Author: Uri Blass
Date: 06:38:12 08/28/02
Go up one level in this thread
On August 28, 2002 at 09:24:14, Rolf Tueschen wrote: >On August 28, 2002 at 09:13:45, Uri Blass wrote: > >>On August 28, 2002 at 08:52:48, Rolf Tueschen wrote: >> >>>On August 27, 2002 at 07:43:26, Uri Blass wrote: >>> >>>>On August 27, 2002 at 06:51:11, Rolf Tueschen wrote: >>>> >>>>>On August 27, 2002 at 04:44:28, Uri Blass wrote: >>>>> >>>>>>On August 26, 2002 at 18:13:56, Rolf Tueschen wrote: >>>>>><snipped> >>>>>>>>Here is the list of the programs above 2600. >>>>>>>>You can see that the porgrams played usually more than 400 games >>>>>>> >>>>>>>Yes, Uri, I knew it. But! I wrote "that could be added up". >>>>>> >>>>>>I did not understand what you mean by "could be added up" so I ignored it. >>>>>>I understand your point that learning is important in their matches when it is >>>>>>not important in real tournaments when programs are allowed to change book >>>>>>between games. >>>>> >>>>>Although we have almost coincidence, I can always find new divergences and new >>>>>ideas to explain certain things. So the debate isn't boring at all. >>>>> >>>>>The new idea here is the tendency of CC experts, and you are one of them, to >>>>>always finding new tricks to cripple human chess. I do not say that this >>>>>crippling is the ratio of CC and a conscient behaviour. It's more a fallacy if >>>>>you once concentrate too much on 'performance' without reflecting the overall >>>>>situation in chess. For me CC is a part of chess. >>>>> >>>>>Let me elaborate this point. >>>>> >>>>>For me the learning function was a technology which allowed the chess program to >>>>>come closer to correct chess! But if I read you and others, I understand that >>>>>the learning has only one single goal, namely to prevent that a clever opponent >>>>>could repeat certain lines _no matter_ if the line itself is good or bad >>>>>objectively, but because later, perhaps due to some different failure, the game >>>>>was lost. For me, honestly, this is a pervert understanding of chess. I for one >>>>>would say that if objectively the position is good, or at least not bad, it >>>>>shouldn't be excluded by the learning function only because later the game was >>>>>lost. The perversion is even that if a certain continuation is favorable for the >>>>>machine, that it's still deactivated because tue to later events the game can't >>>>>be kept open by the machine. This is foolish! I would recommand that the >>>>>programmers should better study possibilities how they could teach their progs >>>>>to understand the positions of such lines. Please take all what I write with the >>>>>certainty that I do not know exactly what really happens with e.g. the learning >>>>>function. So please do not reject the whole context only because a technical >>>>>detail might be false. It should also be clear that I do not address you in >>>>>particular, since you are trying to get to the meant most of the time. Because >>>>>otherwise we end up in endless circles of repeatitions. >>>>> >>>>>Let me mention the books too. I am not against books! But I am against books >>>>>whose sole meaning is to put the machine into the position to play position that >>>>>they couldn't play out of principal weaknesses. This is what I call cheating or >>>>>boasting or fraud, all terms please without juridical implications but only in >>>>>relation to fairness and gentleman sports. - >>>>>For me it is absolutely out of imagination why CC experts are using such >>>>>technology. Is it the dream of perfection? I don't think so, because why then CC >>>>>experts can sleep a single night without nightmares when they think of the known >>>>>absolute weaknesses of the progs. Why don't they concentrate on these weaknesses >>>>>instead of using GM books, whose deep meaning the machine can't understand at >>>>>all!? >>>> >>>>I do not think that programmers in general are not interested in fixing the >>>>weaknesses of programs in the opening but it is not a simple task. >>>> >>>>programmers change the evaluation function of chess programs to do them better. >>>> >>>>I think that having an epd file of the most common positions in human-human >>>>games that most programs cannot find the right move in dew minutes may be a good >>>>idea. >>>> >>>>I believe that it is possible to fix the evaluation and the search rules of >>>>programs in order to help them to find the right moves. >>> >>>If you meant that these positions are added to the GM books with looong >>>variations included, all coming from human masters and not from the program >>>itself, I would not appreciate it. >> >>No >>I meant to add knowledge to help programs to play better by search and >>evaluation but before adding knowledge it is important to know exactly what is >>missing. >> >>I think that having the positions when programs blunder may be prodctive for >>programmers. >> >>I do not think good programmers are going to hurry to fix a mistake in one >>position by knowledge because the problem is to define small number of rules >>that can help in a lot of positions so they first need to observe a lot of >>positions when programs blunder and only later define some rules when the target >>is to avoid most of the mistakes without creating different mistakes. >> >>Uri > >Thanks for the clarification. I'm sure that working on such a dictionary of >actually unsolvable positions would initiate a new aera of CC! > >BTW it's a pity that the month long work of GM Benjamin for DB2 is kept in >secrecy. It's so bad for CC. > >Also, your idea is part of a more general approach for computers analysing in >principal _all_ opening moves from 1 to 30. With good coordination this could >become a worldwide project. The idea comes from a German who made the proposal >in CSS. > >(BTW would you support the exclusion of the lines in question from the GM >books?) > >Rolf Tueschen No I do not think that we can forbid programs to play lines that they do not understand but I think that teaching them to undersatnd more lines is an important target because it may help them to play better if the opponent take them out of book. I believe that today it is possible to beat programs even with lines like 1.e4 f5 if you find the way to continue for the right positions that they do not understand. Uri
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.