guide the search starting at 26 men, then the computer e.g. Sesse should calculate until it hits the table base.
"that number is already culled quite a bit (10^40 already represents 6 orders of magnitude of culling, in fact, much like checkers)."
++ No, not at all. The 10^46 is an obsolete and incorrect number. I quoted the exact numbers from the John Tromp research. The number of possible positions contains illegal positions, and non sensible positions like 5 light square bishops.
The 10^6 culling for checkers is about relevant positions: positions which are legal and sensible, but which need not be assessed for the proof. E.g. for chess after 1 e4 all positions with a white pawn at e2 are no longer relevant. Or after 6 Bxc6 dxc6 all positions with a white light square bishop or a black pawn at d7 are no longer relevant. So the reduction to about the square root of the number of possible, legal, and sensible positions stems from the logic of the proof.
Example: the Four Color Theorem was proved by letting a computer color 1482 relevant maps, not by coloring all possible maps.
"checkers is a much simpler game with half the board space and extremely limited move options in a given position, so the calculations required are much easier."
++ On that all agree.
Checkers: 32 squares, 24 men, 2 types of pieces, compulsory capture.
Chess: 64 squares, 32 men, 6 types of pieces, optional capture.
"First, you claimed that chess positions would be culled to 10^20, and that a 10 man tablebase would suffice. Now you say it can be done with 10^37 and only an 8 man tablebase."
++ The opinion of Sveshnikov made me rethink. I was too pessimistic.
"There will be ~38,176,306,877,748,245 positions in the finished 8 man tablebase."
++ That is irrelevant: Sesse or other takes care to calculate from the humanly prepared 26 men opening tabiya to the table base. A table base with more men shortens the calculation time but the method stays the same.
"then it would take how long to solve chess compared to checkers?"
++ 5 years per Sveshnikov, 5.5 years for 1 ECO code per my calculation above
"there's nobody who would like to see computers solve chess more than I...but it's nowhere near happening."
++ Interesting:
you want it to happen but you think it will not happen.
I do not want it to happen, but I think it will happen.
"If a breakthrough occurs that changes the equation, then it can be revisited"
++ The John Tromp research was a new input.
The Sveshnikov opinion was a new input.
The arrival of 8 men table base will be a new input.
New versions of Sesse or other with improved performance will be new inputs.
New applications of quantum computers will be new inputs.
I'm trying very hardly to find a single line in all that which makes some sense, but I'm afraid I will fail.
#1172
"You realize that Sesse is Stockfish running on a supercomputer, right?" ++ Right!
"Sesse's evaluations/nodes per second are not a calculation that can be used to solve chess."
++ Yes, they can: they guide the calculation towards the table base. Consider the top 4 Sesse white moves, consider the top 1 Sesse black move, continue until you hit the table base. The table base gives the final evaluation: draw or not.
"Quantum computers are not going to be applicable to the problem"
++ We will see about that. All statements about the future always are opinion.
"as you have been informed at least a dozen times by many posters" ++ Opinions differ.
"Sveshnikov opinion is not a new input. It was a layman's opinion."
++ No, it is expert opinion by a 2500 grandmaster still active in 2017 and leading theoretician. Compared to him we are all laymen on chess analysis. Experts on engine assisted analysis are: top grandmasters, their seconds, theoreticians, ICCF grandmasters.
"John Tromp's numbers are not your numbers."
++ I took the exact John Tromp numbers and derived my calculation from these.
"He's still at 10^44 or so."
++ Yes, with 32 men, using 9 chess sets and including illegal positions.
#1173
Try harder. You can do it.
tygxc wrote:
#1172
"You realize that Sesse is Stockfish running on a supercomputer, right?" ++ Right!
"Sesse's evaluations/nodes per second are not a calculation that can be used to solve chess."
++ Yes, they can: they guide the calculation towards the table base. Consider the top 4 Sesse white moves, consider the top 1 Sesse black move, continue until you hit the table base. The table base gives the final evaluation: draw or not.
"Quantum computers are not going to be applicable to the problem"
++ We will see about that. All statements about the future always are opinion.
"as you have been informed at least a dozen times by many posters" ++ Opinions differ.
"Sveshnikov opinion is not a new input. It was a layman's opinion."
++ No, it is expert opinion by a 2500 grandmaster still active in 2017 and leading theoretician. Compared to him we are all laymen on chess analysis. Experts on engine assisted analysis are: top grandmasters, their seconds, theoreticians, ICCF grandmasters.
"John Tromp's numbers are not your numbers."
++ I took the exact John Tromp numbers and derived my calculation from these.
"He's still at 10^44 or so."
++ Yes, with 32 men, using 9 chess sets and including illegal positions.
#1173
Try harder. You can do it.
You are making the same mistake every layperson makes...you are assuming that Stockfish evaluations represent perfect play...they do not. The "top 4" is incorrect and flawed, as demonstrated by the fact that engines still get better every release. Even more so since traditional engines only recently got upstaged by machine learning engines which are still in their relative infancy. You can't bet better if you are already perfect.
