Author: Fernando Villegas
Date: 12:05:04 10/22/00
Some level of discussion was started by Thorsten Czub, a week ago, concerning what, according to his view, Gambit Tiger means for the future of this field. Now with the sucess of this program in the dutch tournament the debate is or should be even more interesting. Czub insisted that Christophe Theron is a kind of continuator of the Wittington’s efforts, of the ideas he embodied in CSTAL. He added that Christophe has gone along the path established by Wittington, one that has been described as a rejection of “bean counter philosophy”. Later Christophe retorted that his path has been all the time different, BUT that as a result of his own endeavor he has arrived to a point where he can understand what Wittington preached since long ago. I understand that also, in a way, Christophe by now seems to share at least some of the Wittington philosophy. Now the problem is to understand what is that "thing" that maybe they share and so what is that thing that constitutes, according Czub, a new paradigm. In fact, what is that new thing Gambit has, IF it has it.. After some thought, I dare to propose these ideas and how my musing went on about what this paradigm maybe is. I do it with the purpose to unleash better and more acurate ideas. My goal is to understand, not to preach. I have nothing for certain in this matter, of course. a) First I believed Gambit was just a case of extreme setting of a more or less “normal” program. In any program, if you make strong changes in just the values of the piece table, it will behave entirely different. You can even make of it a sacrificing animal just valuing the queen as a pawn. b) Then I realized probably is not the case. Changing greatly those values in a normal program will cause odd behaviour, but not smarter playing at the level Gambit Tiger has shown. c) Then I have thought it should be or could be something new, no matter if that “new thing” was intended to be so or we are facing one of those cases of serendipity that you can pick up for the hundreds in the fields of sciences and technology. The same Theron has said the surprise he himself has felt respect the awesome perfomance of Gambit Tiger. d) The first idea, that probably has been in the mind of all us, is that perhaps we have a case of “speculative” search replacing a normal, “bean counter” one. But then I have considered this is not the way to understand the "new thing". In first place “speculative” is a very speculative word that tends to say very little. We should reject it for good. What is more, If you think on it you will see that ANY program is speculative as much as his calculations stop long before the ending of the game. Is like adding a column of numbers and stopping in the middle; you know how much they add until then, but if they ask you the final number, you only can do a guess. In chess, short of the final row of moves, you ever make a guess: you take into account everything you can and you go as deep as you can, but short of the final move, your selection of the best move is just an educated guess; you dim it as “best” with incomplete data. In fact always there is a chance that it was not the best move after all, precisely because of new facts that appeared only later. So, every -or almost- chess "calculation" is not so much calculation as a kind of speculation. e) But then, if speculation in that wide sense is unavoidable in chess -and in many other dynamic processes of decision- I have realized that there are at least two very different ways to do it. The normal one is trying to get the most of the search capabilities inside the frame of data that is searchable. That is what we unsconsciously refer when we talk of calculation. In other words, when you calculate -or the program do that- what happens is that you try to see, ponder and measure better and better the data that already are inside the scope of the searching function. The explicit or implicit philosophy of this approach is this: if you cope the entire room of significant facts that you already have AND if you make that room of searchable data bigger and bigger, then you diminish the degree of uncertainty and you can do a better guess the next time. In certain sense this is like walking giving the back to the directions of march, so you are looking at what you live behind, not to what is coming. Then, if you see carefully and take into account the changes in the landscape you can actually see, then you can guess more or less what you are going to face with your back... until a point. This is what a normal chess program tends to do, I believe. Even “extensions” are just a camouflaged way of looking back because after going to “future” the evaluation is about the data that are really gathered. I dare to say that any event of the future that appears to be fully examined is by that same reason not really a future event but an extrapolation of a past one. Future involves per se a high degree of uncertainty. f) I presume, then, that the other method of “guessing the best”, in this case by Gambit Tiger, try to handle the uncertainty not so much with “data” that are clearly defined in order to choose a move, but on the contrary, trying to chose a move capable of producing the realm of facts that, although not clearly stablished, seems to shape a better scenary. In other words, the program try less calculate than to produce a battle scenary fitted to his code. So his unit of analysis is not so much data compared by scores, but situations compared by his capacity to give a better chance. That is to say, maybe Gambit does not take a chance about moves evaluated and qualified as such looking behind his horizon of search, but take a chance about a course of action accepting a bigg room of uncertainty. g) If it is so, then the issue is not about “knowledge”. Is we talk of knowledge, then we must precise what kind of definition we have of knowledge. I think the issue here is not to have or not to have more or less lines of code with hundreds of algorithms trying to translate certain principles of chess playing theroy, BUT a different knowledge where the core is not the measure of movements according to a looking-back glance, but a knowledge oriented to handle the program in the middle of a fussy landscape where what matter are opportunities. So I would bet that Theron has not fulfilled his task just replenishing his codes with theoretical lines or patterns, but shaping algorithms capable of steering his programs to situations where his tactical capabilites shine at his most. I hope at least I was clear about what I think. To be correct or not is another thing. I wait your comments... fernando
This page took 0.02 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.