Announcing forced mates: a poll
Steven J Edwards
should greatly assist with search time requirements for certain
classes of endgame positions. In some cases, even modest "tablebases"
will allow detection of forced mates up to 55 ply long (mate in 28).
(Once the program has latched onto a winning tablebase sequence, it
can play nearly instantaneously.) This brings up an interesting
question when such a situation arises when playing against a human who
is losing:
In a tournament setting, should the program (or its operator) offer to
tell the human opponent that the program has a forced mate?
In favor:
1) The human may feel that he/she has better things to do with his/her
time than to struggle in a known losing situation. Therefore, such a
player should appreciate hearing the news as soon as possible on order
to do other things. These could include preparing for the next round,
getting a snack, leaving for home early, etc. Hours of time could be
saved.
2) If such notification is assured, that lack of any would indirectly
inform the human that putting up a fight in a questionable position
just might be worthwhile. This would help prevent premature
resignation or inapproriate draw offers.
3) The announcement can be done in a discreet fashion, say, by writing
the message on a slip of paper and passing it to the human during the
program's clock time. There's no need for the operator to stand up
and loudly proclaim "Mate in twenty five, thanks for the game, see ya
later!"
Opposed:
1) The Laws clearly state that a player must not be "disturbed" by the
opponent during a game. Perhaps there are some humans who would
interpret advance mate notification as a "disturbance", even when the
intent is to save time and effort.
2) The program could be wrong. There have been some mistakes with
early work in tablebase construction; one problem with a certain KPK
tablebase was not detected until some five years ago in 1989.
However, I think most researchers are quite confident with the current
tablebase situation.
3) The human may feeling unduly pressured into resigning. Even when
the resignation is clearly justified, the pressure could be seen as an
attempt at an unfair psychological attack that could affect the
human's performance in any later games in the event.
4) There is the chance of (near) future extenuating circumstances.
For example, an unforeseen hardware failure with the computer could
cause it to be forfeited, even in a winning position, just because the
program is unable to respond within the prescribed time period.
My preference, if I were the human opponent, is that I would want to
be informed of an impending forced mate. This is because I would like
to save the time and effort of a useless battle. If I happened to
have the extra time available, I could always play out the game, even
all the way to mate, if I wanted.
I would be interested in reading posts by others about their
preferences. I will use the information to determine whether or not
to provide mate announcements in Spector's future tournament practice.
Perhaps other program operators would be interested as well.
-- Steven (s...@world.std.com)
Don Fong
>In a tournament setting, should the program (or its operator) offer to
>tell the human opponent that the program has a forced mate?
option turned on, or off. This yields most of your "In favor" advantages
while nullifying most of your "Opposed" disadvantages.
--- Don Fong
al
s...@world.std.com "Steven J Edwards" writes:
> I am adding some endgame look-up databases to my program Spector that
> should greatly assist with search time requirements for certain
> classes of endgame positions. In some cases, even modest "tablebases"
> will allow detection of forced mates up to 55 ply long (mate in 28).
> (Once the program has latched onto a winning tablebase sequence, it
> can play nearly instantaneously.) This brings up an interesting
> question when such a situation arises when playing against a human who
> is losing:
Stuff deleted
I think the key is that most computer once they have announced/found
a mate, the subsequent moves will be found with increasing rapidity,
and within a couple or so moves down the line the moves will be
instantaneous. In my experience of operating computers against Humans
at tournaments, this is itself is sufficient indicator that one or
other side is heading to mate and if the player is good enough to
realise that it is not themselves who has the win, and they can see
what's coming, they don't need to be told that there is a mate in 5
or whatever. The only exception might be in the case where there is a
mate in 18 announcement (DT a couple of years or so ago), perhaps in
that case the human might not see what is coming for some moves.
On a personal basis, I would prefer not to be told there is a
mate - I would probably regard that as psyching out - might result in
some people getting up for a cup of coffee and a wander around.
Certainly when a computer I have been operating has announced a mate
in X, after the game if the opponent is interested I will tell them.
I regard this as the same as if a human playing me said "Mate
in 5!" - I would think that rather cocky!
Al
Steven Rix
In article <CtArC...@world.std.com>, s...@world.std.com (Steven J Edwards) writes:
->I am adding some endgame look-up databases to my program Spector that
->should greatly assist with search time requirements for certain
->classes of endgame positions. In some cases, even modest "tablebases"
->will allow detection of forced mates up to 55 ply long (mate in 28).
Hmmm, I am not sure I'd be too keen on playing a program which can access
endgame databases. Openings books are just about okay, because they are
subjective to a certain extent and may not be 100% correct. Assessments in
Informator are notoriously dodgy. Endgames analysed to mate in 28 are a
different kettle of fish.
I guess that it'll become obvious that the computer has found a forced
mate (eg in a tricky endgame it starts moving instantly). Maybe Spector
will be programmed to wait a random time between one and two minutes
before moving in order to disguise this knowledge. Apart from the
attraction of a free chess lesson from a computer, I can't think of
any reason to play on in these circumstances. I think that it's fairest
to point out to the opponent that the computer may access databases if
it sees fit, and, given this warning, the opponent MAY ask for
confirmation. This can apply even in situations where there is no forced
mate (R+B v R is hard to defend, particularly against best play!).
