The problem of opening theory in computer chess

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
MatsW

The problem of opening theory in computer chess

Post by MatsW » Wed May 20, 2009 5:36 am

The problem of opening theory in computer chess. What's the point in
measuring the opening book knowledge when arranging matches between
chess programs? By example, Deep Blue never got out of the opening
book before it was practically over in the final game against
Kasparov. Would the chess public stand up and applause if a human
player sat with an opening book in his lap, replicating the moves in
the book, and winning against the world champion? Of course not, they
would boo at him. I realize, of course, that it is a difficult
question, not the least because human players can play from memory as
well, although not nearly as good. Nevertheless, it is high time that
this practice is questioned in computer chess, at least in matches
between chess programs.

I have a suggestion that could measure also the opening understanding
in chess programs. Today, a program needs, foremostly, to have general
middlegame and endgame knowledge about stratagems, while the opening
can be dealt with by cheating, namely by reading directly from the
book. My proposal, which implies that the standard array can be
rearranged, does not involve randomization, but the players must take
decisions from the beginning, using a method of relocation.

The various relocation methods allow the players optionally to
relocate king and/or queen before the play begins, whilst retaining
the castling rights. The players can abstain from this if they prefer
the standard setup. It is a cogent method of rearranging the initial
position to enhance opening ramification, while allowing the players
to remain in control. The resultant positions deviate marginally from
the standard position and would be experienced as natural by most
chessplayers.

The rules are like standard chess except that the players can, before
play begins, swap places of the king + queen and another piece except
the rooks. Thus, when the king is swapped ('relocated'), the other
piece (the 'relocatee') ends up on the king's square. When the queen
is swapped, the relocatee ends up on the queen's square. One
restriction is that the bishops mustn't end up on the same square
colour, and the king cannot become a relocatee (i.e. swapped by the
queen).

The following article (see link below) contains diagrams and links to
email presets and programs that can play the variants. Unlike
Chess960, these variants do not exclude theoretical study. The players
themselves, independently, decide which setup they want to play. They
could specialize in a certain setup and study it at home. They would
work even harder with their databases. But they could choose the
standard setup, too. It's not the question that variants such as these
threaten people's theoretical work. These variants should be seen as
alternatives. They could be used in chess training, and in rapid chess
tournaments. I have studied much theory, too. I am not allergic to it.
It's just that I have become sceptical about the longevity of chess in
view of today's computerization. It seems that it has narrowed down
theory, in a sense. I give this example in my article, which provides
a good example of today's dilemma:

The best opening move is 1.e4. Suppose that Black answers
1...e5. Obviously, there is a lack of strong alternatives to
2.Nf3 Nc6 3.Bb5 a6 4.Ba4, etc. etc. So there are no good
substitutes for the long variations in Ruy Lopez, while the
alternatives are inferior. Italian game (Giuoco Piano) isn't really
dangerous anymore, and one can't put one's trust in the King's gambit.
The Ponziani doesn't promise much.

It's becoming less and less worthwhile to play inferior variants, due
to opening preparations. If only there existed an equally good opening
system against 1...e5, it would be a great relief. Due to a
slight shortfall of variance in chess, it leads to the necessity of
analysing and playing professional opening lines to the umpteenth
move. There is nothing essentially wrong in such a scientific
approach, but it is slightly frustrating from a creative point of
view.

Of course, 1...e5 is not the only serious defence, Aljechin is
a superb alternative, as well as the Sicilian. I only wanted to shed
light on a frustrating quality of the game, i.e. that after
1...e5 you don't really have the same freedom of moves as you
had in the beginning of the 20th century. Theory is narrowing down
because it's not worthwhile to choose the King's gambit when playing
against a prepared opponent. Comparatively, the feeling of freedom
must have been unimaginable when the American trapper, in the 18th
century, moved into Western territory. Today, people travel along the
highways, just as in modern chess.

