Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Just learning capability?

Author: blass uri

Date: 14:02:08 06/13/00

Go up one level in this thread


On June 13, 2000 at 15:55:35, Dann Corbit wrote:

>On June 13, 2000 at 15:34:42, blass uri wrote:
>>On June 13, 2000 at 14:57:01, Dann Corbit wrote:
>>>On June 13, 2000 at 14:48:34, Mogens Larsen wrote:
>>>>On June 13, 2000 at 14:19:31, Tom Kerrigan wrote:
>>>>>That's just your opinion of "understanding." I think that computers understand
>>>>>chess just fine.
>>>>
>>>>Yes, that's my opinion. I'm also of the belief that you're wrong.
>>>>
>>>>>But opening books are an inherent part of computer chess programming. Otherwise,
>>>>>why does every program have one? I don't see the difference...
>>>>
>>>>Most opening books are not a result of programming (except when it comes to
>>>>format, but obviously doesn't count) or output generated by the program itself,
>>>>ie. learning files and games. Thereby it isn't an integrated part of a program,
>>>>but an attachment.
>>>>
>>>>>But humans have teachers and read books. Surely programmers can teach their
>>>>>programs...?
>>>>
>>>>The problem is that computer programs gets the answers without asking the
>>>>question so to speak. Making mistakes and learning is an integrated part of a
>>>>teaching process. Programmers don't teach anything, they offer something the
>>>>program can't refuse nor reflect upon. They offer knowledge that has already
>>>>been processed. I'm sorry if it's a little unclear, but I'm neither a programmer
>>>>nor a grandmaster :o).
>>>
>>>Computers understand tactics.  They don't plan at all (at least none that have
>>>source code available plan).  Tactically, they are clearly better than humans.
>>>
>>>As far as opening books, they are an accumulation.  Consider chess 200 years
>>>ago.  Were the same openings be played as today?  No they won't.  And why not?
>>>Because after hundreds of thousands of games under careful scrutiny, weaknesses
>>>were found in various lines of play.
>>>
>>>It is obvious that humans "use opening books" because the style of the game and
>>>types of attacks used change over time.
>>>
>>>Consider all the games played by GM's and super GM's in the past century.  This
>>>is the equivalent of millions of hours of supercomputer effort at analysis.
>>
>>I do not think that using opening books by computers is unfair but
>>I disagree that opening books are eqvivalent to million of hours of
>>supercomputer because supercomputer is faster than humans and it can calculate
>>in 1 second things that humans needs a lot of years to calculate.
>
>The converse is also true.  Take the LCT II test and let your favorite program
>chew on it.  Let me know when it gets them all right (I am guessing *NEVER*)
>because programs stink at positional choices.  They don't understand gambits
>very well, and they don't understand sacrifices very well.  Very little tactical
>stuff happens in the first ten moves.  Computers have a really hard time seeing
>deep horizons.  Humans who invented the good openings were *PLANNING* which is
>why they turned out so well.  If we were to just try and brute force all the
>combinations, we will never come up with anything worthwhile.  Too many choices
>to examine them all.
>
>The *only* thing computers are better at than humans is tactics.  But they are a
>lot better at that particular facet.
>
>>The main problem is that humans do not know the way that humans  thought in
>>order to invent the opening theory so they cannot explain it to programs.
>
>It's a mistake to try and make computers play like humans.  Humans have a
>trillion neurons in a network.  Computers have (at most) a tiny handful of
>general purpose computing units.  If we try to model the brain with conventional
>CPU's it's going to flop.

I think that humans do not use the trillion neurons to play chess.

I agree that it is not a good idea to try to model the brain because the human
brain cannot remember like computer(even kasparov forgot his home preperation in
one of his games and lost and I do not suggest to teach programs to forget)but I
think that it is a good idea to learn some good ideas that the human brain knows
to use and teach computers these good ideas.

Uri



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.