On Aug 19, 8:28 pm, Ralf Callenberg wrote:
20.08.2007 01:07, Chess One:

Computer vs. human matches is an unfair competition while the computer
has recourse to a huge opening library. Thus, the program doesn't need
to know much about play in the first phase of the game.

Or anything, It plays moves it cannot calculate for itself, otherwise, ipso
facto, it wouldn't need the book.

It is probably stronger with a book than without. That doesn't mean a
program can't do without.

Hey, now here's a guy who obviously has no experience
with this subject; anyone who would choose "probably" to
qualify an obvious fact is hedging his incredible ignorance
for safety.

It's real simple, fella: turn the book "off"; disable it, and
then watch how your computer tries to analyze the chess
openings. In sum, like a patzer except that the computer
is much, much stronger at tactics, so it gets by okay.


[1] Why did you post this to rec.games.chess.misc and not to
rec.games.chess.computer?

Because computer geeks don;t understand the game at all, and its hard to
find even an 1800 player in their ranks.

You mean people like the programmer of Rybka, International Master Vasik
Rajlich? Are computer admistrators incluced by your definition of
"geeks" - as International Grandmaster Gerald Hertneck? In fact the
percentage of chess players earning their life in the software industry
is quite high.

Hey, this is the first time the title "IM" has popped up
among programmers; years ago, one of the testers was
an IM, but he bashed computers by issuing a challenge
to pay money if one could take him in a match. For
many years, even lower titles such as FM or NM were
fairly rare among the actual programmers, making the
chess programs far superior to their creators in strength.


Every now and again there is some pretence that the fact chess engines can't
play chess from move 1 doesn't even matter )))

When Hydra smashed Adams, it had only a very shallow opening book. It
wasn't move one - but definitely not a book reaching into the
middlegame. The Hydra-team is convinced, that their program does better
going its own path. When Rybka was playing his last two matches with
odds - do you think the Rybka-team set up a sophisticated opening book
for such starting positions?

Apples-to-oranges; in order to assess this properly,
you need to compare the Rybka pawn-odds openings
book with GM Benjamin's pawn-odds preparation.
Looking at the games, I would say that the human
prevailed here, but how can you clearly and distinctly
separate the contempt factor issue from the openings
book issue? It's a bit messy.


-- help bot

On Aug 20, 12:09 pm, Guy Macon http://www.guymacon.com/ wrote:
Phil Innes wrote:
I am not a fantascist like Guy Macon who thinks any
science is 'boring'

I was quite specific. I said *YOU* are boring. And you wouldn't
know what Science was if it bit you on the arse. You are a
blowhard who thinks that personal attacks are an acceptable
subsitute for logic and evidence. Go away. You bore me.

In fact, the claim was that IM Innes' *posting* was boring,
not IM Innes himself. I couldn't help bot notice that in
addition to this "observation", there was a claim to not
have bothered reading the boring post, so one wonders
how it was that the "observation" was derived. LOL


-- help bot

help bot wrote:

Guy Macon http://www.guymacon.com/ wrote:

I would like to examine the argument that "all 256 possible
configurations are sound, balanced, and fully playable."

This looks like a *claim* to me, not an argument.

Sorry. I wrote too fast. I would like to examine any evidence
and/or arguments supporting the claim that "all 256 possible
configurations are sound, balanced, and fully playable."

The proposed solution doesn't solve the described problem.
The computer can simply start with an opening library that is
256 times bigger. It would tend to encourage machine-generated
opening books rather than building the books from master-level
games, which is IMO a Good Thing, but it would not remove the
advantage the computer gets from having a huge opening book.
It would instead increase that advantage.

Eventually, such monstrous books could be created,
but right off the but they would either not exist or be
of rather limited quality.

4000-computer LANS are quite common. If one of them was
used to spend every night for a month generating a book
with, say, 24 CPU-hours on the first ply, 12 on the second,
and so on leveling off at ten CPU-minutes per move, that would
be (gets calculator) 6 moves per hour * 12 hours * 30 days
* 4000 computers = 8,640,000 book moves in a month. the key,
of course, is starting with a program that isn't an idiot in
the opening phase.

