Author: Robert Hyatt
Date: 14:24:26 02/13/04
Go up one level in this thread
On February 13, 2004 at 16:54:44, Bob Durrett wrote: >On February 13, 2004 at 15:33:03, Robert Hyatt wrote: > >>On February 12, 2004 at 10:28:32, Bob Durrett wrote: >> >>>On February 12, 2004 at 01:59:22, Gian-Carlo Pascutto wrote: >>> >>>>On February 11, 2004 at 19:29:19, Bob Durrett wrote: >>>> >>>>>If it were possible to obtain a chess-playing program which could be set to some >>>>>playing strength and also set to emulate human play, then the use of such >>>>>programs would be used extensively world-wide by serious chessplayers for >>>>>tournament preparation. I would be one of those. >>>> >>>>Deep Sjeng already has those features. >>>> >>>>-- >>>>GCP >>> >>>That's great! My only reservation is that I wonder how realistic the emulation >>>[of human play] is? >>> >>>If a GM wishes to use "Deep Sjeng" to emulate the play of human GMs, can that GM >>>get what he wants? >>> >>>If an IM wishes to use "Deep Sjeng" to emulate the play of human IMs, can that >>>IM get what he wants? >>> >>>If an amateur wishes to use "Deep Sjeng" to emulate the play of human amateurs, >>>can that amateur get what he wants? >>> >>>My assumption has always been that realistic emulation of GM play is not >>>currently possible, emulation of IM play reguires a very strong engine, and >>>emulation of amateur play may not require a super-strong engine. >>> >>>Bob D. >> >> >>The problem is that you can not "emulate" something you don't understand in the >>first place. And I don't know of anyone that claims to know how a human plays >>chess... >> >>Emulation of a human requires something we don't yet have, and probably won't >>have for _many_ more years... > >Unfortunately, due to my bad choice of terminology, we have a symantics problem. > >What is emulation? As is the case for most words in the English language, words >have multiple meanings. The intended meaning must be inferred from the context. > Unfortunately, I failed to make the context clear. Sorry. > >As explained in another bulletin, my desire, as a user, is to obtain a >chess-playing machine which plays like a human. Whether or not it thinks like a >human is of no interest to me, as a user. I wish for the play of a human to be >emulated and not to emulate the inner workings of human brains. > >Sorry about the confusion. : ( > >Incidentally, the following definition of "emulation" appears on the internet: > >COMPUTER SCIENCE: To imitate the function of (another system), as by >modifications to hardware or software that allow the imitating system to accept >the same data, EXECUTE THE SAME PROGRAMS, and achieve the same results as the >imitated system. > >The problem with specialized fields of study, such as Computer Science, is that >specialists within that field develop their own specialized meanings for words >which also have different common meanings in common usage. This is one of those >cases. > >Bob D. OK... now we are on the "same page". Still a tough problem. The difficulty lies in the fact that a chess program is a collection of "parts". Evaluation. Search. Search extensions. Etc. You can turn them "up" or "down" but I have never seen a program that really could act like a 1500. IE you might turn the evaluation way down, but tactically it will likely be stronger than any 1500 player on the planet, but it just plays poor positional chess to deflate its actual rating. Even a plain vanila search has trouble reaching down to the lower levels, as with no extensions, a program still might average searching to 12-14 plies or more, and 1500 players won't stand a chance. You can tone the depth down further, but it won't make the kind of blunders a 1500 player will make. You can add some randomness to the evaluation, but it will will not feel the same, as it will make a stupid move here and there, but not the same kind of stupid move a human might make, or not in the typical kind of position... It is actually easier to write a _strong_ program than to write one that can be "dumbed down". Several have "personalities" including Mike's version of Crafty, but while they reduce the overall strength, the results don't feel "right" to me, as a human playing other humans...
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.