Author: Robert Hyatt
Date: 14:12:02 06/14/98
Go up one level in this thread
On June 14, 1998 at 12:38:51, Fernando Villegas wrote: > >Hi all: >I cannot believe that after so many post from one side and another still >nobody seem prepared to accept how complex is this issue respect to the >apparent conflict between fast search and knowledgeable but slow search. >I have even seen a post where Mark said that the path to get the bottom >of the game is search, pure and simple. Yes, but, what kind of search? >Search is just a word, not a thing in itself. Is not the answer, but the >question. So we should go to the beginning and do some questions. Let me >recall some of them: >a) To be faster or stronger on anything else that can be quantitatively >measured is not equivalent to be in the right path. You can be faster >to follow the wrong path. You can dig deeper in the wrong place. You >even can discover something -as alchemists did- trying the wrong method >with the wrong assumption and the wrong ideas. Then is not good argument >to say because this and that has been got thought search THEN we must >continue with that no matter what. >b) Is not the same to be acquainted with the previous point than to have >a real an specific different and best method to replace the first. Is >not enough to say you are digging in the wrong place if don t say where >that task should be done. If there is not real alternative, then to dig >deeper is the only one for now. So, is not enough to blame at the > stupidity of current search methods. >c) Nevertheless, although the advocates of smart search instead of >brute force have been incapable of doing much more, that does not mean >they are not aiming at the right directions even if they cannot do much >else than that. To stuck stubbornly in the faster approach and believe >IS the approach just because has given results until now is no smart or >scientific at all. Anything stubbornly and systematically done gives >some results, but they are increasingly less -law of diminishing >returns- and costly. In fact, all scientific and technological >development happens not when a method has failed, but on the contrary, >when it has reached his perfection. The ground is then ripe for a new, >superior principle. To be blind on that after so many historic lessons >is not a showing of intelligence. >d) This new principle surely will be something superior in kind and >level to the actual search, mainly based in the examination of how good >moves are with respect to parameters established in a table. That there >is another method is no fantasy: just take a look a at what experts or >above players do when playing chess. Surely they does not go move after >move looking his possibilities and then measuring how good they are or >not. They look at the position as such and understanding what is >happening -a game of chess is an historic process, not an static math >problem- they decide what should be done and only then they begin to >look for moves and only at the end of all this they perform calculations >to test the moves chosen on the ground of the previous understanding. >e) The previous point is not to say that tactical calculations are >inferior, meaningless, etc. A game is lost or won on the ground of >specific, tactical moves. That is the reason computers are so strong. >But even if you can win because of tactical shots without any concern >for what is happening in the game, that does not prove the game IS >JUST tactical shots. We must differentiate between what chess is and how >players can play it. Millions of games are won of lost by patzers that >have no idea of nothing, and nevertheless a chess result is produced. >One of them won, the other lost. Does that means the winner had >understood the game? Equally, thousands of games are lost by humans >against computers, each day: does that means computer have a real grasp >of the game and so no further investigation about different method is >needed, just more of the same, more speed ? >f) The fact that the supposedly current smarts programs -as CSTAL- does >not show to be that smart and fall in many traps, tactical shots, etc, >does not prove anything, either. First steam locomotives were a lot >slower than horses. CSTAL is a kind of experiment . Maybe even is a >complete failure. But, again, that does not mean a thing. The debate is >not between CSTAL and the rest of the programs, but between search >methods based in the examination of many K-moves, one by one, of another >method that try to understand the position as strong human player do. I >would like to see a debate opened here on the ground of this last >specific point: how a program could simulate that kind of qualitative >understanding of position and his dynamic. Or is not possible? Bob? Don? >Amir? First an introductory statement, then an opinion. Humans and computers are different. *totally* different. Computers don't do things like humans do, they will *never* do things like humans do, until computers evolve into something similar to our brain, with all of its abilities. *not* for a long, long, long, long time... Given that, then I firmly believe in a statement I have made many hundreds of times... you can reach the same point with knowledge *or* search. Yes, you can abstract many types of knowledge... and yes, I can create many types of search extensions... and eventually, we both are going to reach the same point... because it is going to take your "knowledge" just as much computation time to come to a valid conclusion as it is going to take my search algorithm to reach that same conclusion. But, from experience, my search algorithm is going to be orders of magnitude simpler, and more bug- free. Given that, I depend heavily on search, while using knowledge to fill in the holes that I can't cover with search *today*. But as time passes, this knowledge gets replaced by better search with faster hardware, and then I use new knowledge to fill in the holes that are left. You can take a lousy search, and fill in the holes with knowledge. But the knowledge often has more exceptions than necessary, which leads to wrong conclusions. To date, search is working, particularly when augmented with the right kinds of knowledge to help with things that search can't currently handle. But if you slow down too much, pure search will be more than enough to win the game... fritz is a good example here... You might have great pawn structure and piece outposts, but if you are mated, that doesn't count for much...
This page took 0.01 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.