On Tue, 05 Dec 2006 13:04:29 GMT, "Chess One"
wrote:


To correct you on 'how we learn' is first to say that you don't know! It is
only people with no insight who cram! What seems to be the process of
/strong players/ or those with any insight, is /not/ cramming positional
memory. Its pattern recognition.

Rote learning of position is dependent on memory with no implicit
understanding of what is remembered.

Pattern recognition is ... not memory!


How does one stock up on patterns ? Don't we, uhhhhh, remember them ?

Chess One wrote:
David Richerby wrote:
Why is it acceptable to tell the program `A rook is worth five
pawns' but not tell it `In the initial position, you should play
1.d4'?

R=5 pawns is an evaluation of a piece

To answer why 1. d4 - well! Why let the program play at all?

Do you actually have anything to say or did the monkey at your
keyboard get lucky?


What is the program doing with its knowledge of R=5 pawns?

Not exchanging its rooks for knights, bishops or pawns? Not
exchanging its queen for rooks?


Dave.

--
David Richerby Transparent Projector (TM): it's like
www.chiark.greenend.org.uk/~davidr/ a 16mm film projector but you can see
right through it!

"Mike Murray" wrote in message
...
On Tue, 05 Dec 2006 13:04:29 GMT, "Chess One"
wrote:


To correct you on 'how we learn' is first to say that you don't know! It
is
only people with no insight who cram! What seems to be the process of
/strong players/ or those with any insight, is /not/ cramming positional
memory. Its pattern recognition.

Rote learning of position is dependent on memory with no implicit
understanding of what is remembered.

Pattern recognition is ... not memory!


How does one stock up on patterns ? Don't we, uhhhhh, remember them ?

A fair question Mike, in fact one with a very strange and diffident
implication.

It is answered by de Groot: that we do not consciously take these patterns
in, but 'for the master player' attain 75,000 of them. In his conclusions he
was unable to state if one thing [in]formed the other [that is, he could not
assert a causative relationship], a point!; if this is memory as distinct
from activating a matrix of recognition that already exists.

Deep waters.

I should personally welcome informed opinion on this, and even from other
non-chessic studies]. (I am familiar with what Gardener has written on these
subjects.)

Cordially, Phil

"David Richerby" wrote in message
...
Chess One wrote:
David Richerby wrote:
Why is it acceptable to tell the program `A rook is worth five
pawns' but not tell it `In the initial position, you should play
1.d4'?

R=5 pawns is an evaluation of a piece

To answer why 1. d4 - well! Why let the program play at all?

Do you actually have anything to say or did the monkey at your
keyboard get lucky?

Anan?

What is the program doing with its knowledge of R=5 pawns?

Not exchanging its rooks for knights, bishops or pawns? Not
exchanging its queen for rooks?

This is not anecdotal exchange of stories, but requires some knowledge of
what is happening with computer and how to talk of it. Did you really not
understand the difference between a computer evaluation and a forced move by
programmer. Maybe your post is a joke -if so, sorry not to have understood
it.

PI




by Transparent Projector (TM): it's like
www.chiark.greenend.org.uk/~davidr/ a 16mm film projector but you can
see
right through it!

"David Richerby" wrote in message
...
Martin Brown wrote:
I agree it is a shame that one of the top engines has not been run
against a GM sans book.

What does `without book' mean. How can you stop me including the
opening book as part of the executable rather than as a separate data
file? Why is it acceptable to tell the program `A rook is worth five
pawns' but not tell it `In the initial position, you should play
1.d4'?

I propose limiting the opening book to under 10 moves to simulate human-like
play conditions. Otherwise one could imagine a GM team creating a super
opening book that virtually assured the machine an advantage with White and
equality or better with Black, with all moves drawn from a table. Even at
top levels players forget moves, mix up move orders, and prefer slightly
inferior positions in which they feel comfortable but which are practically
forced wins for the computer.

Chess One wrote:
David Richerby wrote:
Chess One wrote:
What is the program doing with its knowledge of R=5 pawns?

Not exchanging its rooks for knights, bishops or pawns? Not
exchanging its queen for rooks?

This is not anecdotal exchange of stories, but requires some
knowledge of what is happening with computer and how to talk of it.

Yes, thanks, I have that. Specifically, I have a BA in Computer
Science and have written a chess engine.