--
Steve Rix (ste...@chemeng.ed.ac.uk)
"A morbid, Edinburgh-based Chemical Engineer" - and no misprint!
f...@capa.watson.ibm.com
>or whatever. The only exception might be in the case where there is a
>mate in 18 announcement (DT a couple of years or so ago), perhaps in
>that case the human might not see what is coming for some moves.
DT did not announce mate in 18, or rather we did not announce the mate. The
operator on site was a local player who agreed to help. We forgot to mention
to him that some people, including us, consider it bad taste to announce mate.
On the other hand, we were not the one who had to sit through the moves.
It can be a lot of fun for the spectators when the mate got played out between
computers as it happened in the 1989 World Computer Chess Championship between
DT-1 and Hitech. All of a sudden, both computers were blitzing out long
sequences of moves, and bang, checkmate, game over.
John Hoggatt
|>
|> In article <CtArC...@world.std.com>, s...@world.std.com (Steven J Edwards) writes:
|> ->I am adding some endgame look-up databases to my program Spector that
|> ->should greatly assist with search time requirements for certain
|> ->classes of endgame positions. In some cases, even modest "tablebases"
|> ->will allow detection of forced mates up to 55 ply long (mate in 28).
|>
|> Hmmm, I am not sure I'd be too keen on playing a program which can access
|> endgame databases. Openings books are just about okay, because they are
|> subjective to a certain extent and may not be 100% correct. Assessments in
|> Informator are notoriously dodgy. Endgames analysed to mate in 28 are a
Excuse me if this has been done to death, but how is use of 'tablebases'
different from a human sneaking into the gents' with Smyslov&Levenfish?
Just curious.
-------------
John Hoggatt, FIDE master. --or, kill me.
T. M. Cuffel
John Hoggatt <j4ho...@mccoy.srv.PacBell.COM> wrote:
>
>Excuse me if this has been done to death, but how is use of 'tablebases'
>different from a human sneaking into the gents' with Smyslov&Levenfish?
I guess the question here is whether the computer accessing stored
materials is equivalent to a human acessing written materials.
I have to say no. A human is allow to access material stored in the brain;
a computer shouldn't be penalized for having a better memory.
Also, EVERYTHING the computer does is stored in one way are another. It
may be easy to single out endgame databases, or even opening books, and
claim the computer is cheating, since the computer is looking up
information, and not computing it. But there is a lot of other
information a computer will look up as well. The computer will look
at information stored as the evaluation function, and see that a rook
is worth more than a knight, and that isolated pawns are bad. It
will look at information stored as the search heuristics, and see how
to look ahead and avoid losing material.
If a computer cheats when it looks at stored information, all chess
playing computers cheat.
John Hogatt
the strict identity of *remembering* and *looking up* which s/he proposes is
accepted for computers [though not for humans], then there is no problem with
'tablebases' or anything else.
The second point, if followed up in its implications, could be used to call
it cheating for a human to have been taught chess rather than working it
out for himself. Clearly there's a distinction to be drawn between
learning a rule of thumb [which is all 'isolated pawns are bad', etc., are]
and carrying your reference materials around and accessing them at will.
Back to the first point: Is it being suggested that human memory and computer
memory are the same thing? I suggest that, while they are analogous in many
contexts, it is potentially misleading to use the same word for both sets of
phenomena.
I would not be happy to play against an opponent who could dash into the
library and look up the guaranteed perfect way of playing in some situation,
and then bring back her findings to the chessboard -- especially if I couldn't.
I think that this describes what is actually going on when a computer uses
a tablebase, and that Cuffel's straight equation of tablebase usage with
human memory misses the essential inequity of allowing this type of lookup
in tournament games.
It's always interesting to see the findings to be gleaned from brute force
computation in the endings, but storing all that for later verbatim use
in tournament play against humans seems out of line to me. I also wonder
about its importance in the AI scheme of things. We already know that
computers can store and retrieve and sort data, after all. Where are the
snazzy algorithms?
------------------
John Hoggatt, FIDE master -- Just my opinions, nothing more.
Steven J Edwards
>In article <311gvj$5...@lace.Colorado.EDU>, cuf...@cs.colorado.edu (T. M. Cuffel) writes:
>|> In article <311cib$g...@gw.pacbell.com>,
>|> John Hoggatt <j4ho...@mccoy.srv.PacBell.COM> wrote:
>|> >
>|> >Excuse me if this has been done to death, but how is use of 'tablebases'
>|> >different from a human sneaking into the gents' with Smyslov&Levenfish?
>|>
>|> I guess the question here is whether the computer accessing stored
>|> materials is equivalent to a human acessing written materials.
>I would not be happy to play against an opponent who could dash into the
>library and look up the guaranteed perfect way of playing in some situation,
>and then bring back her findings to the chessboard -- especially if I couldn't.
>I think that this describes what is actually going on when a computer uses
>a tablebase, and that Cuffel's straight equation of tablebase usage with
>human memory misses the essential inequity of allowing this type of lookup
>in tournament games.
Perhaps the feeling of unfairness would be lessened a bit if the one
dashing into the library for a book had in fact authored the book!
