Author: Fernando Villegas
Date: 18:40:52 01/11/99
Go up one level in this thread
On January 10, 1999 at 23:11:10, Oliver Y. wrote: >On January 10, 1999 at 13:17:36, Fernando Villegas wrote: > >>On January 10, 1999 at 09:04:50, Oliver Y. wrote: >> >>>I have drawn a number of games against human masters, down a piece with little >>>or no compensation whatsoever. >>> >>>Earlier posts about true sacrifices might be related to this topic, and I >>>apologize in the unlikely event that I am duplicating an old discussion. >>> >>>I think current programs are particularly bad at having a chance at swindling to >>>save completely lost positions. >>> >>>If there's any interest, I can post two games where I was down a piece, >>> >>>a) I further sacked an exchange, so I'm down a rook; then I exchange my queen >>>for his 2 rooks and a knight..eventually I wind up with a mate in one using a >>>rook, bishop, and knight against his queen and bishop...which I miss in time >>>trouble. >>> >>>b) I sack another piece for some pressure, all along I am dead lost, so this >>>game score would be an embarrassment to the FIDE 2250+ master to post... >>>He blunders, and I have a won position, which I promptly turn into a perpetual >>>mate due to time pressure. >>> >>>Sorry, I guess I should just post the games already...but there's really no way >>>in the next 20 years you'll see programs finding effective swindles against >>>humans... >>> >>>That would be a true sign of Artificial Intelligence, IMMHO. >> >>Hi Oliver: >>I do not see an special extraordinary difficulty in that, at least in conceptual >>terms if not in practical one. Of course your expression "effective swindles" >>left open a door to rejects anything that is not defined as "effective", but >>thinking in swindles as such, the attitude to swindle, effective or not, in a >>practical game, I am not even sure some kind of swindling features were not >>present even in some very old programs. I remember Chess Champion by the >>Spracklen -1980- was willing to swindle -although clumsily-in lost positions and >>the same with his sucesor, Excellence and Par Excellence. I do not know for >>certain with current programs, but in fact, if that feature does not exist, what >>should be done -simplifying a lot- is not more than what we do in such >>circunstances. What we do? >>a) to detect that the opponent has an edge againts us, enough for a sure defeat >>in the long run and so understanding that "normal" playing has not sense >>anymore. >>b) Trying to transform the current form of disadvantage in another more complex >>form of disadvantage in order to get chances, even adding more abstract, >>material disadvantage for the sake of confusion and complexity. >>c) And as we do and as much as material disadvantage already exist, positional >>or justs attacking factors could overrun totally -or almost- material factors. >>In that way the engine could look for enterprising moves even if they are >>materially unsound in the very first ply of the search. >>fernando > >Hi Fernando, > >Excellent. A) is the key, and it should seem relatively simple to program. >This would seem the next logical extension of what Rebel was designed to do. > >B) and C) are much more difficult to do in DEAD LOST positions--I am at a loss >to even define the problem of how to mine for human patterns of weakness... > >Happy New Year... The same for you. And going again to the issue, have you seen how old programs, when lost, try to postpone the mate just pushing pieces to sacrifices, etc in order to stay alive one, two or three moves more? That's a kind of very primitive thinking-without-material-factor in the first place. I suppose you mean dead lost positions those where you are three pawns behind or the positional equivalent to that. I presume that, reached that limit, normal material and positional parameters should be replaced by something else. Again, we have something of the sort in the "combination" setting of many programs, where the engine is taught to search for sacrifices that could go after the enemy King, no matter how much the cost is. In fact, if you are dead lost, only direct attacks to the King have a sense. You can only search for mates. I am not a programmer, but I can imagine that that special module, kamisake kind, could be putted on each time that conditions of dead lost game was reached in order to search for sudden king attacks, no matter what. Fernando Fernando
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.