Did you really not understand the difference between a computer
evaluation and a forced move by programmer.

Can you not see that a carefully-crafted evaluation function can force
the engine to choose a given move from any specific position?


Maybe your post is a joke -if so, sorry not to have understood

No, I'm being absolutely serious.


Dave.

--
David Richerby Permanent Adult Atom Bomb (TM): it's
www.chiark.greenend.org.uk/~davidr/ like a weapon of mass destruction
that you won't want the children to
see but it'll be there for ever!

Ange1o DePa1ma wrote:
David Richerby wrote:
Martin Brown wrote:
I agree it is a shame that one of the top engines has not been run
against a GM sans book.

What does `without book' mean. How can you stop me including the
opening book as part of the executable rather than as a separate
data file? Why is it acceptable to tell the program `A rook is
worth five pawns' but not tell it `In the initial position, you
should play 1.d4'?

I propose limiting the opening book to under 10 moves

What does `limiting the opening book to under 10 moves' mean? How can
you stop me including the opening book beyond move ten as part of the
executable rather than a separate data file? Why is it acceptable to
tell the program `A rook is worth five pawns' but not tell it `After
1.e4 e5 2.Nf3 Nc6 3.Bb5 a6 4.Ba4 Nf6 5.O-O Be7 6.Re1 b5 7.Bb3 O-O 8.c3
d5 9.exd5 Nxd5 10.Nxe5 Nxe5 11.Rxe5, Black should play 11... c6'?

to simulate human-like play conditions.

How does it `simulate human-like play' conditions? GMs know Marshall
gambit theory well beyond move twenty: why does the computer have to
work it out from scratch?


Dave.

--
David Richerby Generic Atom Bomb (TM): it's like a
www.chiark.greenend.org.uk/~davidr/ weapon of mass destruction but it's
just like all the others!

Ange1o DePa1ma wrote:

"David Richerby" wrote in message
...
Martin Brown wrote:
I agree it is a shame that one of the top engines has not been run
against a GM sans book.

What does `without book' mean. How can you stop me including the
opening book as part of the executable rather than as a separate data
file? Why is it acceptable to tell the program `A rook is worth five
pawns' but not tell it `In the initial position, you should play
1.d4'?

I propose limiting the opening book to under 10 moves to simulate human-like
play conditions. Otherwise one could imagine a GM team creating a super
opening book that virtually assured the machine an advantage with White and
equality or better with Black, with all moves drawn from a table.

That might still be an interesting undertaking, but I suspect it would
have to be done by machine at 10 moves or 20ply the book with an
average branching factor of 4 would require over 4TB of storage. Well
beyond anything that would be humanly checkable, but not beyond a
distributed computational effort. Or did you mean out to only 10ply ?

top levels players forget moves, mix up move orders, and prefer slightly
inferior positions in which they feel comfortable but which are practically
forced wins for the computer.

I'm not sure if I would put it that strongly. Kramnik has demostrated
considerable finesse in holding Fritz10 to draws in the match. He
clearly has a better temperment suited to playing against a machine
than Kasparov. The win with the Najdorf was very classy. And his only
loss in this match was careless but did not affect his play in later
games (impressive).

Regards,
Martin Brown

"David Richerby" wrote in message
...
Chess One wrote:
David Richerby wrote:
Chess One wrote:
What is the program doing with its knowledge of R=5 pawns?

Not exchanging its rooks for knights, bishops or pawns? Not
exchanging its queen for rooks?

This is not anecdotal exchange of stories, but requires some
knowledge of what is happening with computer and how to talk of it.

Yes, thanks, I have that. Specifically, I have a BA in Computer
Science and have written a chess engine.

I may have preceded you into the field in the time when there was no
monitor - and only punched cards and streaming tape, and you had to write
the [lol] 'operating system' integral to your program, or task, as we called
it. Although this was not in the very early days of computation, and vacuum
tubes were then on the way out, eg

What is difficult, David, and I welcome a conversation on this subject which
can occur on the right basis, rather than a perceived basis of others - is
that so often terms are transposed into virtual terms, then the claim is
made that the virtual event mirrors human [or other non-virtual] activity -
which is often true! But what is mirrored is a missaprehension. Often a
false, or very poor, appreciation of the non-virtual event.

And, sad to say, insufficient attention is paid to the human side of this
mirroring, both of what is mirrored, and also that it is in fact human being
who conduct the mirroring, together with all their knowledge [and false
knowledge!] of the world.

