[Smt-talk] Aesthetics of Computer-Generated Music

Eliot Handelman eliot at colba.net
Tue Apr 12 07:37:50 PDT 2011

Passing the Turning Test suggests an already finished and polished
musical AI. The problem is that computer composition is in its infancy
and it's probably unfair to demand that 1st & 2nd generation programs
perform at the level of established masters. The question is how to
recognize an intermediate result that is perhaps not yet great music,
but which suggests esthetic attitudes and theoretical stances that can
be developed.

For example, there's a huge problem in generating plastic and
differentiated forms without following built-in patterns or
established formulas and without noodling. The test of a system/theory
meant to do this is whether it succeeds in generating a diversity of
such forms. If it does then there is perhaps something to the
underlying theory. That takes one step that can be built upon.

How long should it take to develop musical AI? 2-3 months using
machine-learning? Or rather, maybe, 50 years of concentrated research
in which new styles of computational musicology unfold?  If there's
only "success" and not "progress" then we haven't a chance. This is
not a one-man or one generation job.

-- eliot

Eliot Handelman, PhD

More information about the Smt-talk mailing list