Website founded by
MatPlus.Net Forum Internet and Computing Azlan Iqbal on beauty
You can only view this page!
|Page: [Previous] [Next] 1 2 |
|(1) Posted by Siegfried Hornecker [Friday, Jan 11, 2013 16:36]; edited by Siegfried Hornecker [13-01-11]|
Azlan Iqbal on beauty
The Malaysian top expert Professor Iqbal has written an interesting article on ChessBase.
I have been sent his program Chesthetica a few years back, and it was very interesting. Emil Vlasak had written an article in EG back then.
There was also an article I wrote for the Schwalbe where I referenced that program in a footnote as an example of how attempts are made to evaluate beauty in chess composition, but this article is not published yet.
|(2) Posted by Hauke Reddmann [Monday, Jan 14, 2013 17:10]|
The deeper implications of this finding are...scary, to say the least.
|(3) Posted by Sven Hendrik Lossin [Monday, Jan 14, 2013 17:14]|
Have you tried this tool, Hauke?
|(4) Posted by [Monday, Jan 14, 2013 18:23]; edited by [13-01-14]|
>The deeper implications of this finding are...scary, to say the least.
How so? The program in question sorts PGN games/endgames in order according to some
predefined criteria of beauty. It doesn't (as far as I can say) describe them,
nor does it quantify them. It just say this PGN game (or endgame) is more
beutiful than the next one. By a large margin? or a small one? within some
limits of confidence? Or not? It's silent on those points.
But will it affect anything? Unlikely ...
In the late 60s and early 70s(I think -- I can't find all the relevant stuff
right now) there was an american (F. Vaux Wilson -- who in 1914 became one of Good Companions) who
developed a system for evaluating beauty in chess problems (MOE - Method Of
Evaluation). There was even a tourney (or perhaps several) to evaluate the
system in practice. (Those tournament reports might be worth resuscitating,
to see if the MOE scores and ordinary judge-by-mind agreed.) (One early
version of MOE is cited in another Iqbal paper: A Systematic and Discrete View of
Aesthetics in Chess -- just Google for it.)
MOE was refined over time, though I suspect Vaux Wilson was the only one who worked
actively. But it was published in the Problemist, so it was not dismissed as
crank work -- at least not entirely: Ling, the editor, seemed to have some degree
of interest in it. But judging from comments, it was regarded as a waste
of effort by some.
[Dig, dig, dig] ... I do come up with one version of MOE published in The Problemist in
November 1970 (vol. 9, no. 6, p. 95-96. It's not so detailed that it could be
turned into a computer program easily, but it's not *very* far off.
I also remember seing a pamphlet where a couple of problems were scored according to
MOE, which helped explain some of the thoughts behind it, but finding that again
will take more digging than I am prepared to do right now.
There was also a successor to MOE called 'Move Strategy' ... Problemist, November 1971
volume 9., no. 12, p.179-180).
As far as can judge, it didn't leave much of a trace then.
But I'll have to leave further investigations to others.
Trying out that computer program on the 1000 first studies of HHDB III (release 1, I think),
covering the years 2003-2005, I come up with four studies that tie for 'first place':
N Wortel: 2q2n2/8/4N3/1K5k/5P2/1Q6/5r2/8 +
V Zheltukhov: 1n6/8/P7/8/8/2K2k2/7p/3N4 + (64-Shakmatnoe Obozrenie#080)
I Vandecasteele: 8/8/2B1q3/p7/4N1N1/P1K5/8/n6k + (Humor Tourney EBUR#096)
A Golubev: 8/kp2Bq2/2PP1N2/1P3P2/1KP5/8/7b/8 + (2.c Vecherni Krasnoturinsk-10 JT)
But it should be noted that from those first 1000 studies, only 527 were found to
be OK. The rest were logged as 'incompatible' mostly because of 'wrong score notation',
and that the last ply was not played by white; these studies were ignored. Those four 'top' studies may consequently not
rank higher than shared 500th place, if all 1000 studies had been considered ...
Someone else will have to evaluate them for beauty ...
From a very superficial look: Wortel's study looks like an entry for a competition for the longest sequence of checks,
and the next two also have a tendency to check-check play. The last one at least has
an underpromotion, even if it also involves a check ...
I wonder if the aesthetics are those of the 1840s or if there is more to find in there?
|(5) Posted by Joose Norri [Tuesday, Jan 15, 2013 00:11]|
Later in the 1970s Wilson published very detailed MOE tables in Schach-Echo. (Or perhaps in Die Schwalbe?) Apart from the theme tourneys, in which I think only the nature of the key move was stipulated, he also compared the results of several very prestigious awards against the system. (At least one complete one year run of SE twomovers.) The ranking order came out quite differently (needless to say, I think).
