Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!usc!ucla-cs!ucla-se!turing!bsmith
From: bsmith@turing.seas.ucla.edu (Brian Smith)
Newsgroups: comp.ai
Subject: Re: Music and AI
Message-ID: <3191@lee.SEAS.UCLA.EDU>
Date: 25 Jun 91 15:54:54 GMT
References: <23491@shlump.lkg.dec.com> <3084@lee.SEAS.UCLA.EDU> <10936@idunno.Princeton.EDU>
Sender: news@SEAS.UCLA.EDU
Organization: SEASnet, University of California, Los Angeles
Lines: 35

In article <10936@idunno.Princeton.EDU> eliot@phoenix.Princeton.EDU (Eliot Handelman) writes:
>
>The problem is setting constraints for the planner, ie, what constitutes 
>an admissable "move," so to speak. (Allowing a second simplification of
>restricting possible musical actions, but that's like restricting yourself
>to a couple of tom-toms -- the nature of problem isn't altered).  Musical 
>constraints are exceptionally difficult to get a hold of, because they are 
>cognitive, cultural and volatile, rather than formalistic and invariant. 
>Almost certainly the same is true of creativity.

Yeah, I can buy that.  However, I don't think we should be thinking about
planning in the GPS sense of initial state leads to goal state thrugh a series
of operators.  In a sense, we have to redefine the notion of planning for the
musical realm.  The very fact that MOST composers have to go through extensive
periods of refinement as well as complete goal alteration (or, at least, I
certainly do) leads me to thing that some sort of "weak" or thematic planner
is necessary to perform the task.

As well, I see your point about the constraints, and I've certainly argued that
autonomous composers lack depth due to the fact that they lack cultural,
competence, and performance knowledge.  And yet, to oversimply again, couldn't
we provide the information to the machine in, say, rules or connectionist 
paradigms?  
>
>It's much more interesting to focus on the listener, because apprehension
>is the simplest act of creativity. 

Listening is pretty fascinating stuff, but I think that composition is a much
more interesting problem since one can't compose without listening.  To some
extent, everyone is a composer (i.e. humming arbitrary tunes, melodic contours
in speech, etc.), so it does seem to be a worthwhile area of study.  Once the
knwoledge gained through listening is captured, how do we use it in the
performance domain?

-- brian