Mat Plus Review
Winter 2008
POLEMIC IN THE STUDIES WORLD (1)
For several years now publicly available 6-man EGTBs have presented the studies world, and
especially composers and judges, with a moral problem: should such positions be treated in a
special way, and, if so, why and how? Neither FIDE nor any magazine (not even EG) has
conducted a debate, a debate which is needed today more than ever. Many individual views have
been expressed in many places, with little agreement and sometimes not without heat.
After published (in its no.44 in April 2008) an article by the highly respected St
Petersburg composer Aleksei Sochnev, editor Yakov Rossomakho invited John Roycroft to
respond. John has done so. However, the St Petersburg quarterly is in Russian only, and to bear
fruit debate has to be conducted in more than one language. There is currently only one magazine
in a position to do this, namely Mat Plus. With Yakovs full approval we publish below John's
response in both English and Russian, which we hope will reach a wide audience.
We invite considered and well-argued reactions from our readers.
May I offer the wholly admirable magazine
ZiE a brief reaction to the lengthy article
. ? by
Aleksei Sochnev, one of the leading St
Petersburg study composers and my good
friend for many years?
.
?,
-
This reaction is by someone who, though far
from prolific, is qualfied in composing, spent
26 years in the employ of IBM(UK), and who,
though never a programmer, has had close
association with what he calls oracle endgame
databases going back to the 1970s (and
therefore antedating EGTBs by at least two
decades). He is the author of the very relevant
paper Expert Against Oracle included in Vol.XI
(published in 1988) of the Machine Intelligence
series under the overall editorship of the late
Donald Michie (19232007) who, besides
being an eminent academic with the reputation
of having founded work on artificial
intelligence in Britain, performed a critical role
years before at the renowned secret wartime
codebreaking establishment Bletchley Park,
and in 1985 was Visiting Lecturer to the USSR
Academy of Sciences.
, ,
26 IBM
().
, , -,
,
, 1970-
( , , ,
, EGTB).
,
XI ( 1988 )
Donald Michie (1923 2007),
,
,
Bletchley Park, 1985
.
The present writer has never entered a tourney
with a position derived from an EGTB or
equivalent. He therefore claims to have no bias,
either for or against EGTBs.
, EGTB
. ,
EGTB.
211
Winter 2008
Mat Plus Review
The foregoing unwieldy preamble is I suggest,
important for what follows: credentials, and
declaring an interest could well be a critical
factor in making progress in our still wonderful
studies world.
To ease subsequent discussion I shall use
numbering.
,
,
:
.
.
1.1. Studies are distinguished from problems by
all three results (win for White, loss for White,
and a draw) being relevant. The literature and
development of endgame theory deeply
concern studies specialists but not problemists.
Therefore problems, and how problemists use
computers, may be excluded from the present
debate.
1.1
(
, , ).
, .
, ,
,
1.2 Just as important is the distinction between
two types of chess computer. Chess-playing
programs such as the formidable FRITZ rely on
an evaluation function to determine when to
stop exploration of the tree. On the other hand
an EGTB needs, and has, no such feature
because it simply looks up the pre-computed,
guaranteed infallible, answer(s). See 10.2
below. We are concerned here solely with the
second type of chess computer, the EGTB (or
oracle, a term borrowed from artificial
intelligence).
1.2
FRITZ,
, ,
.
EGTB
,
,
. .10.2 .
, EGTB
( , ,
).
2. Footnote 3 to the FIDE PCCC Codex begins:
The use of a computer does not result in an
authorship of the computer. With respect to
studies and the ongoing development of EGTBs
this sentence needs expanding, as a minimum
by inserting necessarily before result.
2. 3 FIDE
(PCCC) :
.
EGTB,
,
.
3. Discussion is not helped by failure to
distinguish the five different principal roles,
each of which is important, namely: composer;
solver; judge; magazine editor; and the FIDE
Albums (titles!).
4. Since I am no longer involved as composer -for once this is an advantage! -- but chiefly as a
judge, I declare this to be my standpoint.
3.
,
: ,
, , ,
FIDE (!).