Does anybody think the system of Prof. Iqbal is more credible?
|(6) Posted by Siegfried Hornecker [Tuesday, Jan 15, 2013 07:54]; edited by Siegfried Hornecker [13-01-15]|
I think Professor Iqbal - I was surprised to find he is only 34 years old! - has a revolutionary idea! The program as I had it a few years ago was very bug-infested (it crashed all the time), but it was interesting to test it with a few problems. In the end I came to the conclusion that it is on a good way but needs more work.
I did not test the new version yet.
Hauke's implications are interesting, but sadly he didn't specify what he means.
It could range anywhere from having human judges replaced with a program over having humans completely replaced to even worse things. I don't think that will happen in the near future, though.
Thanks for the article, Anders! Here a direct link:
|(7) Posted by Nikola Predrag [Tuesday, Jan 15, 2013 09:36]|
What did Hauke mean by scary implications?
Wouldn't it be wonderful to have the unique esthetic criteria, such that a computer can judge the human creative work? Having these criteria, the computers themselves would be able to create and the humans would not have to bother with creativity any more, we would at last become the mere consumers. No divergences, one Central Mind and billions of uniform ants. At present, few humans are still needed to govern such globalization, it's nice to see that soon the computer will be able to take the lead, to become the Fuhrer :)
|(8) Posted by Jacques Rotenberg [Tuesday, Jan 15, 2013 10:21]|
Many thanks Nikola. Excellent demonstration. May I add that the will of a 'coherence' for problem world have also some common roots with this (rather naive) 'automatic aesthetic' will.
|(9) Posted by Hauke Reddmann [Tuesday, Jan 15, 2013 14:03]|
People, think outside the problemist ghetto! I mean the philosophical implications.
First, it was shown that you need no intelligence (I deliberately do not define
the term :-) to play chess - mere computing suffices. That was a bit annoying to
humanosuprematists, but could be handwaved away. Chess is a mathematically exact
game at all. And you aren't overly butthurt when a Volkswagen drives faster than
you can run, either.
But now this. Esthetics is impossible to formalize! A Little Green Man from
Alpha Centauri declares Hauke Reddmann to be the best problem componist alive!
(OK, I bribed him with...no, I don't tell :-) A blasphemy to even DARE to write
such a program! Das Literarische Quartett mit Roboranicki, Robokarasek, Robolöffler
and Roboguest! Don't you see the utter absurdity!
But OK. The stuff was published in a decent journal (or so I guess), peer reviewed
and declared as sound science. Headline 2100: We phased out mankind. We no longer
had any need for it.
01001000 01100001 01110101 01101011 01100101
|(10) Posted by Hauke Reddmann [Tuesday, Jan 15, 2013 14:06]|
@Joose: Sic Herbert Ahues on it :-) He already shot MOE down in flames
with carefully constructed counterexamples. OK, he's 90 by now, but he
still composes razor-sharp and probably feels as philosophically insulted
by this program as I do.
|(11) Posted by Nikola Predrag [Tuesday, Jan 15, 2013 15:41]|
Well, I hope that the sarcasm in my post was easy perceivable by everyone (not only by Jacques).
|(12) Posted by Sven Hendrik Lossin [Tuesday, Jan 15, 2013 15:44]|
I simply don't understand why they put so much effort in something that can never work and as Anders pointed with a poor result.
It would have helped our problemist community by far more if he invested his power in a program that checks problems for originality.
|(13) Posted by [Tuesday, Jan 15, 2013 18:48]; edited by [13-01-15]|
Sven Hendrik Lossin asks:
>I simply don't understand why they put so much effort in something that can never work and as Anders pointed with a poor result.
Mainly because they are researchers, and research is/should be free. And then perhaps because they don't agree with the idea that it never can work. It may even be that they regard success as something else than problemists do. (And it may just perhaps be because it's possible to get research grants in this area.)
Myself, I suspect that something like MOE can be made to work (with a lot of work) ... which to me would be to do much the same evaluation as a human judge. 'Much the same' doesn't mean 'exactly the same', only that most of the problems end in much the same positions. That's close enough: even human taste differs. It would be something fairly close to the old 'Eliza' program or 'Doctor' that, for a short while, could fool a person into believing a computer actually could keep a conversation going. But it's just mechanical application of scoring rules to produce an evaluation.
It won't replace a human judge -- the program won't be able to write a coherent tournament report. (Though, come to think of it, some judges ... )
The value ... well, it would be the get somewhat closer to understanding how beauty (in chess) is perceived.
The results I reported ... well, I'm not sure they were significant. From the error messages, it was fairly clear that the input data had to be prepared in some way: just throwing studies at the program as I did would be as useful as pouring random liquids into the tank of a car and expecting it to work. Even so, to have reached a level of 1840 in chess aesthetics is not all that bad -- it could have been early 1500s. In a year, we may perhaps hope to see 1900 standards.
>It would have helped our problemist community by far more if he invested his power in a program that checks problems for originality.