In my proposed "relocation variants" theoretical studies remain a
meaningful occupation, although their impact is reduced. This is a
major departure from Chess960. The foremost difference is that the
player can himself decide the setup of his own pieces, whereas
Chess960 is wholly randomized. Additionally, the positions are
"natural" in appearance. Only in one variant (Placement Chess), are
all positions mirrored. Unlike randomized Chess960, the relocation
idea would be acceptable to most chessplayers. See link: (in English
and German)

http://home7.swipnet.se/~w-73784/chess/ ... riants.htm

Mats

User avatar
Bo Persson
Posts: 174
Joined: Sat Mar 11, 2006 7:31 am
Location: Malmö, Sweden
Full name: Bo Persson

Re: The problem of opening theory in computer chess

Post by Bo Persson » Wed May 20, 2009 1:15 pm

MatsW wrote:The problem of opening theory in computer chess. What's the point in
measuring the opening book knowledge when arranging matches between
chess programs? By example, Deep Blue never got out of the opening
book before it was practically over in the final game against
Kasparov. Would the chess public stand up and applause if a human
player sat with an opening book in his lap, replicating the moves in
the book, and winning against the world champion? Of course not, they
would boo at him. I realize, of course, that it is a difficult
question, not the least because human players can play from memory as
well, although not nearly as good. Nevertheless, it is high time that
this practice is questioned in computer chess, at least in matches
between chess programs.
This isn't so. Have you ever seen a physical book present at a computer game? :-)

The fact that the computer can read the openings and remember them almost perfectly, is just one of the differences between humans and machines. The computers also calculate a lot faster. Isn't that unfair as well?
MatsW wrote: I have a suggestion that could measure also the opening understanding
in chess programs. Today, a program needs, foremostly, to have general
middlegame and endgame knowledge about stratagems, while the opening
can be dealt with by cheating, namely by reading directly from the
book. My proposal, which implies that the standard array can be
rearranged, does not involve randomization, but the players must take
decisions from the beginning, using a method of relocation.
That would just create another game, not measure any "understanding" of the current chess game.

MatsW

Re: The problem of opening theory in computer chess

Post by MatsW » Wed May 20, 2009 4:44 pm

Bo, you don't even try to address the issues I discuss. Just because
two pieces have different positions, this doesn't create a totally
different game than chess. But a chess program would have to delineate
a strategical plan that takes account of the problem of the
initiative. Today, a software team must have recourse to an opening
expert, otherwise their program can never achieve fine results against
the highest ranked programs. It is dependent on very sophisticated
theoretical knowledge.

As my method greatly increases the variance, it represents an answer
to the problem of theoretization of chess. In the future a chess
software can have recourse to a super-opening book that covers all the
best lines, say 20 moves in each line, so that the remaining game
becomes more of less a matter of technique. One would be playing
against an opening database for the first 20 moves. If this were the
case, i.e., that regardless of opening line, the program would play 20
moves without thinking, would you still regard computer chess a
meaningful occupation?
/Mats

Marc Lacrosse
Posts: 511
Joined: Wed Mar 08, 2006 9:05 pm
Contact:

Re: The problem of opening theory in computer chess

Post by Marc Lacrosse » Wed May 20, 2009 8:34 pm

MatsW wrote:the problem of theoretization of chess.
Theoretization of chess is not a problem.
It is a chance , an opportunity for new horizons, for endlessly renewed subtlelties.
MatsW wrote: In the future a chess
software can have recourse to a super-opening book that covers all the
best lines, say 20 moves in each line, so that the remaining game
becomes more of less a matter of technique. One would be playing
against an opening database for the first 20 moves. If this were thecase, i.e., that regardless of opening line, the program would play 20
moves without thinking, would you still regard computer chess a
meaningful occupation?
/Mats
Your former post was already showing complete ignorance of chess opening theory and flawed appreciation of the significance of this theory.
But now you go proudly one step further.

Not only is a 20-moves opening book not a super-book as you say but most lower grade master human players have personal repertoires extending far over-there.
And competition books of most top-rated programs extend sometimes to more than 40 or even 50 moves in some lines.

But this does not in any way limit the power of imagination, innovation and refinement in the opening phase of both computer and human chess.
Everyday there are new paths explored or older ones ressuscitated with renewed ideas.

In fact there is nothing like what you seem to call theory in some fixed dried state.

When Kasparov goes for the Scotch in a world championship, when Karpov switches to the Grunfeld, when Topalov plays the Benoni, when Morozevich goes for the Albin, they all go against all established theory of their time and win at the highest level.

