Author: Thorsten Czub
Date: 07:05:59 07/13/02
Go up one level in this thread
On July 13, 2002 at 05:10:45, Ed Schröder wrote: >>Ed isn't saying that a change of best move found makes the engine (on faster >>hardware) stronger, but the lack of move change shows the engine isn't improving >>(or playing weaker). > > >My friend, you got the point, well said. > >Ed the important question is: WHERE in the main line are the changes, and WHY is there a change. IMO you cannot find out by quantifying THAT a change occurs. Todays chess programs do rarely play for an idea or a plan. in fact they SOLVE TEST SUITES when playing chess. they find a best move in a position A instead of finding an idea or weakness to play against in a game. the difference is obvious. if you have no plan what to do, you run arround, like a child, you see something interesting, you watch it, your focus is for a short time (one move) on the new thing, and than you lose focus and walk arround seeing the next "interesting" thing. this is how chess programs play. they have no plan. they have, to make fun about it, infinite plans, but all plans only a few plies. than they switch it. this is random chess. i doubt that you find out about strength or strength progress by quantifying change of moves. to give an example: richard langs programs rarely changed the best move. they mainly changed the line and shuffled the moves BEHIND the first move of the main line. while fidelity programs OFTEN changed first move and main line. Julio Kaplans programs also rarely changed the first move and more created different ways of doing an idea. the question is not IMO how often you change where, but WHY you change where you do it. is the change because you have no clue about the position and try something else, or is the change because the move fits better to you plan. but how to teach chess programs to create a plan that is in the focus for more than ONE or TWO positions in the game ? as long as computer programs do not try to create a plan, they will IMO not make much progress. they will become deeper, they will become stronger in tactics. but they run arround without an idea what to do. they wait until they see the opponent blunder. and they use this. but more than waiting ?? more than random picking of a good move ? no. the target should be (IMO) to teach them to plan. to reach out weaknesses and attack those weak points. and to FOLLOW those plan. even when they see other things. IMO this cannot be done with a program seeing/looking for EVERYTHING. the program must have blinker like a horse, otherwise it would not follow the ideas. so when a program believes a sac is "right", it makes no sense for it to expect and calculate moves where the opponent TAKES the sac. with a blinker it expects the opponent NOT to take the sac. CSTal is an example of a program that behaved nearly opposite way than Genius did. Genius pruned much the own moves, and calculated almost all opponent moves. as a result it played defending, seeing everything but not taking all chances. it played boring. CSTAl pruned the opposite moves much, and looked for many many interesting sacs and ideas in its own move lists. this way it e.g. looked for a sac, and pruned away that the opponent takes it. because why should the opponent take a sac that "works" ? this is a blinker. the program DREAMS about an idea. it was a lot of work to keep those behaviour in balance and make it succesful at those time. but it worked. it worked e.g. because the opponents were often unable to calculate in forward about what CSTal is doing next. they could not. they were completely blind for CSTals ideas. because CSTal was different. when you have somebody coming from a far region, behaviour strange, it is very difficult to calculate him. todays chess program behave almost all the same way. therefore it is so easy for the opponent to find out what it is doing. between humans, this is completely different. each human beeing is different, dreams or not, follows a plan or not, and makes blunder, bluffs, etc. why do todays chess programs not bluff ? they only need to create a thesis that is deep enough the others cannot refute. in the last point of view, almost ANY move is a thesis. and with ANY move nobody knows if this is a bluff. so why not make this as a kind of plan. cstal did. it bluffed. the reason cstal is beaten has nothing to do with the WAY it works, only with the search depth. the same problem BTW. genius had. when the NORMAL programs come deeper than you, you cannot show the strength of your beeing different.
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.