4.
!
, ,
.
212
Mat Plus Review
Winter 2008
5. I claim to be an impartial observer able to
invoke the standpoint of ethics.
5. ,
,
.
6. My view, which Aleksei may have
misrepresented (p24), despite it being set out in
EG (see below) is, I hope, clear enough.
6. , ,
, (.24),
, EG
(. ), , .
7. My position as a judge, based on the belief
that all positions of sufficient interest should be
published, is this, cited from EG166:
8. A pure database study will:
be ranked, not honoured
receive 1 point maximum in a FIDE Album
tourney
7. ,
,
,
EG 166:
8. :
,
1
FIDE
On penalty of rejection:
mining must be explicitly acknowledged
a database position that is both deep and new
occurring anywhere must be explained in
convincing human terms, ie not just with
analysis.
, ,
, ,
,
, , .
[Note the limit of 50 words.]
[: 50 .]
9. EG166 Oct 2006 set out (on p178) three
statements, also in not more than 50 words,
(OK, in English!) of the stance adopted by
other composer-judges. These were by IGM
John Nunn, John Beasley, and the present
writer (see 8 above). The invitation to others
to submit their own statements was taken up by
Michael Roxlau, Jarl Ulrichsen and Sergei
Didukh, but so far by no one else. The
invitation is still open.
9. EG 166 2006 (
.178) ,
50 (OK, in English!),
, -.
. John Nunn, John
Beasley (. 8 ).
Michael Roxlau, Jarl Ulrichsen
, .
.
10. The reasoning behind my stance is the
following:
10.
:
10.1 mining, whatever the skills involved, is
not composing
10.1 ,
,
.
10.2 if a (significant) position is taken from
an EGTB it is automatically analytically
correct. The composer has no responsibilty
for its analysis. On the contrary, he owes a debt
to the database.
10.3 this debt, I suggest, is repaid by
explanation, for the benefit of solvers, judges,
10.2 ()
EGTB, .
. ,
.
10.3 , ,
, ,
213
Winter 2008
Mat Plus Review
and indeed all readers -- even for the
development of endgame theory.
10.4 the composer's contribution to the
precise placing of the chessmen in a selected
position is not nil, but it is minimal.
10.5 there will always be borderline cases.
This is what judges are appointed for. There is
plenty of space for knowledgable discussion,
but this is not taking place. It should.
10.6 central to the above is the tenet that a
tourney award is made to a person for an
original piece of work of significant content.
For a composer to place his name, without
further explanation, above the diagram that he
has submitted to a tourney is to make the
implied claim of it being his own unaided
work.
11. Readers of ZiE may like an example of my
stance in practice. I withdrew from the
triumvirate appointed to judge the memorial
tourney for my lifelong friend Michael Bent.
Why did I withdraw? Because I held to my
published opinion as set out above, with which
my admirable co-judges did not concur. Our
difference was, perhaps, not great: I would
have included in the award a quintuplet study
taken entirely from the relevant database, while
my co-judges wanted to exclude it (as indeed
they did). The force involved was the 6-man
GBR class 0101.02 , or rook and knight against
two pawns. Of course, I would not have given
the composer a tourney honour (prize,
honourable mention or commendation, or even
a special) but only a place. As a result, the
award would have been a mixture of honours
and places. This is my view of the way ahead in
the database world. Is it too revolutionary?
, ,
.
10.4
, , .
10.5 ()
. .
, .
.
10.6 ,
.
,
, ,
,
,
.
11.
.
,
Michael Bent. ?
, , ,
-
(
,
co-judges! .). , , :
,
, (
). 6-
, GBR
0101.02. ,
(,
,
), .
,
.
.
?
John Roycroft, London
FIDE Judge (studies) 1959
17vi2008
John Roycroft,
FIDE () 1959
23.06.2008
214
Mat Plus Review 12
Winter 2009
POLEMIC IN THE STUDIES WORLD (2)
by John Roycroft (United Kingdom)
This contribution is a response to an invitation from
Milan Velimirovic. Milan tells me that he has
received no reaction to Polemic in the Studies World
(1) in MPR Vol.II - No.8 of Winter 2008. He
wondered if I had anything to add. After giving the
matter thought I have decided to accept the
invitation by clarifying, or emphasising, three
aspects. I hope my comments will be found helpful.