And this is no old story and continues to be true today although all of them train with our favorite programs and know "the theory" much better than you (and me).

Moreover, in the field of computer chess an important point is always ignored.

Most seem to prefer the more neutral way of playing chess and evaluating players : an
engine is strong if it demonstrates it is stronger than other engines when having to play all kind of positions. and so some organisations play with varied shortened standard opening books while other ones choose fixed positions coming from a large variety of openings.

Similar is your proposition that is nothing more than a new kind of fisher-random one going inthis same old pathy that tends to favor the ability of an engine or player to play equally well in any kind of legal chess position.

This is a completely skewed way of evaluating chess engines.

Strength at chess is the ability to win more against anybody than you will lose, starting from the same old position that we all know and choosing the path of your liking whatever it could be.

No one ever said that a human chess championship should start from a selected series of fixed openings (one with the =Ruy, one with the Naj) and so on.
There are famous world class players who never played 1.e4 or never played 1.d4.

One day we could perfectly have an engine who is the best in the world because it has found a way to always win playing 1.f4 while its result are poor with 1.e4 or 1.d4. As it may choose to allways play 1.f4 it will win every championship.

And here I come back to the interest of book-making for a given engine.

I can say you this : it is extremely interesting to try to build an opening repertoire for an engine. Not an arbitrary selection of the openings that you like from an human point-of-view. No .
The real selection of playable openings that will best suit en engine "style" and will help it to win more games.

This is a very sophisticated and fascinating process.

I fear that with your " problem of opening theory in computer chess" you are very far of understanding this kind of things ...

Marc

PS I had a good moment of very good laughing when i read that for you it is evident that after 1.e4 e5 2.Nf3 nothing else than 2..Nc6 Bb5 a6 3.Ba4 is playable.
A very good moment.

MatsW

Re: The problem of opening theory in computer chess

Post by MatsW » Thu May 21, 2009 5:30 am

No, no, it is far beyond the capability of the human brain to hold
20-moves lines in memory, for every relevant line. I know lines in the
Aljechin that are 24 moves long (i.e. 48 half-moves), but to know tens
of thousands of such long lines is simply unthinkable. But it's not a
problem for the computer.

Of course, theory is not in some "fixed dried state", but my argument
is that it is continually narrowing down. In the beginning of the 20th
century the practicable opening tree constituted of an enormous
shrubbery with short branches in every possible direction. Nowadays,
practicable opening theory, among professionals, is like a long tall
and narrow tree, with few but very long branches.

One obvious corollary is that 1.e4 is more and more standing out as
the only alternative among super grandmasters with aspirations. Had
Kramnik played 1.e4, then he would still be world champion. It was a
mistake of him, in his youth, to choose 1.d4. The field of theory is
too narrow, so 1.d4 is in today's computerized age easier to defend
against than 1.e4. 1.d4 promises no advantage anymore.

Kasparov went from 1.e4 to 1.d4. After a time he realized his mistake
and went back to 1.e4. This movement from e4 to d4, and back, took its
toll on him, and was probably part of the reason for his burnout and
decision to leave chess. He realized he could not stay with 1.d4
because it is much more narrow than 1.e4, and a prepared opponent thus
has an easier time.

You argue that "...One day we could perfectly have an engine who is
the best in the world because it has found a way to always win playing
1.f4 while its result are poor with 1.e4 or 1.d4. As it may choose to
allways play 1.f4 it will win every championship...


But this can never happen because a majority of the resultant
positions in the middlegame, after 1.f4, are advantageous to black.
At most white can achieve equality with Bird's opening. It is of no
consequence what "playing style" the white program has. The programs
playing black will simply capitalize on the weaknesses in white's
position. In computer chess, it will always generate a statistical
plus for black.

It is not a "very sophisticated and fascinating process" to develop
opening books for computers, it is a waste of life and it does not
contribute to the science of algorithms.