(Yes, I know that this is still against the rules.) I find it amusing
that there are those who would accept a program using an opening book
but not an endgame tablebase. Consider: the opening book is almost
always material taken from human play while a tablebase is authored
entirely by the machine. Which is a better indication of the
program's smarts: the material it took from others, or the stuff it
derived on its own?
>It's always interesting to see the findings to be gleaned from brute force
>computation in the endings, but storing all that for later verbatim use
>in tournament play against humans seems out of line to me. I also wonder
>about its importance in the AI scheme of things. We already know that
>computers can store and retrieve and sort data, after all. Where are the
>snazzy algorithms?
I take the behavioralist approach to AI: if a program acts smart, it
is smart. The other interpretations I leave to the philosophers.
Regarding snazzy algorithms, it turns out that once one such becomes
accurately described, well understood, and widely implemented, it is
no longer regarded as snazzy, AI or otherwise.
I avoided using tablebases in my program until now. This was not
because they are hard to do or because they are not valuable, but
because there were other things that I felt were more important and
had higher priority. Never did I feel that they were unfair in any
way. In fact, in the early days of chessplaying programs, skeptical
humans had great amusment criticising the abyssmal level of endgame
play; perhaps now the programs are only making up for past lack of
endgame knowledge.
-- Steven (s...@world.std.com)
David Spencer
In article <31064c$4...@aban.chemeng.ed.ac.uk> ste...@chemeng.ed.ac.uk (Steven Rix) writes:
> Path: retix!uunet!EU.net!uknet!festival!chemeng.ed.ac.uk!steven
> From: ste...@chemeng.ed.ac.uk (Steven Rix)
> Newsgroups: rec.games.chess
> Date: 25 Jul 1994 12:02:36 +0100
> Organization: Edinburgh University Chemical Engineering Department
> Lines: 27
> Distribution: world
> References: <CtArC...@world.std.com>
> NNTP-Posting-Host: aban.chemeng.ed.ac.uk
>
>
> In article <CtArC...@world.std.com>, s...@world.std.com (Steven J Edwards) writes:
> ->I am adding some endgame look-up databases to my program Spector that
> ->should greatly assist with search time requirements for certain
> ->classes of endgame positions. In some cases, even modest "tablebases"
> ->will allow detection of forced mates up to 55 ply long (mate in 28).
>
> Hmmm, I am not sure I'd be too keen on playing a program which can access
> endgame databases. Openings books are just about okay, because they are
> subjective to a certain extent and may not be 100% correct. Assessments in
> Informator are notoriously dodgy. Endgames analysed to mate in 28 are a
> different kettle of fish.
>
> I guess that it'll become obvious that the computer has found a forced
> mate (eg in a tricky endgame it starts moving instantly). Maybe Spector
> will be programmed to wait a random time between one and two minutes
> before moving in order to disguise this knowledge. Apart from the
> attraction of a free chess lesson from a computer, I can't think of
> any reason to play on in these circumstances.
I can think of a big reason - potential bugs in the SW. Every program
has bugs in it - a few years ago I think even the mighty DT had a problem
in a tournament where in some case it inverted the sign of the value of
mate in some branch of the code - and recently there was the posting from
Bob Hyatt about the bug that had been in Cray Blitz for >20(! - is that
right?) years.
Especially if the machine is replying instantly it isn't that hard to
sit around for a few more minutes.
--
+ ------------------------+
| David Spencer |
| david-...@retix.com |
+ ------------------------+
John Hoggatt
Certainly, I don't accept use of an opening book in tournament play
either -- but that's too entrenched to attack, I'm afraid.
[some elided here]
|> I avoided using tablebases in my program until now. This was not
|> because they are hard to do or because they are not valuable, but
|> because there were other things that I felt were more important and
|> had higher priority. Never did I feel that they were unfair in any
|> way. In fact, in the early days of chessplaying programs, skeptical
|> humans had great amusment criticising the abyssmal level of endgame
|> play; perhaps now the programs are only making up for past lack of
|> endgame knowledge.
Well, for hemidecades, athletes did not consider it unfair in any way
to use steroids. Use of tablebases would be fine in postal chess, but
I continue to regard it as improper in OTB. Still, this is an opinion;
it's only an opinion; it costs no-one anything to disagree.
Cheers,
J. Hoggatt, FIDE master --or, kill me.
Steven J Edwards
>Certainly, I don't accept use of an opening book in tournament play
>either -- but that's too entrenched to attack, I'm afraid.
>| s...@world.std.com wrote:
>|> I avoided using tablebases in my program until now. This was not
>|> because they are hard to do or because they are not valuable, but
>|> because there were other things that I felt were more important and
>|> had higher priority. Never did I feel that they were unfair in any
>|> way. In fact, in the early days of chessplaying programs, skeptical
>|> humans had great amusment criticising the abyssmal level of endgame
>|> play; perhaps now the programs are only making up for past lack of
>|> endgame knowledge.
>
>Well, for hemidecades, athletes did not consider it unfair in any way
>to use steroids. Use of tablebases would be fine in postal chess, but
>I continue to regard it as improper in OTB. Still, this is an opinion;
>it's only an opinion; it costs no-one anything to disagree.
