Author: Dann Corbit
Date: 15:51:29 08/29/02
Go up one level in this thread
On August 29, 2002 at 17:45:32, Uri Blass wrote: >On August 29, 2002 at 17:26:35, Dann Corbit wrote: > >>On August 28, 2002 at 09:19:08, Steve Coladonato wrote: >>[snip] >>>Are you saying that the purpose of the chess engine is not to "solve" chess but >>>rather to implement a programmer's knowledge of chess? >> >>It is absolutely hopeless to try to solve the game of chess with current >>technology. If we coupled together every CPU on earth and ran the strongest >>algorithms on them, it would not solve the game in a million centuries. > >It is hopeless to prove that you solved the game but it does not mean that it is >hopeless to play perfect. Until you have proved it you cannot know if your play is perfect or not. So the goal is still the same. You must prove it to know it. Otherwise, the best you can achieve is "it might be playing perfect" >It is possible that better evaluation function+better search rules are enough to >find always the best move by search even with the hardware of today. I actually doubt that this is true. Alpha beta searching shows that there can almost always be a surprise from compilicated positions. >> >>>If not to solve, then >>>the big advantage the computer has is that it "knows" all openings and all lines >>>and of course does not have to commit anything to memory. >> >>Sometimes, the lines it knows are wrong. Even comically so. >> >>> It is tireless and >>>does not have to prepare for a specific opponent (although I will probably get >>>an argument here). >> >>I think that there is a big advantage to prepare for a specific opponent. What >>openings does he/she/it like the least? If we analyze every move they have ever >>played at long time control on a very fast computer, we can store that >>information into a database. Then we will be prepared for whatever they like to >>do (to some degree). For sure, this is a heck of a lot better than going in >>blind. > >You can do it only if you have games of the relevant program at long time >control and it is a good reason for programmers not to release their program and >not to play public games with it for a long time. You can also do it if you have a copy of the program. >> >>>Also, if not to solve then I am really disappointed. >> >>Sorry to disappoint you. No way a chess engine can dream of solving chess. > >We do not know. >Today no engine is close to it but things may be different in the future. Well, there is always a possibility of a superior algorithm. But I think it is simply computationally intractible. You might get close. Perhaps that is good enough for most people. >> >>> Not that I want to see >>>chess solved but that the development of an engine is just to be the "best" >>>computer playing program and, of course, to reap the commercial rewards. >> >>Most programmers do not write chess programs to reap commercial rewards. There >>are 4-5 good commercial chess programs that sell well. There are 150 freely >>available chess engines. > >The question is how many plan to beat the commercial sometime in the future. I don't even see how that is relevant. There have been private or freely available programs that have beaten commercial programs in important contests and yet they remain either private or free. A goal of winning does not mean a goal of commercial enterprise. I have a goal of winning a chess world championship but I have no intention of ever selling a program. I think those that pursue that course end up being owned by their programs. Or so it appears to me. >>>And, >>>if not to solve, then I don't think the best heuristics nor algorithms are being >>>searched out. >> >>If the algorithms were written with a purpose of solving, then the eval function >>has to return only 1 or 1/2 or 0 for the three possible game states. > >No if solving mean only playing the best move and not knowing the result. If you do not know it is the best move, you will never know if it is or not without a proof. >Maybe it is possible to solve chess with good evaluation and with search that is >deep enough in the right lines. I think it is mathematically impossible, but I might be wrong. > The >>algorithms can be vastly simplified. Unfortuately, we don't have a vigintillion >>years to wait for a sensible reply from the program. >> >>> Just tweaking what already exists. >> >>Do you imagine that someone will invent new chess theory with a program? > >Of course. > >There are already novelties that programs could find by search. That's not what I meant. I meant a new idea like the positional strategy of Steinitz or the control from the corners like Nimzowitch. Chess progams produce data, but they do not produce ideas. At least not so far (that I have seen). ><snipped> >>> I think >>>the amateur programs have an advantage here in that there are no commercial >>>distractions to the development of their engines. >> >>I think the professionals have an advantage since they get to work on computer >>chess 40 hours a week or more. > >I think that this assumption is wrong. >I know that Amir Ban has a full time job not in chess programming. I think most of them work at chess more than Amir does then.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.