Because one of the free programs I downloaded came
with no openings book, I have seen quite a bit with
regard to how well computers can play and in what
style, sans openings books. Generally speaking, the
faster the games, the more important tactics are and
the slower the games, the more likely one can convert
a strategically won position against them. On the whole,
even the strongest programs are way behind human
understanding of strategy, although their vast superiority
in tactics makes this far less noticeable. Take the
pawn-odds match between Rybka and GM Benjamin,
for instance; Rybka won, but in the opening the human
prevailed rather easily. The program was unable to do
anything until some complex middle game arose, and
even then, several games were drawn by repetition, not
by simplification.

I agree with all of the above. It does, of course only
describe current computer programs. There have been some
attempts at making computers better at strategic thinking,
but doing so means longer evaluating each node on the
search tree and thus making the search a lot slower, and
so far evaluationg more positins with a faster but less
sophisticated evaluation beats the alternative. There is
some speculation that adding plies to the search starts
giving the computer diminishing returns. If that's true,
future chess programs may go back to using slower and
smarter evaluations.

The entire concept that openings are a disease which
needs to somehow be remedied, seems to require more
or better-reasoned support. The professionals seem to
like the status quo, and base their careers largely on
openings preparation, and this in turn generates a lot
of premature draws which I expect they do not feel were
"uncontested", on account of the contest in openings
prep. before the games. Even so, it is not good for the
game's image to have all these grandmaster draws, nor
to have important games decided by computers before
the game begins.

I rather suspect that having many draws actually helps
the image of chess in the minds of the ab=verage non-
player.

--
Guy Macon
http://www.guymacon.com/

20.08.2007 15:40, Chess One:

If after 20 years of evolution, and fantastically improved chess programs,
the answer to my question is still not known, I rest my case that any claims
for the engine are not based on facts.

I don't get your point. Computers *can* play starting with move one.
They are just weaker without their opening books. The same would be true
for a GM: take away his opening knowledge, and he would play weaker -
otherwise, why would GMs go through the hassle and invest so much time
in the opening preparation?


[1] Why did you post this to rec.games.chess.misc and not to
rec.games.chess.computer?
Because computer geeks don;t understand the game at all, and its hard to
find even an 1800 player in their ranks.
You mean people like the programmer of Rybka, International Master Vasik
Rajlich?

No - indeed he is an exception to the norm - and look what a difference that
makes! Why do you choose some exception as if it were the rule?

Well, let's have a look at some programs from the last Computer World
Championship:

Stefan Meyer-Kahlen, author of Shredder, not active since several years,
but his last German rating was about 1900, which corresponds to an Elo
of roughly 2000. Johannes Zwanzger, author of Loop - Elo 2335. Vincent
Diepeveen, author of Diep - Elo 2290. So, in a field of 12 engines, 4
programmers (Rajlich included) have an Elo safely above 1800. One may
add one of the well known people of the scene, Matthias Feist, co-author
of Fritz, who has a German rating of about 2000 which corresponds to an
Elo of ca. 2100. Rajlich may be by far the strongest, but your claim,
that computer "geeks" are generally weak players simply does not stand a
proof.

But lets us
not continue with fantasies about things which are testable - the standing
issue is as above.

Just for the record: the above claim about "geeks" and their playing
strength was a fantasy by you, a claim not grounded on anything more
than prejudice.


Every now and again there is some pretence that the fact chess engines
can't play chess from move 1 doesn't even matter )))
When Hydra smashed Adams, it had only a very shallow opening book. It
wasn't move one - but definitely not a book reaching into the middlegame.

I understand you like to make these rather general claims - as if - you were
answering what I wrote, but we both know that they do not answer the
question, or even pretend to - and computer-geeks always do the same
schtick!

What are you talking about? Your claim is, that computer programs can
not play chess from move one, right? I brought examples of programs who
play against GMs quite successfully although they start playing at move
one or at least quite early. So, why does this not address your point?


Sorry, I don't want to encourage more specualtions - my question is actually
a scientific one, which is a different method of inquiry.


Would you be so nice to repeat your question, please? I am not sure that
I got it.