. ,
8
MPR (Vol II, 2008) ,
. ,
. ,
.
(2)
POLEMIC IN THE STUDIES WORLD (2)
I. Computers in general
1.
The view is often expressed or implied that the
computer is a tool for composers to use in any way
they like, with no distinctions to be made. In
particular, according to this view, total freedom
applies to the use of endgame tablebases of the kind
first developed by Ken Thompson, and subsequently
by such as Eugene Nalimov. Such a product is
commonly called an EGTB. For reasons that will
become clear, my preferred term remains oracle
database (or odb).
, ,
. , -
,
,
.
EGTB. ,
, "odb"
().
The computer is endlessly versatile. Versatility may
be the computers salient characteristic. It can be
used to further our interest in games, in business, in
storage and retrieval, in classifying, in calculation,
and in research of all kinds. It can further
communication, for good or bad purposes. It can
support other activities by making them faster, more
complete, and even feasible where they were
previously infeasible. Some of these activities will
be trivial, such as the universal and contagiously
attractive game tetris, while others will be less
trivial. One totally non-trivial application is in the
development of weapons of mass destruction.
.
.
, ,
, ,
-
.
, .
,
, -,
, ,
. ,
,
.
-
.
-
,
.
Let us look more closely at the use of computers to
extend human knowledge in fields where humans
have difficulty. We shall assume that such research
is worthwhile, though it is always possible to
maintain the opposite view. That chess is the
drosophila of artificial intelligence, a verbal quip of
Russian scientist Kronrod sr. which has become
legendary, is not universally held, for chess is not
mainstream artificial intelligence, though it is
common enough to be respectable, with papers
published and symposia held. For our purposes we
shall assume that all readers of MAT PLUS
REVIEW concur in the view that algorithmic
computer generation of EGTB/odbs is an activity
worthy of serious consideration, and therefore with
potentially serious implications.
,
,
- ,
, ,
,
,
,
. ,
Mat Plus Review ,
EGTB/odb
,
.
227
Winter 2009
Mat Plus Review 12
An odb is independently verified to be complete and
flawless for its intended purpose, while an EGTB,
especially when used as an adjunct to a chessplaying program, may be abridged or unverified. If
an odbs content relates to a field or sub-field that is
imperfectly known it holds the evidence, not just
some evidence but all the evidence, needed for its
users, up to the most expert of humans, not only to
refine their knowledge, but to test each attempted
refinement. In the case of an odb we are talking
about a human construct, namely endgame theory.
An odb does not extend endgame theory. It supplies
the evidence for us to extend endgame theory.
Odb
, EGTB,
, .
odb
,
, ,
, ,
,
,
. odb
, . Odb
.
.
It is my contention, based on the foregoing, that an
odb deserves to be treated with the respect that any
other piece of scientific research receives in the
realm of academia. We may or may not be
academics, but, so I aver, in this respect we should
strive to adopt the best academic standards.
It further follows that a composer using a flawless
source in pursuit of composing a study, and in
particular when entering for a tourney, is in duty
bound to acknowledge any significant element of his
entry that is not his own work.
, , odb
. ,
,
.
II. Chessplaying programs and EGTB/odbs
Here is my list no doubt incomplete of relevant
contrasting features of (a) chessplaying programs
such as Fritz, and (b) EGTBs (or odbs).
A chessplaying program:
deals with any number of chessmen on the board
is not generated by algorithm (indeed the word
generated in the EGTB/odb sense does not
apply)
needs an evaluation function to limit the event
horizon, ie how deeply it analyses any particular
position
has its own evaluation function to distinguish it
from rivals
will whenever possible improve or modify its
evaluation function in later versions
is in a commecial environment, competing with
other chessplaying programs in the marketplace
will have a use for an EGTB/odb when a relevant
reduction in force is reached
always takes advantage of increased computer
power and memory as brute force features
takes account of the 50-move rule
can look up an evaluation only if it has stored (eg
in hash table) that value already
will always do its best to propose the strongest
move but its choice will not be guaranteed
, ,
,
, ,
.
