Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Strength of the engine in chess programs

Author: Robert Hyatt

Date: 12:23:27 05/21/02

Go up one level in this thread


On May 20, 2002 at 22:46:46, martin fierz wrote:

>On May 20, 2002 at 20:53:40, Rolf Tueschen wrote:
>
>>On May 20, 2002 at 20:04:21, martin fierz wrote:
>>
>>>On May 20, 2002 at 10:15:44, Rolf Tueschen wrote:
>>>
>>>>Let me demonstrate a little thought experiment. If I would gauge (in 2002) the
>>>>actually most known chess programs against say 1000 human chess players (first
>>>>step) to get some insight into the Elo numbers, I would expect that the top
>>>>programs would at best get Elo performances of 2200 - 2350, if I let the engines
>>>
>>>2200? you must be kidding! my rating is 2240 FIDE and even if i start all my
>>>games against fritz with 1.h3 or some other (quite sensible) moves to take it
>>>out of the book, i have no chance against it.
>>>maybe someone here could experiment with a few top programs using no book
>>>against other top programs.
>>>
>>>>How many years from now it will take to develop a real chessplaying robot who
>>>>could participate in human tournaments completely on his own? Buying new books
>>>>he reads, asking collegues for some information about this or that,
>>>>differentiating between truth, lies and irony.   ;-)
>>>
>>>i wonder why you have a problem with chess engines using opening books. is it
>>>that they did not find these moves on their own? if yes: can i ask you about
>>>your opinion on a computer-generated opening book? that is, an opening book
>>>which the chess engine works on day and night, finding opening lines all by
>>>itself? it stores this information and can retrieve it instantly and without
>>>failure (unlike humans), but unlike today's opening books it has computed
>>>everything itself.
>>>the reason i ask is that my checkers program has exactly such an opening book.
>>>after only a few weeks of analysis of checkers openings, my book contains much
>>>of the human opening theory for checkers, and some corrections of it. everything
>>>was discovered by the engine itself. it could never find some of the moves "over
>>>the board", but this book just serves as a memory for it's analysis - very much
>>>like a human chess master.
>>>
>>>computing such an opening book for chess is much harder, since there are many
>>>more viable moves. but if you went on to write a screen saver application to
>>>distribute the task, who knows - maybe something good will come of it.
>>>incidentally, this is just what dann corbit is doing. jeroen noomen once wrote
>>>me he also has had some success with automated opening book construction in
>>>chess.
>>>
>>>aloha
>>>  martin
>>
>>Yes, of course, that would be a good method, although I doubt that you would get
>>serious results. How many nights the machine should calculate? Who will decide
>>that? And - much more important, what will you do with the losing tickets? Would
>>you outsort them by hand? Honestly, then you could let it as before with the GM
>>books. How many years it will take? Because already the exactly solvable
>>endgames did cost Ken Thompson weeks and months.
>
>the endgames have little to do with this question :-)
>my machine for the checkers book calculates as long as i want it to. the book
>just grows and grows. i don't know what you mean with "losing tickets" - i know
>there are errors in my book - but as time goes by and the analysis gets deeper,
>the errors disappear. i don't throw them out of course, no point in generating a
>book with a computer if you have to fix it afterwards...
>
>
>>But I think your question shows my own in a different light now, hopefully. The
>>machine is either prevented to play garbage which it wouldn't find out in weeks
>>on its own and it is told to play moves, the quality of it couldn't discover
>>either. Not only the move itself but possibly a 30 moves variation...
>the computer does find the moves on it's own in my example - i know nothing
>about checkers and cannot feed it any good moves. yet my opening book is very
>good - just because my checkers engine finds good moves. it could not discover
>many of the moves over the board, but then all grandmasters put in lots of
>analysis at home (and with computers too!) before they play something over the
>board - it is very similar!
>
>>Doesn't sound fair in games against human players, no?
>so no, if the human GM is allowed to analyse at home, why should the computer
>program not be allowed to do so?

Why bother asking logical questions?  This discussion is _not_ based on
logic.  It is based on emotion and agenda...  The human commits things to
memory.  But the computer cheats if it does this.  The human remembers
things during analysis, but the computer cheats by "writing them down"
if it does so.  Humans memorize things between rounds, but the computer
cheats if it does.  This will go round and round forever...





>
>>As to the question of your own record against FRITZ, I would beg you to read the
>>first two postings I wrote in this thread. With "normal chess" you will be in a
>>disadvantage by definition against the machines because you must always
>>calculate at a given depth and you can't correct mistakes. More appriate is to
>>play directly the weaknesses of the machine. Of course that has no longer to do
>>with training for human chess. You must always search for a way to exploitate
>>the computer's horizon.
>this is easier said than done!
>
>>Take a look at the many games Eduard Nemeth has played.
>>He is even weaker than you. And then you must become familiar with typical
>>characteristics of computer play. Exchanges. It's a totally different form of
>>chess.
>i know... nothing against eduard, his games are wonderful.  but i think that in
>a match against a computer like fritz,junior,hiarcs,tiger, at standard time
>control, on decent hardware, he would be totally chanceless. i have played games
>with fritz where one takeback gave me a draw, even in blitz with 1s increment.
>but i always needed that one takeback. maybe on a longer time control i would
>have made that draw. but that happens once in ten games. you always only see
>eduard's wins. i'm sure he loses too, but doesnt post these games here :-)
>
>>Excellent calculating power you must have,
>i do not - remember, i have 2240. if i had excellent calculating power, i would
>have 2400+. by my definition of "excellent calculating power" of course...
>
>i know a lot about computer weaknesses, and i really tried to play for them -
>but it is not easy. much stronger players have tried and failed. i don't believe
>the opening book makes such a big difference in the computer's rating against
>humans.
>
>aloha
>  martin



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.