I don't think that tablebases would help much at all when suffciently
long (several hours) search time is available, because most programs
have transposition tables. The contents of these are constructed
on-the-run and, given enough time and memory, will serve in place of a
tablebase library. It is pretty amazing to see a program looking
thirty, forty, or even more ply ahead in a locked pawn endgame because
of the transposition table support. I would think a human would have
better chances against a well stocked tablebase library (positions
that may never be reached) than against a similar amount of storage
assigned to a transposition table that gets used every move.
Although the tablebase concept is simple, it is still a significant
accomplishment to construct and implement a practical system from
scratch. That's why a number of chess programmers use Ken Thompson's
CD-ROMs to help out with the effort. But even with outside help,
there is still a bit of work involved. Wouldn't you accept the
analogy that this is similar to an athelete who has spent many hours
exercising and performing other training activities? Shouldn't a
program be rewarded for such effort? Shouldn't a lazy, unprepared
athelete be denied reward for lack of such effort?
My suspicion is that many chess programmers use an opening book mostly
to avoid embarrassment in the opening, and similarly, a tablebase
library is used to avoid embarrassment in the endgame. A secondary
reason for both is to save time on the clock. I doubt that in either
the opening or the endgame is there much visceral joy in winning by
reason of one's program having a bigger memory than the other program.
-- Steven (s...@world.std.com)
Robert Hyatt
once more...
This is your brain.... this is your brain on...
nope, wrong commercial... :^)
In any case, the computer *does not* flip thru a book. it simply has a
"photographic memory".... :^) BTW, let's *NOT* start this debate again...
--
Robert Hyatt Computer and Information Sciences
hy...@cis.uab.edu University of Alabama at Birmingham
(205) 934-2213 115A Campbell Hall, UAB Station
(205) 934-5473 FAX Birmingham, AL 35294-1170
Robert Hyatt
>in tournament play against humans seems out of line to me. I also wonder
>about its importance in the AI scheme of things. We already know that
>computers can store and retrieve and sort data, after all. Where are the
>snazzy algorithms?
>------------------
And just how many times has Karpov, Kasparov, Korchnoi, Spasky, Fischer,
et. al, done just EXACTLY this when preparing some cute opening innovation
for an important game? How is that different? In fact, during some of the
more "interesting" Belle vs Cray Blitz "wars" in the early 80's we were
doing this regularly. Belle (Ken Thompson) insisted on playing the two
knights against our 1. e4 as black, and then played the Na5 variation
allowing us to keep the pawn but having our pieces driven back to our
side of the board with a cramped position. We kept "improving" our
line and finally started winning these. We used our "book" to store
the analysis, just like we store all of our opening moves. QED, the
opening library contains the same things that a human's memory does,
just arguably more of it....
Bob
Robert Hyatt
Seems more likely for such bugs to occur in a program that is "thinking"
as opposed to one looking things up in a database. Bugs should be rare
there....
al
As an arguement for playing on to the bitter end against computers I
am afraid I have a sorry tale to tell! I was one of the operators of
a three "team" of Mephisto Academy's at the British Championships in
the Major open section (Basically under about 2300-2400 - all players
who have not qualified for the main championship down to about 1800
ELO or so). One of our machines ran very short of time on the
external clock due to an operator error (he had entered the wrong
move and in taking it back had screwed up its' timings). Got tight
against a player from UAE (I think) in a position where it was
clearly winning - a rook up or so as far as I can remeber. Any way
goes in to hibernation with a couple of moves to make and flag falls
just as it announces mate in 5!! Sickner. Don't think this would
happen nowadays with clock adjusts etc. so don't hold out too much
hope folks - besides mate in 5 only takes a couple of seconds :-)
Al
Steven Rix
In article <CtKCs...@world.std.com>, s...@world.std.com (Steven J Edwards) writes:
->
->>|> I avoided using tablebases in my program until now. This was not
->>|> because they are hard to do or because they are not valuable, but
->>|> because there were other things that I felt were more important and
->>|> had higher priority. Never did I feel that they were unfair in any
->>|> way. In fact, in the early days of chessplaying programs, skeptical
->>|> humans had great amusment criticising the abyssmal level of endgame
->>|> play; perhaps now the programs are only making up for past lack of
->>|> endgame knowledge.
What exactly are we talking about anyway? Are you saying that in the
course of a normal game, if the position reaches KRB v KR, Spector will
suddenly start using a database and may well announce mate in 30? Or are
you just saying that because of the use of hash tables etc, your program
may well lock into a mate in 28, in a material situation not previously
analysed by Ken Thompson and the endgame database people?
->My suspicion is that many chess programmers use an opening book mostly
->to avoid embarrassment in the opening, and similarly, a tablebase
->library is used to avoid embarrassment in the endgame. A secondary
->reason for both is to save time on the clock. I doubt that in either
->the opening or the endgame is there much visceral joy in winning by
->reason of one's program having a bigger memory than the other program.
I think there's a distiction between belief and certainty. If you take
opening analysis from a random Chess Informator, there will probably
be deliberate misassessments included; if you include everything, then
you can conceivably weaken the program. A bigger book does not necessarily
mean better play. On the other hand, with endgame databases, you can
announce mate in 50+ moves, no problem, cast iron certainty.