II. EGTB/odb
(, )
,
(Fritz), EGTB/odb.
:
( ,
EGTB/odb)
,
,
,
50
,
.
228
Mat Plus Review 12
Winter 2009
will have an ELO rating
will have no use of a metric.
, EGTB/odb:
An EGTB/odb:
is generated by algorithm, i.e. it does not analyse
but works backwards from the set of all positions
of checkmate with the given force
is complete for its (limited) field, usually for 5man, 6-man, and, in 2010 increasingly, 7-man
positions
has no use for an evaluation function (ie there is
no 'event horizon' and no need of one)
has no need of improvement
is not in commercial competition with rivals
has no use for an ELO rating, since its play is
already perfect, i.e. with an infinite ELO rating
needs a metric such as distance to mate (DTM)
or distance to conversion (DTC) or an
equivalent, for its process of generation
has no use for a chessplaying program
looks up the pre-computed value (won, drawn,
lost, unknown) of any input
supplies, if the position is a win, the metricdependent win depth (number of moves), and all
the best moves
ignores the 50-move rule (or equivalent), if only
because many endings include longer winning
solutions.
III. Authorship
Following a 15-page article Nalimov tables is
there a problem? by Aleksei Sochnev in the
Russian issue 44 (April 2008), its
issue 47 (April 2009) ran a large number of articles
(all still in in Russian), but then announced closure
of the topic. Nevertheless the St Petersburg quarterly
allowed IGM Oleg Pervakov to contribute on the
same topic in its issue 48 (August 2009).
The composition grandmaster reported development
in his thinking over time. I hope to report accurately
his concluding remarks. The essence of the dilemma
facing us, he says, is the question of authorship. In
his choice of that word I sense support for the view
expressed in the present article. The word
concentrates the mind. For instance, the Codex
promulgated by the FIDE PCCC, and updated or
modified from time to time, states in a footnote that
the existence of a chess position in an EGTB/odb
does not constitute authorship of the position.
On the other hand, the publication of a name (with
no accompaniment) above a diagram is a prima
facie claim authorship.
,
,
()
, 5-, 6-, 2010
, 7- .
(
)
,
,
[. .: ,
2999]
, ,
(
)
(, , , )
, ,
,
50 ,
,
III.
15-
?
44
( 2009), 47
( ),
. ,
48 ( 2009).
. ,
.
, , .
,
.
. , ,
(PCCC) ,
,
,
.
, (
) ,
.
229
Winter 2009
Mat Plus Review 12
Our thinking need not stop there. Since their
development
EGTBs/odbs
have
gradually
introduced a new and major phenomenon, calling
for a re-think of concepts previously familiar.
Specifically, the concept of authorship in the context
of the composed endgame study. I suggest that the
following important distinctions can and should be
made.
,
EGTB/odb
,
.
.
1. . ,
.
(
? ), .
[. .: ,
].
(
).
1. Authorship of idea. The computer has no ideas in
its head because it has no head. Ideas are a human
(does any animal have ideas? We simply do not
know) prerogative, because an idea is abstract.
There is no copyright in an idea. The diagram
name has an uncontested right to this authorship.
(Anticipation and originality are something else.)
2. Authorship of position. The position is in the
database before the user chooses it. Searching for
candidate positions is his aim in consulting the
EGTB/odb. The skills involved in conducting such
searches and making such choices still await debate
a debate that is much needed.
3. Authorship of soundness. The diagram name has
contributed nothing, absolutely nothing, to the
soundness of the chosen position. That soundness
that is guaranteed.
Friends, please think on these things. The computer
may be amoral, but its users do not have that
excuse!
2. . ,
.
EGTB/odbs. ,
,
- ,
.
3. .
.
.
, .
, - !
, 18- 2009 .
John Roycroft
London, 2nd November 2009
(translation: Ija Ketris)
230
