Author: Rolf Tueschen
Date: 05:02:15 10/10/02
In the Berliner Zeitung Wüllenweber gave an interview: http://www.berlinonline.de/aktuelles/berliner_zeitung/sport/.html/183583.html "Fritz wählt Eröffnungsvarianten, die aus der Theorie kommen. Er fällt viele Entscheidungen nicht durch Berechnung, sondern durch Nachschlagen in seiner Datenbank mit früheren Großmeister-Partien. Was Kramnik äußerst geschickt macht: Er spielt Varianten, die - rein statistisch betrachtet - eigentlich für ihn schlecht sind. Das Programm freut sich dann über den subjektiven Vorteil, nun die Damen abzutauschen. Objektiv macht Fritz alles richtig, doch dann gelingt es Kramnik immer wieder, die Maschine zu überraschen. Er tut Dinge, mit denen Fritz nicht rechnet. Die für eine Maschine nicht logisch zu erklären sind. Fritz berechnet jeden möglichen Gegenzug auf den möglichen Gegenzug. Dabei entsteht ein Suchbaum, der so schnell wächst, dass das Programm trotz massiver Geschwindigkeit nicht so tief in die Stellung abtauchen kann wie ein menschlicher Großmeister. Ein hochintelligenter Mensch wie Kramnik kann im Gegensatz zu Fritz in abstrakten Mustern denken. Das ist ein Vorteil des Menschen. Einer wie Kramnik weiß eben, dass irgendwann in zwanzig Zügen - mehr als jede Maschine der Welt berechnen kann - die weißen Bauern am Damenflügel schwach werden. Das ist ein Stellungswissen, eine Intuition, die Fritz nicht hat. Unser Problem ist, dass das Programm in seinen Schwächen im Moment vorgeführt wird." Without going into the details of programming I can show why programmers like Matthias have so many difficulties. Because they don't understand how easily human chess masters play chess. 1. The belief in statistics in chess is nonsense! You must be capable of playing all kind of positions. Weak, weaker, good and better positions. The most difficult positions, and that is known to real chess players of a certain class, are the _won_ positions. Because here you can only lose your face! The pressure is immensely high! So the argument that Kramnik played weak positions so that Fritz "believed" that it could be good for him to exchange Q is simply stupid. Because it shows how primitive and naive Fritz is. Matthias, who is also responsible for the ChessBase database is simply confused by the differences between real chess and training with databases and engines. He, and with him many more "experts" took the whole mass of information for aproof of strength! Which is wrong! I have said for years that real masters, once they entered into fighting mode would make putty puke out of all the million bases... Such hyperbole is only really dangerous for weak amateurs! With no memory. Even 2000 players have outstanding good memory and can play through hundreds of lines! I know what I'm talking about! The problem begins always when the opponent begins to deviate from the main path. Then you must play on your own! And Fritz is simply too stupid to play on his own. So better forget about the Elo 2700 numbers... 2. Matthias is claiming that human players could rely on patterns and must not calculate all to the end. So Kramnik would see that in 20 moves (that is 40 plies!!) the pawns will be weak. Yes, fine, but fo programmers it would be easy to find routines for such "weaknesses". Certain configurations of the pawns are simply weak! And programmers must understand that. Of course with 1400 they can't. But even Vincent with over 2200 show little interest in such topics. Of course it could get painful in the end if you realise that there are thousands of such patterns to judge. But the machine will go through that in seconds! You must only invent code. Instead of just adapting to the new hardware twice a year. 3. Matthias is talking about human intuition! That is wrong! Perhaps he must rely on intuition on his already good level of an amateur. For Kramnik this is not intuition but he simply knows what patterns are weak and what patterns are unclear. The nonsense Fritz played was already visibly weak for 1800 players! That is not intuition, that is experience from hundreds of games. And so on... Moral The bad side of CC issimply that thesingleprogrammers developped a general view on chess from the frog's perspective. What they must learn is that chess can be trained. And they must implement the patternsinto code one by one... Of course this is a year long practice. Not something to be done in some weeks or months. Rolf Tueschen
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.