Then, there's the "everybody else does, so why can't I?" approach. If in
computer v computer matches, both sides are using databases, there's no
problem. However, it seems unfair to compete against a human, though,
when all the machine is doing is looking up a best move each time in order
to achieve the shortest checkmate.
Steven J Edwards
>In article <CtKCs...@world.std.com>, s...@world.std.com (Steven J Edwards) writes:
>->
>->>|> I avoided using tablebases in my program until now. This was not
>->>|> because they are hard to do or because they are not valuable, but
>->>|> because there were other things that I felt were more important and
>->>|> had higher priority. Never did I feel that they were unfair in any
>->>|> way.
>What exactly are we talking about anyway? Are you saying that in the
>course of a normal game, if the position reaches KRB v KR, Spector will
>suddenly start using a database and may well announce mate in 30? Or are
>you just saying that because of the use of hash tables etc, your program
>may well lock into a mate in 28, in a material situation not previously
>analysed by Ken Thompson and the endgame database people?
Both. Although Spector has/will have the code to handle any arbitrary
tablebase upto (2^31)-1 entries, KRBKR is not on the initial list as
it requires some 167772160 bytes of storage, and I don't have the free
disk space at the moment. But even with a simple KPK + KQK tablebase,
Spector will be able to announce mate in 26 (17 moves to promote in a
favorable KPK postion, then 9 moves to mate from a favorable KQK
position). Transposition tables are built during the game and so
consume time on the clock whereas tablebases are all constructed
before the game. But to build a tablebase to detect and solve any
KPPPPKPPP position is probably outside current technology. Yet such a
position with (almost) totally locked pawns could be handled by a
current technology transposition table because only the specific case
needs solved, not the general one. By the way, there are some
2.4*10^15 KPPPPKPPP positions and over a quadrillion bytes of storage
would be needed for the tablebase; just over fifty bits of addressing.
>I think there's a distiction between belief and certainty. If you take
>opening analysis from a random Chess Informator, there will probably
>be deliberate misassessments included; if you include everything, then
>you can conceivably weaken the program. A bigger book does not necessarily
>mean better play. On the other hand, with endgame databases, you can
>announce mate in 50+ moves, no problem, cast iron certainty.
Perhaps "balsa wood" should be substituted for "cast iron" because of
the large body of anecdotal evidence that many games lost by computer
programs are due to bugs. Sometimes they're opening library bugs,
sometimes they're bugs in the search, sometimes they're bugs in the
hardware. So, there could be (and have been) bugs in the tablebase
and its access subsystem.
>Then, there's the "everybody else does, so why can't I?" approach. If in
>computer v computer matches, both sides are using databases, there's no
>problem. However, it seems unfair to compete against a human, though,
>when all the machine is doing is looking up a best move each time in order
>to achieve the shortest checkmate.
I think if something was unfair in human competition, it would be
unfair in program competition as well. I see nothing wrong in using
tablebases in either. It's just memory.
If it ever got to the point that a game could be solved entirely by
table look-up, then that would kill programmers' interest immediately.
What would be the point with further work? Tablebases just happen to
be a good way to solve an engineering problem in what is a very, very
small portion of the entire chess game tree. By using them, a
researcher is able to concentrate the development effort on the more
interesting parts of the whole problem.
Chess is in no danger of ever being solved by enumeration, at least
not in this universe. But other games are in danger. There are
already several victims like 3-D order four tic tac toe whose solution
has eliminated all further work in the game. Checkers (American
draughts) is in danger of being solved with tablebases by the efforts
of Jonathan Schaeffer and company, and if they achieve this they will
undoubtably be known as the very last checkerplayer program
researchers.
-- Steven (s...@world.std.com)
John Hoggatt
A human's paper library contains the same things that a human's memory does,
but its use during a tournament would constitute illegal use of references.
I would accept keeping whatever Mr. Hyatt wants in RAM, but as soon as
anything on the order of: '23... c4-c3!! [TN] and Black wins' is read out
of a file or ROM during a tournament game, I believe that it is properly to
be considered a transgression.
If a program derived an opening book or endgame tablebase on the fly or,
having accessed such an aid *prior* to a tournament game for use *during*
the tournament game, then kept the aid permanently in RAM *throughout* the
tournament game, as a human must: then I would consider use of that aid
proper. If the program, during a tournament game, reads anything at all
from a file or ROM, I continue to consider it illegal.
--------------------
J. Hoggatt, FIDE master. --Pacific Bell may not agree.
David Spencer
> ...
> If it ever got to the point that a game could be solved entirely by
> table look-up, then that would kill programmers' interest immediately.
Well, then would come the work of a program that tries to find patterns
in the table to teach humans heuristics to play better.
Paul Colley
>If a program derived an opening book or endgame tablebase on the fly or,
>having accessed such an aid *prior* to a tournament game for use *during*
>the tournament game, then kept the aid permanently in RAM *throughout* the
>tournament game, as a human must: then I would consider use of that aid
>proper. If the program, during a tournament game, reads anything at all
>from a file or ROM, I continue to consider it illegal.