Did you really not understand the difference between a computer
evaluation and a forced move by programmer.

Can you not see that a carefully-crafted evaluation function can force
the engine to choose a given move from any specific position?

I see that that is no exact answer, but a paradigm shift by way of emphasis.
In previous conversations with computer-folk they are content to discuss the
/strength/ of their engine in emulation paradigms.

I wish to contrast this with computational evaluation in a /learning/
paradigm.

Maybe your post is a joke -if so, sorry not to have understood

No, I'm being absolutely serious.

You are quite right in what you say, except that it does not really respond
to my point, which is missaprehended, and rested on pattern-recognition,
rather than positional memorization to emphasise strength.

And here we have the human dimension of the programmer who dutifully
reflects the activitiy of most chess players, who cram positions, especially
in the opening. But the very interesting that de Groot revealed about chess
players is that 'master-level' [by which he means masters and upward] do
not!

Their primary source, regardless of what opening memorizations they attain,
is not cramming positions. Should some chessic authority be needed to
substantiate this point, I believe Kasparov has written much about it.

The traditional Russian School of chess uses pattern-recognition
publications as their primary encouragement for players. Example: thousands
of diagrams upon various pattern-recognition themes, so that the knight fork
can be seen operating in a 2-move sequence, or in an 8-move sequence! By
spotting the pattern then you increase your awareness of all such patterns,
and miraculously can look deeper into the game. Add a prophylactic 9th move,
then another /thema/ is possible, this time 'reversing the pin' on your
Queen by a rook, since the 8-move knight-fork has engineered an
intermezzo-move with the pinned piece which then allows it to move in a
forcing way, and the Q can capture the rook.

Then you have 'seen' 17 moves. Perhaps you can add a few other moves so that
after QxR you are assured you will not lose your shirt, and now have a
comfortable material advantage.

But absolutely none of that has to do with positional memorization. It is
not as if the patterns are memorised at all, but revealed as possibilities
by virtue of practice. And this is a huge distinction between /levels/ of
play from that of the rote learner, as well as the /mode/ of play of the
'master'.

De Groot did not even conclude that rote-learning can produce master-level
play, and can even inhibit it. Not a factor much pointed out by
book-publishers or chess authors. I think it may or may not do any harm to
specific players - and is worth further study.

But to return to chess programmers - almost universally they adopt the
strategm employed by rote learners. Chess was once considered a useful means
to understand how machines could learn by emulating people, so that we could
then understand people's processes. When the target is shifted to
non-learning, and 'merely' winning, then chess ceases to be interesting
since nothing is revealed when emulation of play is combined and confounded
with pattern evaluation.

About 10 years ago, [maybe you are familiar with this?] an interesting and
cheap device was deployed in all sorts of consumer electrical products, and
by the generic name of 'field-gates'. Essentially a programmable chip which
recorded faults in new products, but the faults themselves could not be
anticipated by the programmer since they were unknown and massively-various,
and one chip of as little as 100 cells could not contain enough information
to pre-record anticipated errors. Bigger chips were also too costly.

These chips were essentially able to self-organise and determine value.
Latterly, these self-evaluating programmable hardware devices were
themselves emulated, a process called 'exigesic emulation', so that
streaming data of unknown providence and value could be assessed by the
emulated cell-group itself.

The level of development between say, an implanted chip in an electric
toaster, or telephone, compared to a chess playing computer replete with
program was vast, and yet the simple device was capable of making its own
value determinations - I say again - its own determination of the worth of
the data - and so could, if we transpose it to chess, not need the
programmer to tell it that the bishop is now worth a pre-programmed 3.2
points because certain favorable conditions are present on the board, but it
could make that determination of value itself, and perhaps even disagree, in
that it provides a value of 3.05, then 3.15, then... in real-time response
to actual changing conditions on the board.

While this might be the work of computing in the C22nd, its value can be
assessed now - since it is not dependent on any [no matter how good!]
pre-programmed value, and takes each position as it comes to constantly
re-evaluate its worth.

I note, [laugh], and just in passing, that Fischer is famous for evaluating
his own bishops at 3.25 pawns. And even if the whole chess world wanted to
deal with the awkward decimals involved, because we thought there was some
justice in Fischer's sense of his bishops, this would still not provide the
same evaluation sense as real-time evaluation by the engine itself.