I had hoped that people would read and discuss my idas about
relocation rather than engage in diffuse rhetorics about the nature of
opening play. I regard relocation as a viable alternative to
the standard position, not as its successor.

http://home7.swipnet.se/~w-73784/chess/ ... riants.htm
/Mats

User avatar
hgm
Posts: 23785
Joined: Fri Mar 10, 2006 9:06 am
Location: Amsterdam
Full name: H G Muller
Contact:

Re: The problem of opening theory in computer chess

Post by hgm » Thu May 21, 2009 6:29 am

The Mad Queen's variant has more important shortcomings than over-analysis of its opening array.

A more natural solution for those that want to play a more exciting and tactical game with a smaller draw rate is to simply switch to another variant, sch as Gothic Chess. Then you kill two birds wth one stone: opening theory is hardly advanced for such variants, and offers much more opporunity for innovation at lower depth into the game.

Terry McCracken
Posts: 15844
Joined: Wed Aug 01, 2007 2:16 am
Location: Canada

Re: The problem of opening theory in computer chess

Post by Terry McCracken » Thu May 21, 2009 6:50 am

hgm wrote:The Mad Queen's variant has more important shortcomings than over-analysis of its opening array.

A more natural solution for those that want to play a more exciting and tactical game with a smaller draw rate is to simply switch to another variant, sch as Gothic Chess. Then you kill two birds wth one stone: opening theory is hardly advanced for such variants, and offers much more opporunity for innovation at lower depth into the game.
What's wrong with 3-D Chess if you want a real tough game?

I think I should reinvent the game instead of all these variants. :wink:

User avatar
hgm
Posts: 23785
Joined: Fri Mar 10, 2006 9:06 am
Location: Amsterdam
Full name: H G Muller
Contact:

Re: The problem of opening theory in computer chess

Post by hgm » Thu May 21, 2009 7:57 am

Wich 3D variant did you have in mind? There are many.

The problem with 3D Chess on a more or less cubic board is that the number of cells increases very quikly with size, and with a resonble filling fraction that would make the game duration very long. 5x5x5 is aready 125 cells, and yet the maximum distance is so small that there hardly is any difference between sliders and short-range leapers.

Of course there are also '2.5D' variants, that merely stack 'stories' of a reasonable-sizeboard. That also allows for more easy equipment (e.g. a glass level mounted above a normal Chess board.

MatsW

Re: The problem of opening theory in computer chess

Post by MatsW » Thu May 21, 2009 8:16 am

Gothic Chess, and all the other Capablanca Chess variants, are all
extremely tactical games in that there are several super-knights on
the board. There are double-threats and forks everywhere. So these
variants are more "mad" than Mad Queen Chess (orthochess).

The tactical madness makes those variants less attractive to the
general population, because most people will lose in about seven
moves. Comparatively, in medieval times chess was very popular, also
among women. But when the mad queen appeared on the board, together
with the long-shooting bishops, the game soon lost in popularity. But
the experts became even more enchanted with the game.

So I think we must conclude that the Capablanca variants, including
Gothic Chess, will never become popular. They cannot compete with
orthochess. By an increased tactical complexity we cannot achieve a
popular variant. But to increase opening ramification in orthochess is
something necessary for the future. Comparatively, Anglo-Saxon
checkers has been solved by the computer. Likewise, many opening
systems in chess have today been solved. /Mats

Uri Blass
Posts: 8608
Joined: Wed Mar 08, 2006 11:37 pm
Location: Tel-Aviv Israel

Re: The problem of opening theory in computer chess

Post by Uri Blass » Thu May 21, 2009 9:18 am

I think that the the best moves are simply unknown.
It is not clear that unpopular lines are inferior.

It is not clear that 1.e4 e5 2.Nc3 is inferior relative to 1.e4 e5 2.Nf3

Theory is basically relatively unimportant in computer chess and my opinion is that if you remove book from engines they will usually lose less than 100 elo.

Edit:The fact that 1.e4 is the most popular move of super grandmaster prove nothing because super grandmasters are patzers relative to correspondence players that use rybka to help them and in part of the cases the winners of correspondence games at the high level use 1.d4

Edit 2:Here is one example from a tournament that I play when a GM with a correspondence ICCF rating above 2600(that is equivalent to more than 3000 in normal chess) lost against 1.d4

http://www.iccf-webchess.com/MakeAMove.aspx?id=190512

Uri

Post Reply