Author: Ed Schröder
Date: 10:10:39 12/13/99
Go up one level in this thread
On December 12, 1999 at 18:44:00, Amir Ban wrote: >On December 12, 1999 at 14:54:27, Enrique Irazoqui wrote: > >>On December 12, 1999 at 13:29:29, Amir Ban wrote: >> >>>On December 12, 1999 at 09:48:31, Enrique Irazoqui wrote: >>> >>>Dear Enrique, >> >>Hi Amir, >> >>>I'm surprised to read that you subscribe to this fast vs. knowledge nonsense, >>>which is as false as it is popular. >> >>It is not that I subscribe or unsubscribe. It is just that as a flower collector >>I do not know if it makes sense or not. I see hints in both directions and >>realize that I have no idea, and that's why I posted about it. Talking to other >>programmers I hear both sides: ones, agreeing ardently with you; others, >>opposing with equal passion. >> >>> The simple truth is that all programs are as >>>fast as their author can make them, and have as much knowledge as their author >>>practically knows how to put in them. >> >>Are you sure? I see some evidence to the contrary, particularly in the endings. >>Some of the strongest engines have no idea about such elementary stuff as bad >>bishops or Philidor endings. > >By "elementary", you mean that it's in the opening pages of your old textbook ? >That doesn't mean it's the most important thing. I think many programmers >started to write a program with good intentions to cover everything the textbook >says, but found there are more urgent things to do. Indeed so many that >sometimes you completely forget about the textbook. > > >I also hear that some programmers get rid of >>knowledge in order to speed up the search, which causes the horror of other >>programmers. It seems that there are two opposite schools of thinking about al >>this. >> > >There are no two schools of thought. That's the fallacy. There are tradeoffs >everywhere, and every programmer tries to strike the right balance. The better >ones manage to find the right compromises, or better yet, to find ingenious ways >to avoid making compromises. > >Whatever the compromise, the part you keep is always called "knowledge", while >the part you discard is not :) > > >>>>Mind you, I also think that without intuitions, whatever that is, exact, >>>>verifiable thinking tends to sterility, so from my let's call it feminine >>>>intuition (astrologically I am the intuitive cancer, double cancer in fact, soon >>>>triple I guess :(, what crap this astrology), and going back to this comp-comp >>>>vs. human-comp discussion, I sometimes wonder. To make it short, when looking at >>>>the Rebel-Baburin and Rebel Sherbakov games, I "know" that the fast finders >>>>couldn't play as well as Rebel. >>> >>>Untrue. J6 finds the critical choices in Rebel - Sherbakov to be rather easy, >>>and in my opinion understands Baburin - Rebel better than Rebel. It thinks that >>>at some points Baburin mishandled a white advantage (e.g. 28. Qc7 ?). >> >>I don't have it yet, so I can't tell about Junior 6. But would Junior have >>played some Rebel moves that defined the game, like 14...Ng4? Only Rebel and >>Shredder 4 seem to play this specific kind of game, and Rebel's play was >>impressively consistent. Fritz, for instance, was fooling around all the time. >> > >Deep Junior on 2x350 finds it 5 minutes or so, with score -0.14. > >By the way, doesn't white have something good with 28.Rf1 (Qd5 29.Rg1)? I get >+0.50 after a few minutes analysis. > > >>>>Following the games with Fritz 6 was >>>>overwhelming evidence in this direction. >>> >>>I understand it was much too optimistic for black, >> >>Yes it was, but it was not only a problem of optimism but of not having a clue >>of what was going on in the Baburin game. >> >>> but then, so was Rebel, or so >>>we are told. All programs are stupid when their evaluation is way off. This >>>happens to Hiarcs, or every other vaunted "knowledge" program, quite often. >>>Fritz, by the way, often shows understanding that would make the so-called >>>knowledge programs green. >> >>Here we disagree. I have seen this better understanding of Fritz at times, but >>rarely. And I like Fritz, but for other reasons. >> >>> On the other hand, why this alleged >>>>positional, human-like (?) superiority wouldn't also show up in comp-comp games, >>>>so "knowledgeable" computers would compensate with it for their slower tactical >>>>speed? Because it doesn't compensate and comp-comp is decided by tactics. >>> >>>That's wrong. Computers kill other computers all the time when their opponent >>>doesn't understand a position. It compensates for order of magnitude in speed. >> >>This is wrong in the way you put it and quite central in my perplexity. Let's >>assume as an hypothesis of work (is this English?) > >Working hypothesis, yes > > >that a tactical test gives >>the same performance ratings than thousands of computer-computer games. If this >>were true, and I think it is, it would mean that what makes a difference in >>comp-comp is tactics, while the "knowledge" that different programs have >>built-in is of similar value, different here and there, but similar on the >>whole. > >Do I understand that your secret test suite is a tactical one ? If so, I don't >believe in it. You would find that J6 scores less than J5 (which didn't score >much higher than J4.6). > > >So you have that a program can be helped tremendously by its knowledge in >>some given positions, but the same will happen to others in the same or >>different positions and to the same extent. Then the tactical ability will >>prevail, and that's why in the SSDF list the fastest finder is on top, followed >>by the second fastest, and so on. > >That's incredibly simplistic. Do you really think that eveyone's positional >knowledge more or less evens out so tactics prevails ? I would much sooner >assume that everyone's tactical ability is more or less the same, so positional >understanding would decide. That's simplistic too, but closer to the truth. > > > Now, that's comp-comp. My question is if >>human-comp is any different. I mean, some claim that their programs are better >>suited or tuned for human-comp and that they have "knowledge" that fast finders >>don't. If this is true, why is that this extra knowledge doesn't make them >>perform better in comp-comp? >> > >I would certainly expect this supposed extra knowledge to show up in comp-comp >games and make a difference, otherwise we have cause to suspect an empty claim >was made. > >I think those who claim that some particular program is specially geared for >comp-human games are aware of a fine record it has, but are ignoring that other >programs have an equal or better record in comp-human events. For example, I >would guess that Junior and Fritz are "prime suspects" for being "comp-comp >specialists", while in fact both have now a long and excellent record against >ranked players (Fritz mostly in active chess). Also, without great statistical >significance, in the WCCC99 two programs, Shredder and Cilkchess, with little >experience against humans drew respectably their GM opponents, while Ferret with >its long record on ICC was the only one to lose. Amir, can you post Junior's 40/120 results against GM's. I vaguely remember yours were good too. Ed >Amir > > >>>>Is >>>>this "superior" understanding only the adaptation of a program to human playing, >>>>with the only value of making human life more miserable in chess, and we believe >>>>this anthropocentric approach greater? Is there really a difference between >>>>comp-comp and human-comp? So what's up? I really wish we would be less of a >>>>flower collector and more of a botanist. >>>> >>> >>>Just my opinion. >> >>Just my perplexity. :) >> >>Enrique >> >>>Amir
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.