Subject: RE: Virtual Concerto
From: Michael Rempel (Michael@VIDIR.COM)
Date: Tue May 18 2004 - 12:43:52 EDT
I think we may be dancing around the same issue from different perspectives.
I have answered you below but this is my over all impression.
> >Without a database and transition set, one must rely on
> algorithms. There is
> >little to recommend algorithms over databases. Algorithms
> are harder to
> >create, and in the final analysis rely heavily on human
> interpretation to
> >decide what is 'good' or 'interesting' enough to survive as
> It's a little misleading to speak of "algorithms," since the
> issue here
> is more like
> "problem description," which is about what, rather than how (which an
> algorithm answers).
Problem descriptions are algorithms on a higher meta level.
You are creating boundaries, not absolute internal cohesiveness.
> >Conceptualization of what is actually taking place implies a
> basis for
> >judgment of the validity, value, interest, or (insert
> measure here) of the
> >musical event that has been randomly or algorithmically derived.
> I disagree -- it involves a representation of what, potentially, is
> heard, or how it's heard. I don't see
What and How are value driven questions. Rules are implied, whether they are
explicit or not, whether they are subject to change or not.
> "criticism" as part of this, although it was or is Rowe's approach.
> Think of it more as
> the problem of sustaining an intelligible conversation with a bot. Of
> course the translation of
> terms -- the musical equivalent of "intelligible conversation" -- is
> something up to the metacomposer
> to figure out. It's part of your job description. In the
> early part of
> the 20th C, there were articles
> published in esthetic journals purporting to supply formulas for the
> calculation of "beauty" -- of
> the form B = A + B + C/2. Of course this can't happen because
> b. is in
> the ear of the
> listener. And besides, any a priori critical principles could be
> implemented in the generation
> routines, so there's no point trying encapsulate any such principles,
> assuming they exist. What
I don't follow this argument. It may be lack of coffee on my part but it
> CAN happen is modeling of (your version of) cognitive process, or of
> attentional models,
> in conjunction with your best ideas about music "works." How
> do things
> foreground or cohere
> or transform? When your computer hits a note, will Jimmy
> Durante wish to
> stop the music and
> hear that note again, is it possible to work out the convening logics
> that give rise to that highly
> satisfying shove into the climactic register? Could your program
Emotional content. Judgment required.
> recognize this in a few broadway
> tunes, computationally would it rely on the idea that such a
> note is the
> "highest", or could
> it trace through the delineation of confluencing factors? All you're
measuring internal cohesion and deriving maximum impact of emotional
> trying to do is to express
> ideas about how your (not my) ear thinks along with the
> music, if I may
> invoke Adorno's
> "structural listener" (Socio. of music) here. Can that ear be
> intelligibly simulated? Perhaps,
> but can that ear be programmed to know what it likes? This I doubt.
Within limits. I.E. what is music probably can not be programmed, but what
is good music can and is constantly. Even the simple admiration of an
excellent player of an instrument is enhanced greatly by the ability to play
that same instrument at all.
In that music is a language of empathy, computers are a natural antithesis.
So it is a lot of fun to mess with them to make empathetic language. But the
only math of empathy I understand is in me. Teaching it to a computer is on
the order of difficulty of getting a blind man to paint a masterpiece. But
it is fun to try.
> >This is
> >essentially a reactive database or algorithm set, rather
> than a pro-active
> >compositional one.
> If you take the approach you've just outlined, yes. In my
> program the
> thing that tries to understand
> works together with the thing that tries to compose. (The thing that
> programs has a big
> headache at the moment.)
You made a rule set about composition vs. judgment. It is quite analogous to
the lower meta level of additive vs. subtractive synthesis.
> >Either way human judgment is imposed, albeit at arms
> >length with some programmer imposed degree of randomness.
> That's ok. If someone ever gets to square one, maybe
> somebody will get
> to square two in a few decades.
I am saying that the simple database stuff is square one. If you mess with
it and find it organizing along certain dimensions, you can then encode
those dimensions and mess some more. Eventually you get to a model of the
thing by appreciation rather than only direct a priori cognition. It makes a
lousy PhD thesis, but it works wonderfully.
> >If your objection is to simple combinations with little or no musical
> >theory, then by all means your objection could stand. Except
> for the cases
> >where music is interesting because the rules are broken or
> bent, and it
> >still 'works' somehow.
> Ok. But I'm not personally endorsing the idea that music is about
> "rules," and that
> good music is about "following the rules." On the other hand many
> pieces of music
> create their own particular law validated only by that single
> composition. For instance
You are making my point for me. Laws or rules don't need to be universal. I
never implied they were. Internal consistency and establishing a premise are
valid. Working with established rules gives you more freedom however
(ironically) because you can assume more without loosing your listener, and
you need not articulate every permutation to derive a satisfying climax.
> the patterns of Prelude 1 of the WTC I are unique to that piece, at
> least as far as I
> know. The idea of a symphonic movement saturated with a
> 4-note pattern
> like a generalizable idea (eg, should symph. 6 have consisted of the
> same idea with
> a 5-note pattern?) Metacomposing is, I think, about trying to invent
> conditions under which
> something might recognize the potential of a vast material
> simplification and arrive at
> something potentially addressable only in its own terms.
You bite off a huge chunk with that statement. It would be interesting to
see you articulate what 'recognize the potential' means even on a linguistic
level, let alone a computational one. Try doing it without rules :-)
Internal consistency I accept completely. On the whole it is the easiest
concept, but computationally it is difficult to maintain as a meta
principle. The question is how do you derive or constrain adequately a
musically satisfying internally consistent set of terms? An example: Is a
phrase that takes 45 minutes to resolve itself, while wandering dangerously
close to resolution several times satisfying? Answer: maybe. What constrains
it? Answer: sustaining interest for 45 minutes. What sustains interest for
45 minutes? Segmenting time, and exploring variations individually and in
combination in interesting ways leading to mini-climaxes that could resolve
but don't. ... etc.
The questions seem to grow, rather than diminish, and each question could
have many answers. As meta composer I need to answer the question "what are
good rules" for creating internal consistency.
My Answer: I cant tell you but I can help you organize your ideas, and given
a rule set I can compare it to other existing rule sets, and give you an
approximation of an esthetic valuation based on experience and personal
Isn't my answer really what academics do? And as soon as you or I write a
program that can do some aspect of "it", other academics will write "it" off
as trivial, and of no academic value. The reason for which is that they can
no longer rely on that particular bottle of mysterious elixir as a source
for their income and prestige. "It" has become available to the whole world.
Sociologically speaking academic value relies on narrowcast-ability of a
given thing. Each academic needs to own a set of ideas exclusively in order
to achieve or maintain position in academia. Where as software is, by it's
nature, a broadcast medium.
> >My fundamental assertion is that if it sounds good it is good.
> I've changed my mind about a few pieces over the years.
Ditto. No issues.
> >Composition by any means whether by algorithm, database or
> human, good sound
> >can only be human judged.
> Make way for the robot invasion.
Doubtless the invasion has hardly begun, but despite a long career as a
systems analyst / architect I maintain that living cognition and emotion is
The ultimate. Nothing has higher value.
Don Quixote step aside. I shall now take a turn at that windmill. And I the
greater fool than you, for I know it is no dragon but something much worse;
unkillable rock and steel.
> -- eliot
This archive was generated by hypermail 2b27 : Sat Dec 22 2007 - 01:46:01 EST