There is no clear difference between different forms of computer
memory---particularly not between the different forms of read/write
memory. A long time ago there were computers with only magnetic drum
storage (nothing resembling RAM); today there is virtual memory, cache,
EEPROM, and networks linking huge numbers of computers. I don't see
any natural dividing line along this scale of memory.
Even banning read-only memory is an obviously silly thing to do.
Microcode is just another form of ROM, one that is essential to the
running of most modern computers (the Z80 being one of the last common
non-microcoded CPUs). I have a chess computer in a calculator; the
entire program runs from ROM.
And banning reading from a file? So much for any chess program on a
Unix system. You're also going to have problems distinguishing between
I/O and file storage---mercury delay-line memory was just a glorified
speaker and a microphone, after all.
Hmmm---are you going to require the entire operating system to be kept
in RAM? That'll remove a few more computers from competition.
And the entire point is somewhat moot, anyways. There are computers with
gigabytes of RAM; the table bases you're complaining about would fit into
RAM on some computers without a hiccup. Requiring them to be stored in
RAM just makes life difficult for micro-computers, not necessarily that
bad for the big boys.
Oops, back to the problems with networks. Are parallel chess computers
allowed? If so, I could network together thousands of computers over
the internet just for the sake of their RAM storage. If parallel
computers are banned, there goes Cray Blitz and Deep Thought etc.
And one final complaint:
> then kept the aid permanently in RAM *throughout* the
>tournament game, as a human must
Huh? RAM is a very poor model of how human memory for chess knowledge
works. Files and CD-ROMs are a better model; but those are what you
want to ban!
RAM is more similar to human short-term memory, and it is for sure that
no human can keep even a fraction of the rules of chess in short-term
memory, let alone any chess skill.
- Paul Colley
University: col...@qucis.queensu.ca
Home: paco...@ember.uucp watmath!ember!pacolley +1 613 545 3807
"Anyone who attempts to generate random numbers by deterministic means is,
of course, living in a state of sin." - John von Neumann
John Hoggatt
encounters. The 'athlete jumps into a car' point which many have made does
carry some force. I've never played a computer in a tournament, and perhaps
that is handled differently, but I was once reprimanded for writing suggested
alternate moves and exclams on my scoresheet [I thought that was a little
extreme at the time, true.]. I was barred from storing anything for later
reference, even in that rather silly instance. I don't see why the rules
should favour the nonhuman competitor in this vital area.
|> My suspicion is that many chess programmers use an opening book mostly
|> to avoid embarrassment in the opening, and similarly, a tablebase
|> library is used to avoid embarrassment in the endgame. A secondary
|> reason for both is to save time on the clock. I doubt that in either
|> the opening or the endgame is there much visceral joy in winning by
|> reason of one's program having a bigger memory than the other program.
Perhaps I didn't make myself totally clear. I don't object to use of
opening books or endgame tablebases in games between computers, only in
computer vs. human encounters. A human, after all, is not allowed to write
down his own ruminations for use during the game, nor to read Informants.
I don't see the essential difference here. As computers get stronger I
expect more chessplayers to drift toward this point of view [from the
highest motivations, of course!].
The exact distinction between allowable 'learning' and unallowable
'reference lookups' needs to be refined. For me, the essential question
is whether a playing aid is RAM-resident throughout a given tournament
game. If it is, then I would consider it allowable. If a tablebase or
an opening book is being scanned out of a disk file or some other semi-
offline thingie during a tournament game, I would consider it to be an
illegal use of a reference. If the computer _writes_ anything to a semi-
offline thingie and refers to it during the course of the game, I'd consider
it to be illegal note-taking.
But, heck, I'm just suggesting these criteria. There's a fairness problem
if there are no restrictions on this sort of thing, but it's not worth a
jihad. If Mr. Edwards or anyone else can show a flaw in my preconceptions
about this issue, I'll be happy to switch sides on it today.
And, with that cheery thought --
--------------------------------
J. Hoggatt, FIDE master -- or, kill me.
david carlton
> If in computer v computer matches, both sides are using databases,
> there's no problem. However, it seems unfair to compete against a
> human, though, when all the machine is doing is looking up a best
> move each time in order to achieve the shortest checkmate.
Why? Is it unfair for the human to use human intuition when playing
against a computer? Computers and humans each have different
strengths and weaknesses, but that doesn't mean that it's unfair for
each side to take advantage of their strengths and their opponents'
weaknesses. Especially in this case - if humans were sufficiently
devoted, after all, it would be possible for them to memorize the best
moves for the endgame positions that are likely to come up. It would
take a long time, and be a silly way to try to improve your chess
game, but I don't think that it's out of the realm of possibility -
people are quite good at memorizing data, especially when it's
structured (as in this case). And if a person did that, nobody would
dream of saying that it's unfair for that person to use it.
david carlton
car...@husc.harvard.edu
I'm having fun HITCHHIKING to CINCINNATI or FAR ROCKAWAY!!
J E H Shaw
A computer, in a standard human tournament, should not have
access to any chess information on a hard disk, CD-rom etc.,
once the game has started. A 128Mb monster with ECO in RAM
is acceptable, a 520Kb machine with a 10K book on disk isn't.
It's irrelevant whether the computer has created (say) an endgame
database itself; I wouldn't be allowed to bring in written
notes of my adjournment analysis or anything else.
