Author: Terry McCracken
Date: 14:16:14 08/30/03
Go up one level in this thread
On August 30, 2003 at 15:44:23, Timmay wrote: >Ok, first off, the name Tim was taken, so I took the next closest thing....grin. >I have liked South Park in the past, I haven't been able to watch it lately >though. Too busy! > >On the more serious side towards Uri...Computer vs. computer results simply do >not matter. If one computer beats another its most likely because of the >opening. NO computer is perfect. So in one position that they got to by chance >(or via opening book specifications) Junior 8.0 might have a superior code, or >more applicable code to the specfic situation I should say, whereas in another >position Junior 7.0 might be superior. So it goes both ways. It's more likely >the talent of the new openings book or time management that does it than a >better evaluation or search (based on the kind of moves it makes that I've >seen). > >Big things that impact results are both the opening...essentially the specific >position the engine comes to, and secondly time management. For instance Junior >8.0 plays quite a few dumb moves by human standards. But since it plays a >computer the computer won't gradually take advantage of it later on like a human >would, so Junior 8.0 still might win. > >How do I judge what computers are good? Simply by objectively looking at the >moves they make. I don't care about any of those automated computer vs. computer >results where the opening books take over and if one engine has bad time >management and in a critical position (to human standards) moves in 15 seconds >and commits a big blunder where the opponent takes a little bit longer next move >and discovers the refutation, I mean that stuff is by chance and worthless. The >only way to judge strength is humans vs. computers. > >If Junior 8.0 makes a silly move, only the experience of a human master can take >advantage of it. Kasparov could have, but he didn't want to lose to another >computer. A computer simply can't. > >So since computer-computer results are unreliable, the only way we have to test >computers is via test positions, or games against humans. It is simply not >possible to know that there is actually an improvement between two engines (once >they get to a high tactical level) unless you do millions of strategic test >positions between each computer which is not feasible. Just because there was a >game ortwo or three when Chess Tiger 15.0 beat 14.0 or where Junior 8.0 beat >Junior 7.0 or computer that Junior 7.0 couldn't beat before, it doesn't mean >it's better. > >**Important point on the why of the previous paragraph. The reason it doesn't >matter is because if a program beats another, and it is NOT because of opening >book or time management (which is rare) it is because the program it played is >deficient in the code that is used by the winner since practically speaking >they're about equal in the raw tactical department. For instance programmers >might think implementing a big king safety code is important simply because they >win many games against other top computers because those computers "don't" have >a big king safety code. However, in doing so, the moves are objectively weaker >according to strong humans who really aren't deficient in ANY codes....grin >Aside from distraction or occasional blunders. Humans can ward of king safety >code computers attacks quite simply and prophylatically if they have experience. >So what I'm looking at in computers is the raw strength. Do they make good chess >moves or not? None of this "is my computers code going to beat that computer's >code?"** > >Also, it's not fair to say that a computer understands chess as well as a human >master. You mentioned I can't judge them because they're rated 2600+ or whatnot. >But that's ridiculous, Junior is maybe 2900 in tactical positions, but in quiet >positions I'm stronger than it is. And if I'm "on" and not distracted I could >compete in the tactics as well. I've defeated both Juniors before, so I think >I'm rightly able to critique programs because I have a sense for what moves are >good or not which is the only way to judge a program, by moves, not by results >against other computers. If there's an occasional blunder I miss then virtually >ANY program could help me with that. > >Computers only do only what they are taught to do whether it be right or wrong. >Even I (nonethless Garry who analyze deeper than me at many occasions and is >usually a bit more accurate) understand chess better than any computer program >ever could. As I said before ANY reasonable tactical program could prove a human >wrong in a tactical blunder that he commits. That doesn't mean the computer is >better, it just detects mistakes when they can happen occasionally. But if we're >talking about the "best" computers, being the most talented chess players, not >just a computer which spots tactics that humans occasionally miss, it has to do >with the quality of moves it makes in the quiet positions. The best programs in >that regard are the ones I mentioned, Fritz, Chess Tiger 14.0, Nimzo 8.0, and >maybe Shredder. Sorry Timmay!:) You're not really telling us anything we don't already know, and some things that are obviously debatable. Terry P.S. I kicked Junior 8s chips in a blitz game, as it failed to recognize my attack that was about to shred it's King, untill it was too late. But I find the best programmes have similar shortcomings, but for the most part they are tough to beat and easy to lose to. You never touched on HIARCS 8? Anyways, I'm waiting for HIARCS 9 to be sent out to me, on Sept. 9. A Chess Programme Junkie!:o)
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.