So, "approaching the tablebase" with a set of flawed evaluations doesn't work. You are trying to build a suspension bridge with steel girders on one end and dried mud on the other. And since your bridge in this case is well over half made of mud (8 piece tablebase vs. 32), it's even worse. To top it all off, your bridge crosses the wrong river: engine evaluations (using current engines) reaching the tablebase 100% of the time still does not solve chess.
All statements about the future are, by force, opinion. That's never been in question. The issue is how informed an opinion...and while many people have tried to inform you, you have ignored them all and signed on with a dead GM who was 70+ and looking for notoriety when he tossed out his (unsupported) opinion. Being a famed theoretician in 2-3 lines doesn't cut the mustard for this application.
False. Computers will become too strong. Maybe not winning all the time but maybe never losing either.
Sep 21, 2021
0
#1177
Chessflyfisher wrote:
False. Computers will become too strong. Maybe not winning all the time but maybe never losing either.
It's not a question of how great the superiority of computers over humans becomes, it's whether or not computers will be able to prove irrefutably that chess is or is not a forced win in every possible variation.
Present-day technology is insufficient for a brute-force analysis of every possibility. Whether or not such a calculation eventually becomes possible, or a brilliant new idea that simplifies the process will emerge can only be a matter of opinion.
My own opinion is that sooner or later this question will be solved. Today there is no definitive answer.
ok
Sep 21, 2021
0
#1179
mpaetz wrote:
Chessflyfisher wrote:
False. Computers will become too strong. Maybe not winning all the time but maybe never losing either.
It's not a question of how great the superiority of computers over humans becomes, it's whether or not computers will be able to prove irrefutably that chess is or is not a forced win in every possible variation.
Present-day technology is insufficient for a brute-force analysis of every possibility. Whether or not such a calculation eventually becomes possible, or a brilliant new idea that simplifies the process will emerge can only be a matter of opinion.
My own opinion is that sooner or later this question will be solved. Today there is no definitive answer.
This is probably the most realistic prediction. With the current tools it is impossible to solve but I'm sure there will be technological breakthroughs in the veeeeeery distant future. From there this will influence research in a major way not only in regards to chess.
We will probably not see it in our lifetimes.
#1175
"you are assuming that Stockfish evaluations represent perfect play"
++ No, engine evaluations are always flawed. There are only 3 possible correct evaluations: draw, win, loss as in a table base. That is why to solve chess the calculation must not stop until the table base is reached. However, the flawed engine evaluation may guide the calculation. Based on the flawed engine evaluation we can select the top 1 move for black. If always selecting the top 1 move for black is good enough to reach a table base draw, then chess is solved. For white the proof must look at more possibilities, e.g. 4 per move. If we can prove black can draw after 1 e4, 1 d4, 1 c4, 1 Nf3 then it becomes trivial to prove that black also can draw after 1 e3, 1 d3, 1 c3, 1 Nh3. It is also unnecessary to look at e.g. 1 e4 e5 2 Ba6. The flawed evaluation helps to guide the calculation towards the true evaluation by the table base.
"engines still get better every release." That is true, but if the top 1 black move per engine software version X on hardware Y is enough to reach a table base draw, then that is good enough. Likewise the top 4 white moves on software version X on hardware Y will more or less correspond to the top 4 white moves on software Z on hardware T. This is clearly shown in TCEC, where evaluations by both engines are shown. In like 98% of cases they concur. The 2% where they disagree are the cases where one engine beats the other. Even then the top 4 moves tend to agree.
"You can't bet better if you are already perfect."
++ No they are not perfect and they are flawed, but they are good enough to select the top 1 move for black and the top 4 moves for white to calculate towards the table base.
"Approaching the tablebase with a set of flawed evaluations doesn't work."
++ It did for checkers: starting from the opening, using Chinook to calculate towards the table base.
"reaching the tablebase 100% of the time still does not solve chess."
++ It does. The table base gives the correct evaluation: draw or not. The huge problem is to calculate that deep and that wide that the calculation hits the table base 100% of the time. Already in TCEC the engines running on poor hardware and with a short time start to hit the 7 men table base around move 10 and more around move 20. A top engine on top hardware like Sesse will hit it more especially when guided by the human assistants.
#1159
Work on 8 men table base is in progress.
It is not necessary to go beyond 8 men: we can calculate towards the table base.
Schaeffer solved checkers with a 10 men table base, not a full 24 men table base.
Interesting, thx for the update. But 10 men is a little bit less than half of 24. While 8 is not close to the half of 32. We also have to take account that the complexity of chess is much higher than in checkers so I think going higher then 8 might help more then trying to calculate towards 8 men. It already seems to be an effort to create 8 men but 9-10+ is even more insane in regards to the numbers. Even if we manage to get to 9-10 it doesn't mean that it's gonna be easy as the numbers are just mindboggling from what I've seen so far.