And that real-time 'Intelligence' is exactly what the AI community hoped
would be explored in chess. Instead we have such entities as Deep-Blue from
which absolutely nothing was learned, as as World Computer Chess Champion
programmer, Crafty's inventor, Bob Hyatt said to me, from which nothing was
learned.

These are two possible development routes for chess, and while one optimises
performance in emulation mode, which could be characterised by the world
'strength', nothing at all is larned from this, and in fact by
commercialisation of chess-playing products we ensure it does not. The other
route provides far more modest results of performance, yet yields real-time
appreciations, sometimes reflecting our own missaprehensions about what is
going on, but in any case, enables us to understand machine /learning/ as a
process, rather than machine /result./

Perhaps an interesting idea would be to send this note to Rybka, and ask its
programmer for a response?

Cordially, Phil Innes

Dave.

--
David Richerby Permanent Adult Atom Bomb (TM):
it's
www.chiark.greenend.org.uk/~davidr/ like a weapon of mass
destruction
that you won't want the children
to
see but it'll be there for ever!

"David Richerby" wrote in message
...
Ange1o DePa1ma wrote:
David Richerby wrote:
Martin Brown wrote:
I agree it is a shame that one of the top engines has not been run
against a GM sans book.

What does `without book' mean. How can you stop me including the
opening book as part of the executable rather than as a separate
data file?

Why would you want to cheat seems like a better question. The unfortunate
joke about computer strength is that the computer cheats in order to attain
it by not playing chess according to the rules. Period.

Look-ups are not okay.

Why is it acceptable to tell the program `A rook is
worth five pawns' but not tell it `In the initial position, you
should play 1.d4'?

I propose limiting the opening book to under 10 moves

What does `limiting the opening book to under 10 moves' mean? How can
you stop me including the opening book beyond move ten as part of the
executable rather than a separate data file? Why is it acceptable to
tell the program `A rook is worth five pawns'

Because that is a 'rule' of chess?

but not tell it `After
1.e4 e5 2.Nf3 Nc6 3.Bb5 a6 4.Ba4 Nf6 5.O-O Be7 6.Re1 b5 7.Bb3 O-O 8.c3
d5 9.exd5 Nxd5 10.Nxe5 Nxe5 11.Rxe5, Black should play 11... c6'?

to simulate human-like play conditions.

How does it `simulate human-like play' conditions? GMs know Marshall
gambit theory well beyond move twenty: why does the computer have to
work it out from scratch?

It is a false paradigm to compare computer :: human play. Both parties are
subject to the Rules of chess, which is the correct paradigm against which
to compare anything.

Then what is fair or unfair according to each participant is a variable
which makes no difference tot he laws of the game, and if computers are
truly superior, then so be it! All else is evasion of the initial point at
the top of the message, that the emulated strength of the engine is achieved
at the expense of breaking the rules, which is to say that it is not playing
legal chess at all.

If you want to make a level playing field on the basis of 'fairness',
specifically to more nearly equate opening knowledge, then the player would
also need to have access to 'look-up' material. This already has a name, and
is called [Kasparov's term?] Advanced Chess.

Possibly time of access to information would also need to be addressed so
that the human player would not suffer from the fact that it takes time to
consult their own database? But that is a detail.

As you note, when 2 human players engage in Advanced Chess, the games are
not ratable, and in fact there exists no means of integrating that play with
the general pool of ratings.

This is an interesting chess variant, yet it needs to be called Advanced
Chess, rather than confounded with other forms, and with the therefore
risible claims for performance at Chess - which it pointedly is not playing.

It is rare to find programmers who will admit the fundamental honesty of
this logical proposition - and more common to encounter an attitude of
resentment that the distinction is pointed out, and to be characterised as
'anti-computer', which is an untrue extension of the proposition.

When I play Rybka I want a strong emulation engine, as strong as possible,
and strong by whatever means. But when I play legal chess I don't wish for
the same things, and would be discontent if opponent played 'by whatever
means' since that does not recognise the limits of the game itself which is
defined by its rules, as is what we call 'cheating.'

Phil Innes


Dave.

--
David Richerby Generic Atom Bomb (TM): it's
like a
www.chiark.greenend.org.uk/~davidr/ weapon of mass destruction but
it's
just like all the others!