If it's "easy" to start with the databases in RAM, then do so.
Otherwise, don't use them against humans under tournament conditions.
All IMHO, of course -- Ewart Shaw
--
J.E.H.Shaw, Department of Statistics, | JANET: st...@uk.ac.warwick
University of Warwick, | BITNET: strgh%uk.ac.warwick@UKACRL
Coventry CV4 7AL, U.K. | PHONE: +44 203 523069
An ex-algebraist who lost his ideals, his associates, and finally his identity
John Hoggatt
Neither files nor RAM is a good model of what a human memory does. I
agree that there are serious problems with that classification, which I
regretted post-facto, tho' it seems to have been good flame-bait.
Mr. Coffey has completely blown up my suggested distinction between allowable
'remembering' and illegal 'referring'. I had no huge confidence in it, since
it was at its root based only upon storage speed; unfortunately, I have no
other suggestion to offer. Perhaps it really is a non-distinction for a
computer. However, if 'remembering' and 'referring' really can't be dis-
tinguished, a troublemaker can argue that just about everything a computer
does constitutes illegal use of references, which would mean that there is
no way that a game between a computer and a human could ever be considered
rateable, by a restrictive reading of the rules. I am not comfortable with
that position, frankly.
So what is being said here? That it's just fine for a human to sit in
a game with no access to Informant or even his own personal notebooks while
a computer can just read theory from any old place and has access to the
infallible endgame tablebases (which, while the program may have derived
bits of them, are like as not bought over-the-counter)? Does that honestly
seem equitable?
Do we have to bar computers from playing humans in tournaments at all,
in order to prevent illegal use of references? Or are we stuck with saying
that rules about illegal use of references, by their nature, apply only
against humans?
Perhaps humans should simply be allowed to refer to their own notebooks
and write notes to themselves when playing computers. That, strictly
speaking, wouldn't be legal either, but it would be a stab at maintaining
equities. The whole thing about this matter that bothers me is that we
seem to have rules out there [uses of references, jotting down memory aids,
even touch-move] which are enforceable only against humans.
-----------------
al
j4ho...@mccoy.srv.PacBell.COM "John Hoggatt" writes:
deleted
>
> If a program derived an opening book or endgame tablebase on the fly or,
> having accessed such an aid *prior* to a tournament game for use *during*
> the tournament game, then kept the aid permanently in RAM *throughout* the
> tournament game, as a human must: then I would consider use of that aid
> proper. If the program, during a tournament game, reads anything at all
> from a file or ROM, I continue to consider it illegal.
> --------------------
But all that requires is a computer with a very large amount of RAM -
I don't see the difference in keeping it in RAM, ROM, or hardrive
etc. I think what we have hear is that machines with access to
endgame CDROMS (or hard drives if enough money is thrown at the
problem!) will play certain types of endings perfectly (and the
number of endgames they play like this will grow), and with near
perfect openings, apart from the odd TN and missed line. However
their middlegames will still need to be played ("before the endgame
the Gods have placed the middlegame") and there is still obviously
scope for the human player to win here - humans have intuition or
whatever which the computer program doesn't. I believe that there are
endgame databases in Draughts/checkers and opening books which cover
a far larger total area of the game but still a human is the champ,
or was when I last heard :-)
Personally I think that all openings knowledge is terrible
and that as a moderately lazy chess player I wish all players started
the game having to think rather than reel off 20 moves of theory -
therefore I can't object to computer opening books as I feel the same
about booky chess players. Besides with some computer programs the
only way to bveat them is to let them use their opening books until
they end up in a position that they haven't a clue what they are
doing e.g. a gambit line where the first thing the program wants to
do is get it's pawn back.
Just my 2pence worth
Al
Mark Crowther
: As an arguement for playing on to the bitter end against computers I
I think I was playing in this Championship. What was interesting
was that at the start of the tournament people were allowed to
not play the computer if they didn't want to. [I obviously availed
myself of this] After about 4 rounds the controllers said the event was
becoming potentially hard to pair, so you couldn't refuse to play the
computer. [I told the controller I simply would default if this happened
and still would today, there is a principal involved here for me. I don't
think computers have a place in normal tournament play.] Anyway this isn't
my main point. The UAE team delegation leader at this major open was
Ravikumar, we went down to see the pairings in the evening (after we were
told that we couldn't avoid being paired against the computer) he was
paired against the computer. The following afternoon we found the
strongest UAE player was playing instead ... [This was Plymouth 1989
I believe, incredably hot. A friend of mine actually played and beat the
computer, whereas I kind of lost the thread of the event after losing a
really violent game against someone.] Our recollection (this is through
the haze of time) was that the computer operator went to the toilet and,
well, appeared to have lost a little more time on the clock when he returned
than he might have expected ... Say no more ;-).
--
Mark Crowther - M.D.Cr...@bradford.ac.uk.
Ralf W. Stephan
> But all that requires is a computer with a very large amount of RAM -
> I don't see the difference in keeping it in RAM, ROM, or hardrive
> etc. I think what we have hear is that machines with access to
> endgame CDROMS (or hard drives if enough money is thrown at the
> problem!) will play certain types of endings perfectly (and the
> number of endgames they play like this will grow),
I don't see this, and I would like to ask the original poster
what endgames he will add to the Spector program. Unless he
made a breakthrough, KRPKR will eat dozens of megabytes of disk
space, and this is the only one I would be bothered with.