Some similar research a couple of years ago looked very much like research into optimization, which just happened to use chess problems as the 'arena' in which the optimizations were made. Easy to misunderstand as research into chess aesthetics for uninformed onlookers.
|(14) Posted by Sven Hendrik Lossin [Tuesday, Jan 15, 2013 20:47]|
But didnt they do the second step before the first one?
Before I can judge problems I have to look for forerunners and these forerunners have a great influence on the problems originality which should be an important part of the beauty.
So the first step for my opinion is to optimize the search algorithms in this regard and do the comparison process afterwards.
|(15) Posted by Kevin Begley [Tuesday, Jan 15, 2013 22:51]; edited by Kevin Begley [13-01-16]|
I don't [entirely] trust human judges [even after months (sometimes years!) of consideration, even without knowing the composers, I couldn't even trust myself, entirely]
I'm certainly not ready to trust human algorithms for computer judges.
But, there is no denying that this is interesting research.
Eventually, such research may enable computers to dominate certain aspects of formal chess composition.
Regardless, I expect there will always be areas where human creativity is unchallenged.
At least until self-aware circuitry presents a rival for the human brain (unlikely, in this lifetime).
While I agree that forerunner research is a key component of good evaluation, it is not necessarily the logical first step for a computer evaluation algorithm.
Eventually, that may be connected -- and, it is worth nothing that databases are already implementing techniques to search for forerunners -- but, there's no reason that this should forestall an holistic evaluation effort.
You can not properly measure improvements (over some earlier comparable work), if you can not comparatively assess one against the other.
So, this does seem a logical starting point.
How will humans ever appreciate the moment when computers do become better judges of chess problem art?
How do we know, for sure, that they are not already there?
The truth is, we can't begin to trust computers with artistic evaluation, because great art will never reduce to a quantified thematic checklist -- it is a deeply personal experience, founded upon shared circumstances.
Great art conveys a feeling, which computers are unable to detect.
Until circuitry can establish (indeed demand for itself!) a shared emotional connection with humanity (e.g., sadness, hope, joy, loneliness, etc), it's impossible to trust that they share our circumstances.
And, the purpose of art, in my view, is to convey something meaningful to an audience which shares our emotional perspective.
At present, computers are unable to demonstrate an ability (so far unique to humans) to appreciate art on a level of sentient experience.
Therefore, we are probably right to discard their output in such matters.
Still, computers would not be the first to overlook great art, in favor of some biased checklist of formal technical achievements (we humans make similar oversights, quite frequently).
And, we presently have no means to discard the erroneous output of human judges.
If we want a credible technique to measure the quality of a computer judge, we have to begin by measuring human judges; and, the only reliable way to do this (as far as I'm aware) is statistical analysis.
Baseball relies on statistical analysis (done by computers) to grade umpires -- but, is far from ready to trust the strike calls to computers.
The point is: computers helps human umpires to improve (and significantly, if we believe the statistics).
Similarly, I expect that computerized evaluation of chess problems will offer an invaluable tool to help human judges improve (particularly in formal and technical areas).
We should all welcome progress in this area -- because, I think most would agree, our umpires could use some help!
|(16) Posted by Hauke Reddmann [Wednesday, Jan 16, 2013 12:00]|
Kevin wrote: "Great art conveys a feeling, which computers are unable to detect."
So there, but what makes you so sure that your "feelings" aren't computable
from a Very Large Program either? Just google "Valentin Braitenbach" - the most
elementary feelings can be emulated by ludicrously simple contraptions.
Right, essentially I'm exhuming Leibniz' vision here. Personally, I think humans are
fundamentally more than computers, but this though might reduce to a stack overflow
in my Hybris Subcircuit 2013AFFE. ;-)
|(17) Posted by Kevin Begley [Wednesday, Jan 16, 2013 21:51]|
The development of a sufficiently complex neural matrix should give rise to a human level consciousness.
And, from this, there is no reason not to expect that a vast set of shared emotions will quickly develop.
Computers will be the judges -- it's only a question of time.
|(18) Posted by Nikola Predrag [Thursday, Jan 17, 2013 01:44]|
I'll try to rearrange a joke: A man was granted one wish. He asked "when I wake up tomorrow, I want may hands being long enough too reach the floor". When he woke up, he noticed that he has no legs.
A computer would match a human if the latter would lose his humanity. Unfortunately, this looks quite achievable (:
|(19) Posted by Ian Shanahan [Thursday, Jan 17, 2013 12:09]|
Aesthetics, which is not based upon fixed axioms or logic to the exclusion of all else, transcends Gödel's Incompleteness Theorems. Computers - at least those built to date - cannot.
|(20) Posted by Siegfried Hornecker [Friday, Jan 18, 2013 08:10]|
Humans - at least those built to date - cannot.
|Page: [Previous] [Next] 1 2 |
MatPlus.Net Forum Internet and Computing Azlan Iqbal on beauty