Author: Bob Durrett
Date: 16:39:36 12/02/02
Go up one level in this thread
On December 02, 2002 at 18:02:34, Mike S. wrote: >On December 02, 2002 at 13:24:18, Bob Durrett wrote: > >>On December 02, 2002 at 12:00:13, Art Basham wrote: >> >>>(...) >>>Also, maybe "planning" was a poor word to use here... >>>I rather think of it as "seeing" into the future, >>>and then, making it happen...:-) > >>Predicting the future??? Hmmm. >> >>Well, as far as I know, no human has been able to do that with any degree of >>certainty. Perhaps you mean predicting the opponent's moves, after several >>intervening moves. Two people who play often together may be able to more or >>less accurately predict each others moves and plans, based on past history and >>knowledge of the other's style, preferences, etceteras, but not with absolute >>certainty. > >"Seeing into the future" seems to me to point to a basic difference between a >human's and a computer's method to go for a certain goal: Human planning means, >I see the goal *first* which I decide to go for (or to check if it's possible), >and *then* search for ways (variants) to achieve that goal. The chess program >OTOH searches - more or less - all the variants A complicating factor is the pruning and other tricks used to minimize the number of nodes which will be examined by the engine. Fast engines do not examine nearly all the possible nodes in a typical case. It sounds like you are describing a procedure, used sometimes in human chess, in which a "fantasy position" is first envisioned and then analysis done to see if that position can be realized. In that process, several fantasy positions might be considered [if time permits] and modifications made to the fantasy position(s) if the analysis shows that the original position(s) cannot be realized. The problem I see is that some good realizable "fantasy positions" might exist in unexamined portions of the complete tree. The engine would never find those good realizable fantasy positions. The methods used by the chess engine would have to be such that the best realizable fantasy positions would be found using the chess engine's search algorithm. It is not immediately obvious that currently used search algorithms guarantee finding all such realizable fantasy positions. Incidentally, for the sake of completeness, we should specify that, for our purposes, a fantasy position is defined as "realizable" iff it is possible to reach that position by legal moves from the root position currently being analyzed. >first and then chooses from >them, comparing all the possible "goals" (= end positions of the variants) which >can be reached. > >This is the major difference of the methods IMO. Sounds right to me except as noted. > >Some time ago, I used the following example in an article about the topic of >"vision" (imagination) in chess. The following training position is from a game >Yates-Nimzovitch 1923, where I tried an alternative idea for White: > >[D]r4r1k/p5pp/b1p5/3pNp2/3Q1P1P/q2P2P1/P5B1/1R4K1 w - - 0 27 > >The simple tactics-orientated plan is to bring the queen to h5, creating threats >connected to the idea Ng6+ etc. - So I imagined the queen being at h5 first, >decided (or evaluated) that it would be to my favour, and then searched for the >moves which would achieve that goal: > >27.Qf2 Qc3 28.Qf3 Rab8 29.Rxb8 Rxb8 30.Qh5 Bxd3 31.Ng6+ Kg8 32.Ne7+ Kf8 33.Ng6+ >Kf7 34.Ne5+ Kg8 35.Qf7+ Kh8 36.Qe6 h6 37.h5 Qc2 38.Nf7+ Kg8 39.Nxh6+ Kh8 40.Nf7+ >Kg8 41.Ng5+ Kh8 42.Qg6 Kg8 43.h6 Qb1+ 44.Kh2 Qb7 45.Qh7+ Kf8 46.hxg7+ 1-0 > >That's how I played against Hiarcs 4 at 15 seconds per move, IIRC on P133. I >took that as an acknowledgement for the idea, because Hiarcs 4 was much stronger >than me. Unfortunately Black can defend better. Maybe you can spot the major >hole in my plan :o). > >After Black's move 35...Kh8, we have a nice example of *pattern recognition* >because I think it's much easier for humans to find 36.Qe6! with a common >threat, than for computers. Of course today they'll be quicker finding it, than >1996. Note, that engines will evaluate the position much to Black's favour for a >relatively long time, until they spot Qe6: > >[D]1r5k/p4Qpp/2p5/3pNp2/5P1P/2qb2P1/P5B1/6K1 w - - 0 36 >36.Qe6! best move, although I'm still not 100% sure if White is winning (see >variation above; Black's alternatives at move 37 would require further >analysis). > >Regards, >M.Scheidl
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.