6 pieces would be in the range 100 MB - 1 GB.
Thanks in advance,
ralf
--
You are in a maze of twisty articles, all alike.
al
M.D.Cr...@bradford.ac.uk "Mark Crowther" writes:
Stuff deleted
> strongest UAE player was playing instead ... [This was Plymouth 1989
> I believe, incredably hot. A friend of mine actually played and beat the
> computer, whereas I kind of lost the thread of the event after losing a
> really violent game against someone.] Our recollection (this is through
> the haze of time) was that the computer operator went to the toilet and,
> well, appeared to have lost a little more time on the clock when he returned
> than he might have expected ... Say no more ;-).
> --
> Mark Crowther - M.D.Cr...@bradford.ac.uk.
>
excat details- thought it was a take back problem - I was
concentrating on my own machine of course!! I can appreciate players
not wanting to play against computers but I have to admit that if it
came down to playing a 2200 person or computer I would choose the
computer! The problem with computers is that people who buy them want
a reasonable idea of how strong they are and the best way of getting
a good approximation is to play against people in serious play - the
USCF grades a few years back were a bit of a joke i.e. too high as
the manufacturerswere able to choose which events their grades could
be based on - grades from 33 games in the British Championships major
open is a true test I feel.
My 2pence worth - don't feel too strongly either way really!!
Al
--
al Is filigree Siberian hamster
al
I did say certain types of endings. Storage space is not an issue
IMHO - with quad density CD's not TOO far away that gives you 2.5 Gb
storage on one CD - who knows where CDs or the equivalent will be in
ten years. Even hard drive sizes have increased dramatically from
10's of megabytes 10 years ago to 1000's of megabytes. IMHO
technology is not the issue when it comes to improving computers to
the top level -CPUs, memory etc. increase quite rapidly without any
imputus from chess - increasing program sophistication is the area
where biggest gains will come - who knows, maybe the algorithm for
playing KRPKR endings will only be 100 bytes long in 2020 :-)
Al
Jeff Mallett
A computer's memory is its hard disk and RAM. It's like a human's memory
as much as it is like a book or a paper and pencil. So, if you allow a
human to memorize moves (now let's see, what comes after 1.e4 c5...) then
you should allow a computer to also start out with moves memorized in its
memory. A computer happens to have a good memory and only forgets due to
events such as hard disk crashes. So what? Human's have other skills
that computers don't have. To say that a computer can't start out with
any opening/ending knowledge other than basic rules would give humans a
huge and underserved advantage (at least at high levels of play).
As to the possibility of making a distinction between RAM and non-RAM, I
think that it may not make too much sense. For one thing, on what grounds
is this reasonable, access time? the fact that RAM goes away when you
shut down the computer? What about static RAM and RAM disks? The code
for my program (Innovation II) is not even loaded into memory all at
once at start up time, is my program now required to make sure that all
code containing chess heuristics is not paged out during the game? For
the record, Innovation's KPK, KRK, and KQK databases are kept in RAM
for the entire game for speed purposes, whereas the opening book is not.
Actually, I see why a human playing a computer might feel disadvantaged
playing a computer in using databases (and yes, Innovation has announced
forced mate in 20-some moves using them in a tournament). My approach
though is not to make a program that competes at a human level, but
rather one that tries to play the best chess possible. If we chess-
programmers didn't use databases our chess programs would be weaker,
just as human postal players would be weaker if they couldn't consult
their books during postal games.
For the things are right now, humans must take solace in the fact that
once both sides are out of book (i.e. the middlegame) the human still
reigns supreme! (given sufficiently long time controls, etc.)
Jeff Mallett
j.ma...@genie.geis.com
"I can, therefore I am." - Simone Weil
Marc ROGER
wrote in rec.games.chess:
[...]
> So what is being said here? That it's just fine for a human to sit in
> a game with no access to Informant or even his own personal notebooks while
> a computer can just read theory from any old place and has access to the
> infallible endgame tablebases (which, while the program may have derived
> bits of them, are like as not bought over-the-counter)? Does that honestly
> seem equitable?
Well, you should never have to play against a computer if doing so
seems unfair for you. So where's the point to argue that computers
have 'illegal' (?) access to databases ? If you don't like it (which
I could perfectly understand and even agree), leave it.
> Do we have to bar computers from playing humans in tournaments at all,
> in order to prevent illegal use of references? Or are we stuck with saying
> that rules about illegal use of references, by their nature, apply only
> against humans?
No, users should be asked beforehand if they accept to play against a
computer (my opinion).
> seem to have rules out there [uses of references, jotting down memory aids,
> even touch-move] which are enforceable only against humans.
Obviously, the game of chess is made to enjoy humans. If playing against
machines does not enjoy you, just don't do it.
Computers are just there for our pleasure, because we like the game and
we think that they are interesting opponents we can learn from.
Allowing them to play only if we could limit their ability to access db
is meaningless (as RAM/ROM/file system distinction is).
> J. Hoggatt, FIDE master --or, kill me.
-- Marc.