Greetings,
Ralf

help bot wrote:
On Aug 20, 12:09 pm, Guy Macon http://www.guymacon.com/ wrote:
Phil Innes wrote:
I am not a fantascist like Guy Macon who thinks any
science is 'boring'
I was quite specific. I said *YOU* are boring. And you wouldn't
know what Science was if it bit you on the arse. You are a
blowhard who thinks that personal attacks are an acceptable
subsitute for logic and evidence. Go away. You bore me.

In fact, the claim was that IM Innes' *posting* was boring,
not IM Innes himself. I couldn't help bot notice that in
addition to this "observation"

"Consider the source"

Not a valid point in debate - but very useful guidance in life.

--
Kenneth Sloan
Computer and Information Sciences +1-205-932-2213
University of Alabama at Birmingham FAX +1-205-934-5473
Birmingham, AL 35294-1170 http://www.cis.uab.edu/sloan/

20.08.2007 11:16, Guy Macon:

In my opinion, the basic strategy of using human-to-human games
from the past as the basis for an opening book will become less
and less useful for computers as they improve to the point where
they are much, much better than the best humans.

What you seem to forget is: a move taken from opening theory is not
based on a single human. It is evaluated in an evolutionary process over
many cycles of refinenment. The GMs nowadays use the aid of computers to
find new moves. So it doesn't matter, that computers are so much
stronger than humans - the level of refinement in the opening theory is
still far beyond what a single machine can achieve. Whether you let it
think for 10 minutes or an hour per move, doesn't make much of a
difference. It is still far below the time - of humans plus computers -
which was put into the moves of current opening theory. The only thing
you can get this way is an opening book which might better fit the
program than "normal" theory. But this would mean, that the created
opening book is best suited for the program which created it, while
another machine might not take a lot of advantage from it.

If a computer that is playing at, say, 2 minutes per move time
control has on its hard disk a complete tree of all possible
moves and responses up to ply N with its moves precalculated
at, say, 20 minutes per move, it will play those first N plies
better by using the book moves.

Did you make an estimate, how much time it costs for an N of, say, 10?


All of this takes computing power, but it is computing power
that is used before the start of the game, not computing power
that could be used to play better during the game. It is also
a task that is particularly well suited to being done on a
large LAN during non-working hours or even in with distributed
computing such as is used by SETI@Home and Folding@Home.

Before you go on with your idea, you might to a rough estimate as
suggested above. It might turn out, that even with a lot of computers at
hand, this is not feasible. I don't know it, but it's your idea, you
should give it a rough test up front.

Greetings,
Ralf

On Aug 20, 1:48 pm, Ralf Callenberg wrote:

Well, let's have a look at some programs from the last Computer World
Championship:

Those who compete for the championship are likely
among the strongest, not necessarily a representative
sample of the "geeks".


Stefan Meyer-Kahlen, author of Shredder, not active since several years,
but his last German rating was about 1900, which corresponds to an Elo
of roughly 2000. Johannes Zwanzger, author of Loop - Elo 2335. Vincent
Diepeveen, author of Diep - Elo 2290. So, in a field of 12 engines, 4
programmers (Rajlich included) have an Elo safely above 1800. One may
add one of the well known people of the scene, Matthias Feist, co-author
of Fritz, who has a German rating of about 2000 which corresponds to an
Elo of ca. 2100. Rajlich may be by far the strongest, but your claim,
that computer "geeks" are generally weak players simply does not stand a
proof.

Once upon a time the same list would have averaged closer
to 1500; things are more competitive now.


But lets us
not continue with fantasies about things which are testable - the standing
issue is as above.

Just for the record: the above claim about "geeks" and their playing
strength was a fantasy by you, a claim not grounded on anything more
than prejudice.

And the odd belief that he is nearly-an-IM, far above
most of the players you listed above. His alter-ego,
Rob Mitchell, keeps telling me that in his expert(??)
opinion, IM Innes is better than I am, though he knows
not how good I am so the question is, on what basis
does his mind allow itself to gauge his own strength?


What are you talking about? Your claim is, that computer programs can
not play chess from move one, right?

Closer to: they do not attempt to play chess from
move one, having been programmed to access an
openings reference by-rote instead.


I brought examples of programs who
play against GMs quite successfully although they start playing at move
one or at least quite early. So, why does this not address your point?

