Author: Gordon Rattray
Date: 03:48:16 08/11/02
Go up one level in this thread
On August 11, 2002 at 05:46:36, GuyHaworth wrote: > >Sacrificing to reach 'EGT territory' and a guaranteed win has been seen several >times. The fact that EGT-wins are 'attractors' was noted in "3-5-Man Chess >Maximals and Mzugs", GH/PK/JT/CW [ICGA_J v24.4, Dec 2001]. > >"Any port in a storm" comes to mind. > >Others can add their examples but I can remember: > >a) a Shredder game which folded down to KBBKN >b) Deep_Junior-Deep_Fritz (elimination match) Game 4 which folded to KRPKR > - as in ICGA_J v24.4 > - you would expect KRPKR to be targetted relatively frequently in this way > >For an extreme example, put the bK on a1, the wQ on b3 and other force >marginally winning for White elsewhere. Then play 1.Qa2/b1/b2 KxQ ... > > >Aesthetics are a personal thing: one might say that watching the victor ditch >surplus force to guarantee the win is aesthetic. I would say that seeing small >advantages turned more often by infallible play into wins adds to the >aesthetics. I agree that aesthetics are a personal thing. I guess it's similar to whether you like Karpov's style of play or Kasparov's, etc. I personally like the sort of finishes that EGTs often produce, even if it involves throwing surplus material away. For me, it highlights a very logical strategy. Since chess involves finding the "best" move in the shortest amount of time, getting into a situation where "perfect" play can be done almost instantly sounds ideal. It doesn't matter to me that shorter wins are sometimes missed as there is no penality for this. I also like the fact the computers often have their own way of doing things - they don't have to play like humans. ;-) > > >If the reward-system for chess is generalised from 0/0.5/1 to: > > - draws scores 'd' points > - win scores 'd + w + a*moves_to_mate/b': w>=d, a >= 0, b>0 > >then you are more interested in the quickest mate. > >[ As in the Argentina tournament with Hiarcs, (and football in the UK) I am in >favour of creating better incentives to win: I throw this in here! ] I think that creating better incentives to win, such as the ones mentioned above, would introduce more "luck" into the game, and hence spoil it. For example, supposing that I noticed in an earlier round that a player falls for a quick mate (e.g. White playing moves like e4, Qh5, Bc4, and Qxf7+ mate), and I now face this opponent myself? Should I attempt a similar mate, even although I know that my moves could be punished if he/she plays more correctly? If I play more sensibly, and take longer to win, I don't score as much... despite play more correctly... Also, given a level position, should I play a move that gives me a win, but only if my opponent fails to spot a very subtle defence that would cause me to lose? Chess would become more about risk taking, and gambling on what you may or may not get away with against a given opponent. And that's not what chess is about. The current scoring system 0/0.5/1 reflects the game better. > > >There is an argument for training chess-engines to be more competent in >endgames, either without EGTs completely, or with win/draw/lose EGTs only. This >would arguably improve their pre-EGT endgame play. > >The learning/evaluation process requires a definition of 'competence', and I >have just proposed one such [ICGA Computer Olympiad 7 Workshop Proceedings, >Maastricht 2002]. > > >The point about EGTs being 'unfair' in human-computer play has been raised more >than once. They are only literally 'unfair' if their use breaks a rule of the >competition. > >However, I think there is an argument for giving the carbon side of the >competition some bionic help against the silicon. The October K-computer >matches will have a sense of anti-climax if they are won by errors that others >easily see. One might argue that humans should have access (via computers) to >opening-books and EGTs to level the playing field somewhat on the memory side. > >In my 'Bionic Chess', we could also allow the human a 'vet this move by N-ply >search' service ... and more time than the computer gets. I don't see anything wrong with "bionic chess". However, for matches which are termed such as "man versus machine", then it has to be just that - no adding or removing to/from what we generally accept as being "man" or "machine". So keep the EGTs with the machine, and no PC asistance for the man (or woman ;-)). Gordon > >g
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.