Author: Robert Hyatt
Date: 19:23:18 12/07/99
Go up one level in this thread
On December 07, 1999 at 20:41:22, Ricardo Gibert wrote: >On December 07, 1999 at 20:04:29, Robert Hyatt wrote: > >>On December 07, 1999 at 16:54:15, Ricardo Gibert wrote: >> >>>On December 07, 1999 at 14:22:19, Robert Hyatt wrote: >>> >>>>On December 07, 1999 at 00:01:08, Ricardo Gibert wrote: >>>> >>>>>On ICC, I often see computers winning games against strong players by >>>>>"unfriendly" means. Consider what happens when the position is dead drawn, but >>>>>the computer player does not realize this and makes an unending series of >>>>>aimless moves that drains the human opponent of time on his clock. It isn't any >>>>>secret that computers have "faster reflexes". This is boring and inflates the >>>>>apparent strength of the program. >>>>> >>>>>I propose that computer programs should offer/accept draws when the following >>>>>conditions hold: >>>>> >>>>>1) There have been no pawn moves or captures played by >>>>> either side over the past 10 ply played. Of course >>>>> the 50 move rule counter is perfect for this. >>>> >>>>This is total nonsense. I have seen _many_ games where no pawn move or >>>>capture was played over a 5 move (10 ply) stretch. This has _nothing_ to >>>>do with the game called "chess". >>> >>>Please explain why you consider this to be total nonsense. It is clear that this >>>is reliable indicator of whether substantive change has taken place on the >>>chessboard. >>> >> >>Please read the second sentence in the above paragraph. I don't see how that >>can be misunderstood... >> >> >>>> >>>>Crafty will offer a draw when the eval says "Drawscore" for N consecutive >>>>moves, where N varies depending on the opponent. For GM players I believe it >>>>is 3 consecutive moves. For IMs maybe 5. >>>> >>>> >>>> >>>> >>>>> >>>>>2) The evaluation has remained relatively stable over >>>>> this period of moves. Fluctuating within a _very_ >>>>> small range. >>>> >>>>Again this won't work. Some programs vary a little, and some vary a >>>>lot... so eval change means little in this context... >>> >>>It is meaningful, because no pawn moves or exchanges have taken place _and_ the >>>program has not found any way to improve its position. The combination of the >>>above 2 conditions detect when the program is just churning its pieces around >>>uselessly. Programs do that with blocked position draws or when they maintain >>>seemingly favorable positions by avoiding exchanges that the EGTB indicates >>>would produce a drawn position. Computers win a lot of these positions on time, >>>because the human player can't move instantly like the computer can. Unfriendly >>>chess. >>> >> >> >>It isn't reasonable for the reason I gave. A program might well have 4 >>different possiblities for making progress. Three of the four might fail and >>turn into draws... the fourth might win. I would hardly give up after my >>first attempt. I have watched GM players try 3-4 different approaches to a >>particular endgame before they figure out how to win it... because the other >>approaches lead to drawish position and they backtrack. >> > >I already understood this in my original post. That was exactly why I said it >would not work perfectly and offer draws in winning positions. This can be >ameliorated by increasing the number of ply from 10 to some higher number. >Obviously, if it is set to 100 the "problem" you indicate would disappear, so >improvement in this way _is_ possible. > >BTW, I thought of another possible condition that may also _help_ meet the >possibility of the computer offering a draw in a position it might have won. >You can count up the number of 2-fold repetitions in the last n (I will let you >specify n) ply where a pawn move or capture has not occured. With enough of >them, even if there had been a win missed, the computer probably could not reach >it without allowing a 3-fold repetition anyway. > >There must be some reasonable way of doing this for friendly blitz games. It >does not have to be perfect. > >> >> >> >>>> >>>> >>>> >>>> >>>>> >>>>>Notice the computer player could possibly offer/accept a draw when it is >>>>>material up. It is also possible that the position could be winning for the >>>>>computer, but I think that's OK, since the computer has demonstated an inability >>>>>to find the win. When a position is a winning one, the score should degenerate >>>>>in favor of the side that has the winning position. I know this is not perfect, >>>>>but restricting this to blitz or bullet would >>>>>keep the chess "friendly" and entertaining. >>>> >>>> >>>>A computer demonstrates its inability to win by reaching a dead drawn position >>>>by the 50-move rule or by repetition, or by insufficient material. A program >>>>might try 2-3 different 'plans' before it finds one that doesn't lead to a >>>>forced draw. >>>> >>> >>>You can increase the threshold number of ply to 20 if you want. It would be >>>adjustable. There are a lot of dead drawn positions that the computer does not >>>detect. What I propose is an attempt to remedy that. Please note that I only >>>propose this for those contexts where it is desirable to keep the chess friendly >>>and not when there is something substantive at stake like in a tournament. I >>>thought I made this clear. What do you have that is better in such cases? It >>>doesn't seem like you really read my post carefully. It is an adjustable feature >>>that can also simply be turned off and should be when playing another computer >>>or in a tournament. What could be wrong with that? >> >>The human can already control this. If he wants to play a relaxing game, he >>plays with increment. If he wants to play a sharp time-controlled game, he >>plays a zero-inc game. It is all up to the human. Some computers (mine at >>least) will offer/accept draws when it is reasonable, if it doesn't think it >>has any sort of winning chance at all). >> > >yes, but you ignore human nature. Again, read my reply to Kappler's post in this >thread on this. > >> >> >> >> >> >> >> >>> >>>> >>>> >>>> >>>> >>>>> >>>>>A second proposal I have to make the chess more "friendly", is to keep the >>>>>computer from forcing wins from sheer speed of play. Force the computer to >>>>>consume a little more time per move so that it does not win on time just by >>>>>virtue of its inhuman speed. You can have this trigger a draw offer when it gets >>>>>low on time, _then_ if it is refused, you can have the computer take the gloves >>>>>off and play at full speed. >>>> >>>>This is already done. It is called "playing with increment". If a human >>>>chooses a zero increment game, then he has to play to win or draw within >>>>that time limit. That is _his_/_her_ choice, and has nothing to do with the >>>>computer. I see no reason for the computer to play within that clock time >>>>limit but let the human off if he gets low on time. >>> >>>Read my response to this that I wrote to Kappler's post in this thread. >>> >>>> >>>>If the human insists on playing zero-inc games, then as the saying goes "he who >>>>lives by the sword, dies by the sword." >>>> >>>> >>>> >>>> >>>> >>>> >>>>> >>>>>The "drawback" to all this is that computers employing the above 2 ideas will >>>>>wind up with lower ratings, but I think those ratings will then reflect their >>>>>strength due to chessic reasons rather than non-chessic ones. Computer chess >>>>>programers egos will take a hit when their programs ICC rating goes down, but >>>>>they will gain in the long run by virtue of having produced a more enjoyable >>>>>program that is bound to thereby be more popular. In a serious competitive >>>>>setting or against another computer, these "features" should be turned off of >>>>>course. Perhaps this could be tested on ICC with unrated games to see what the >>>>>impact would be on playing strength. >>>> >>>> >>>>I have been playing chess for a _long_ time. I have won and lost games on >>>>time. I consider the 'clock' to be a "chessic reason" for losing a game. It >>>>is part of the game, included in the rules... >>> >>>I have also played chess for a _long_ time and it has always been considered bad >>>etiquette to try to win dead drawn positions on time in skittles. Maybe you play >>>in a "tougher" neighborhood than I do. >> >> >> >>You just play in a very protected environment. I have seen this happen at >>USCF open events (blitz tournament). At a FIDE event. Even at long time >>controls with a mad scramble at the end... >> >>again, avoid it by playing with increment... not by expecting your opponent to >>let you off the hook after you choose a time control you can't live with... > >How do you know what "environment" I play in? You keep forgetting I am limiting >this feature to friendly games. You mention USCF & FIDE events. So what? What >about them? I wasn't talking about those. Like I said before, you don't seem to >read my posts very carefully. I simply reject the concept of "friendly games" between a human and a computer. Computers aren't "friendly". I have gone way out of the way to make crafty do as well as possible here, by having it offer draws when it thinks the game is drawn, resigning when it is hopeless, accepting draws when it thinks it is appropriate, etc. But this was done to encourage GM players to keep playing. I've not had one GM complain about the way it plays on in some drawn positions, as they understand and accept that in some positions it is drawn but the program doesn't know...
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.