The latest odds match had Rybka using a special
pawn-odds opening book; just thought I would throw
that out there since Rybka is, after all, the best of
the best.


Sorry, I don't want to encourage more specualtions - my question is actually
a scientific one, which is a different method of inquiry.

Would you be so nice to repeat your question, please? I am not sure that
I got it.

This reference to science is simply IM Innes' way
of complaining about the fact that none of the big
whig programmers took up the gauntlet and tested
their engines sans openings book; my interpretation
is that they were scared to face facts, like the fact
that most chess programs using references to look
up their moves are relatively weak sans-book. They
simply never had to compete without the reference
work so their weaknesses were not rooted out.

It is odd that so much time has elapsed without
anyone taking an interest in this. One reason it is
important is for game analysis; imagine you played
a game where your opponent (or you) went out of
book very early, and you want to know objectively
how badly you played, and what was a much better
way to play the position; further imagine that you
have no access to human GMs, but only your
chess program. In this situation, my program will
often as not suggest ludicrous moves as "best
play", so I have just wasted my time until we get to
the tactical errors (where it is very useful indeed).


-- help bot

On Aug 20, 2:01 pm, "Guy Macon, Engineer" "http://
wrote:

I was quite specific. I said *YOU* are boring. And you wouldn't
know what Science was if it bit you on the arse. You are a
blowhard who thinks that personal attacks are an acceptable
substitute for logic and evidence. Go away. You bore me.

In fact, the claim was that IM Innes' *posting* was boring,
not IM Innes himself. I couldn't help bot notice that in
addition to this "observation", there was a claim to not
have bothered reading the boring post, so one wonders
how it was that the "observation" was derived. LOL

Oooh! A pedantry fight! My favorite!! :grin :0


"We pedants don't 'fight'; we quibble." -- Prof. Knowital


The use of the plural ("boring posts" instead of "a boring post")
clearly implies that I found all previous posts by Phil Innes to
be boring

Hmm... now we are getting a transmission which
seems to imply having read *all* previous postings
by IM Innes... very strange! Nobody, and I mean
no body, could possibly have read all previous
postings by IM Innes. No way, no how. Except
of course, IM Innes himself.


and thus that Phil Innes himself is boring. (They were
and he is.) Based upon past performance, I predicted that the
post before me would also be boring and so decided to delete it
unread.

Fair enough. Just don't try and tell us that your
"evaluation" was that his post was boring, while at
the same time admitting you didn't even read it;
that just makes you look dumb. (You don't *want*
to look dumb, do you?)


I could be wrong, of course; perhaps this is a rare
exception -- a Phil Innes post that isn't dripping with boring
contempt for all who are not the great Phil Innes. Was it?


Not exactly. But if you already understand his
position, so to speak, on computer chess, his
meaning was not entirely obfuscated. Sometimes
near-IMs can be very lazy writers; in that posting, the
nearly-IMnnes made very little effort to communicate,
and came off as a computer-basher.


In fact, I used the phrase "You bore me." Far be it for me to
question the English parsing skills[1] of the great help bot[2]


It's okay, so long as you don't try to compare my
parsing to the gold standard of parsing, Parser Blair.
(Doing that makes me look bad, you know, and like,
I don't need your help for that.)


but it is difficult to understand how you could possibly read
"You bore me" and conclude that I was writing that Phil Innes'
posting was boring, not that Phil Innes himself is boring. Who
do you imagine I was addressing? His post?


Idiot. Here is the exact quotation I was referring to,
and which you -- a complete idiot -- have attempted to
cover up:

"Until that day, I have better things to do that to read boring
posts."

I say you are a complete idiot only because it is so
dreadfully obvious that you cannot erase google posts;
they are right there, where anyone can see them. Try
scrolling up and reading what you actually wrote. (With
a mind like yours, it is unwise to ever rely on memory.)


Footnotes:

[1] According to longstanding Usenet convention, any post that
contains a spelling or grammar flame must have at least one
spelling or grammar error. At least I don't think that
"fantascist" is an actual word...


Nonsense; the spelling errors are simply a reflection
of IM Innes' imperfect understanding of English, as his
mother language is Andean.


[2] Correct usage would, I believe be "help 'bot" or "Help 'Bot,"
not "help bot." "Bot" being a contraction of "Robot," it should be
preceded with an apostrophe.


