Author: Dann Corbit
Date: 13:09:07 02/15/99
Go up one level in this thread
On February 15, 1999 at 14:32:13, Bruce Moreland wrote: [snip] >Considering them in detail in this case involved forcing a move at the root, >which I don't think qualifies. We can combine that with the original analysis for FIN.08: ---------------------------------- 8 - Fin Spassky - Byrne : c5!! (1974) (1...dxc5 2.d6 Kd7 3.Bxc5 Bd8 4.Bb4 Ke6 5.Kc4 Bf6 6.Bc5 Bd8 7.Bd4 Kxd6 8.Be5 Ke6 9.Bb8) OR (1...bxc5 2.Be1 Kb8 3.Kc4 Kc8 4.Ba5 Bd4 5.Bd8 Be3 6.Bg5 Kb8 7.Kd3 Bc1 8.Bxh4 Bxf4 9.Be1 Kc7 10.Ba5 Kb8 11.h4 Bc1 12.h5 Bg5 13.Ke2 c4 14.Bd2 Bf6 15.h6 1-0 ) ----------------------------------- >A few days ago I made the argument that in a sense all moves that lead to mate >are equally good. I think it was important to do it in that case since it >appeared that there was an attempt to apply some absolute meaning to the notion >of a best move. > >I think this argument would be tiresome if I invoked it in every circumstance. >If someone shows a move that wins a rook, destroys counterplay, and leads to a >position that any reasonable player would agree is won, I am not going to say >that the solution is spoiled because there is another move that wins a pawn and >may lead to a technical win. Well, any move leading to mate can be called a 'best move'. Any move not known to lead to mate might even be a worst move (the only move that surely leads to getting checkmated) unless we can somehow see far enough ahead. I see it somehow like this: 0. Gets you checkmated (only considered if provable) 1. Horrible blunder 2. Blunder 3. Not an obvious blunder 4. Careful tactical blunder checks ok 5. Intense tactical blunder checks ok 6. #5 & Strategic Position checks ok (OK, so programs don't do this so well) 7. Certain draw from inferior position 8. Certain checkmate If we had an infinitely fast computer with infinite storage, we could know all of the above instantly. EPD test suites run the whole spectrum. Often, the classifications are wrong. It is *very* difficult to tell if #6 is valid or not. >I think that there are many positions where computers can offer valid insights. >And by this I mean that they find a good move and offer a convincing line. At some measure of depth. Ten years ago a program that could find solutions 8 plies deep at tournament time controls would be a barn burner. Not today. >I think there are also many positions where you can let the computer search >forever and it is still in the dark. Many of these are middlegame positions, >but there seems to be another category that includes endgame positions of this >sort. That's only because we have not developed the right model yet. Eventually, that will be solved (I think -- I could be wrong). >Perhaps 1. c5 does not win. I haven't looked at the analysis. But I've fiddled >with the position a bit. 1. c5 is a pawn sacrifice. It is possible that other >pawn sacrifices, or even a piece sacrifice, will follow from this position. Then we should be able to demonstrate this or refute it. If it is just guesswork, then the position has no value. >I think that this kind of position is dangerous for computers and it is very >likely that there is a win here and your computer won't be able to see it. It would be very nice to be able to quantify that. If we can do a strategic evaluation of the board, then the next thing that should follow would be an explanation. From an explanation we can experimentally arrive at a quantification. From our quantification we can produce a testable model. >You can burn through 13 full moves in an endgame very quickly. Put a bishop on >the board, put the kings somewhere where they can't do much, and put a few pawns >down for the other side, in such a fashion that they can't be simultaneously >stopped by the bishop. This is a -1 position for the pawns, but after many >moves by both sides one of them queens. This particular instance examined about 4 billion nodes. Now, I will admit that there are positions that fool chess engines. It could be that this is one of them. >>By supplying the pv for both possibilities, someone with a lot more chess >>knowlege than I have can say, "Yes, there may be something here..." or "No, this >>is clearly wrong. The position is Zugzwang here and so the computer simply did >>not consider the proper choice..." >>I will go so far as to say that unless a checkmate for one side or the other is >>undeniably proven then any 'best move' candidate is in question. Here's why: >>Imagine a 5 year old (non-prodigy) who has just learned the rules of chess. >>They look ahead one ply at a time for the immediate move. They have a move that >>would look best to them. But now, after 10 years of playing, the 15 year old is >>much better. Perhaps she looks ahead 'n' moves. She would beat the 'best move' >>of the 5 year old. Now, she faces a GM. The GM thinks ahead 'n'+2 or more >>moves at all times. The GM's best move is better than her best move. The GM >>faces a super GM. The super GM thinks one ply farther than the GM, on average >>(or however the strategy is superior -- probably nobody knows). From the same >>position, the Super GM's best move may be different than the GM's. Now, let's >>take that Super GM and let them think on a move for a whole month. After >>thinking it through very carefully, using database systems and computer programs >>to assist, the Super GM's best move may not be the same move they chose >>previously. >> >>In other words, a 'best move' is only 'best' compared to a worse move unless it >>definitely and undeniably leads to mate. > >I think that in this case it is likely that 1. c5 is the only try for a win, and >that this is one of those marginal positions where this can be seen by fairly >weak humans (although the win itself may be harder to see), but that computers >may have a hard time understanding. I think that discovering and classifying positions such as this could lead to quantifying the problem. If so, then they are the most important thing in computer chess right now.
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.