Again, you are out of the loop; the word "bot" is now
an accepted term, by itself. No need to preface it with
an apostrophe, because the term "robot" is generally
understood to mean a mechanical device, whereas
"bot" refers to internet programs, such as shopping
bots, search bots, and so forth. (Picture a huge,
clumsy machine which rants on and on about "danger,
young Will Robinson!", and you have an idea of what
most people think of when they see the term robot. )


-- help bot

"Ralf Callenberg" wrote in message
...
20.08.2007 15:40, Chess One:

If after 20 years of evolution, and fantastically improved chess
programs, the answer to my question is still not known, I rest my case
that any claims for the engine are not based on facts.

I don't get your point. Computers *can* play

That is a hypothetical statement. It is no matter of fact. What I said was
that no program dares play without its book to achieve a nominal rating.
This is not a supposition. If you can refute this with factual material, do
so, otherwise what I wrote above stands.

To answer the issue - can you remember what it was? - the question is, why
have they not? That is the unanswered question. It is a scientific and
commercial one.

---

There are 2 routes to explore in computerizing chees - one is the current
paradigm which is to make emulation of chess as strong as possible, even if
that means the program contravenes the laws of chess. The other is to
understand something about the evaluation paradigm, a completely untended
sector, which nevertheless is the primum mobile of AI. [which is too
complicated for anyone here, since writers cannot even differentiate
emulation from real]

Since programmers have abandoned this second factor in favor of playing
moves the computer could not achieve, it locks itself into the emulation
paradim, and has nothing whatever to do with learning, which is why
computerization of chess is an abandoned subject in AI.

For computer geeks it is all about strength [commercially to make money],
even if that requires cheating the rules of chess so that it is not chess,
and nothing to do with learning. Review your own writing and tell me in
which camp you are in.

At the end, the initial question remains that the computer plays moves it
does not understand. And maybe you will then consider why this is less than
useful in full expert systems and in AI, since, eg, where does it obtain the
base postions it posits, which it cannot even understand?

That is a systemic approach, or in other words, a scientific one. The
process is all.

Phil Innes

starting with move one. They are just weaker without their opening books.
The same would be true for a GM: take away his opening knowledge, and he
would play weaker - otherwise, why would GMs go through the hassle and
invest so much time in the opening preparation?


[1] Why did you post this to rec.games.chess.misc and not to
rec.games.chess.computer?
Because computer geeks don;t understand the game at all, and its hard
to find even an 1800 player in their ranks.
You mean people like the programmer of Rybka, International Master Vasik
Rajlich?

No - indeed he is an exception to the norm - and look what a difference
that makes! Why do you choose some exception as if it were the rule?

Well, let's have a look at some programs from the last Computer World
Championship:

Stefan Meyer-Kahlen, author of Shredder, not active since several years,
but his last German rating was about 1900, which corresponds to an Elo of
roughly 2000. Johannes Zwanzger, author of Loop - Elo 2335. Vincent
Diepeveen, author of Diep - Elo 2290. So, in a field of 12 engines, 4
programmers (Rajlich included) have an Elo safely above 1800. One may add
one of the well known people of the scene, Matthias Feist, co-author of
Fritz, who has a German rating of about 2000 which corresponds to an Elo
of ca. 2100. Rajlich may be by far the strongest, but your claim, that
computer "geeks" are generally weak players simply does not stand a proof.

But lets us not continue with fantasies about things which are testable -
the standing issue is as above.

Just for the record: the above claim about "geeks" and their playing
strength was a fantasy by you, a claim not grounded on anything more than
prejudice.


Every now and again there is some pretence that the fact chess engines
can't play chess from move 1 doesn't even matter )))
When Hydra smashed Adams, it had only a very shallow opening book. It
wasn't move one - but definitely not a book reaching into the
middlegame.

I understand you like to make these rather general claims - as if - you
were answering what I wrote, but we both know that they do not answer the
question, or even pretend to - and computer-geeks always do the same
schtick!

What are you talking about? Your claim is, that computer programs can not
play chess from move one, right? I brought examples of programs who play
against GMs quite successfully although they start playing at move one or
at least quite early. So, why does this not address your point?


Sorry, I don't want to encourage more specualtions - my question is
actually a scientific one, which is a different method of inquiry.


Would you be so nice to repeat your question, please? I am not sure that I
got it.

Greetings,
Ralf

Ralf Callenberg wrote:

Guy Macon http://www.guymacon.com/ wrote:

In my opinion, the basic strategy of using human-to-human games
from the past as the basis for an opening book will become less
and less useful for computers as they improve to the point where
they are much, much better than the best humans.

What you seem to forget is: a move taken from opening theory is not
based on a single human. It is evaluated in an evolutionary process over
many cycles of refinenment. The GMs nowadays use the aid of computers to
find new moves. So it doesn't matter, that computers are so much
stronger than humans - the level of refinement in the opening theory is
still far beyond what a single machine can achieve. Whether you let it
think for 10 minutes or an hour per move, doesn't make much of a
difference. It is still far below the time - of humans plus computers -
which was put into the moves of current opening theory.

I hadn't thought of it that way. The above makes a lot of sense.

The only thing you can get this way is an opening book which might
better fit the program than "normal" theory. But this would mean,
that the created opening book is best suited for the program which
created it, while another machine might not take a lot of advantage
from it.

It's even worse than that. Such a pre-computed opening book is
best suited for the specific version of the program which created
it. If a newer version is stronger or even has a different playing
style, a new book would have to be made.

If a computer that is playing at, say, 2 minutes per move time
control has on its hard disk a complete tree of all possible
moves and responses up to ply N with its moves precalculated
at, say, 20 minutes per move, it will play those first N plies
better by using the book moves.

Did you make an estimate, how much time it costs for an N of, say, 10?

Of course I did. And as expected, it's huge. Clearly there has
to be a lot of pruning done, just as there is when a computer
evaluates positions. Just for fun...

Number of possible positions at ply N:

N=1, 20 positions
N=2, 400 positions
N=3, 8,902 (8 kilopositions (8 thousand))
N=4, 197,281 (197 kilopositions (197 thousand))
N=5, 4,865,617 (4 megapositions (4 million))
N=6, 119,060,679 (119 megapositions (119 million))
N=7, 3,195,913,043 (3 gigapositions (3 billion))
N=8, 84,999,425,906 (84 gigapositions (84 billion))
N=9, 2,439,540,533,153 (2 terapositions (2 trillion))
N=10, 69,353,270,203,366 (69 terapositions (69 trillion))

References:
http://www.research.att.com/~njas/sequences/A006494
http://www.research.att.com/~njas/sequences/A048987

So a hundred trillion positions for ten plies, more or less.

At 10 minutes per position, that would be a quadrillion
(1,000,000,000,000,000) CPU-minutes. Assume 1,000
computers and that's a trillion (1,000,000,000,000)
minutes / 16,666,666,667 hours / 694,444,444 days /
1,901,285 years -- a mere 2 million years. No problem!

BTW, the same sort of no-pruning calculation give estimates
of a hundred hours for Hydra and to look ahead 10 plies and
around 122 days for Rybka to look ahead 10 plies. but of
course both Hydra and Rybka *do* prune rather than doing a
stupid exhaustive search.

All of this takes computing power, but it is computing power
that is used before the start of the game, not computing power
that could be used to play better during the game. It is also
a task that is particularly well suited to being done on a
large LAN during non-working hours or even in with distributed
computing such as is used by SETI@Home and Folding@Home.

Before you go on with your idea, you might to a rough estimate as
suggested above. It might turn out, that even with a lot of computers at
hand, this is not feasible. I don't know it, but it's your idea, you
should give it a rough test up front.

There are some rather obvious pruning strategies available.
One would be to start with the largest conventional opening
book available (still far smaller than a 8-ply exhaustive
search), re-evaluate each move at an hour per move, discard
all places where the human and the computer don't come to the
same conclusion as to what the best next move is, then start
growing the computer-generated book from the resulting tree.

As I said before, all of this assumes a computer that doesn't
suck at openings. In your first paragraph above you make a good
case for that assumption being untrue and remaining untrue.
That's a real problem. The need for pruning isn't.

--
Guy Macon
http://www.guymacon.com/