Information Technology Reference
In-Depth Information
are strategies to make this seemingly impossible problem at least partially tractable.
One way is to look for the simplest possible implementation of each required com-
ponent, still being sufficiently complex for the overall emergent creative behaviour
to appear. Certain core features of each component may suffice to arrive at inter-
esting results. It is a research problem in itself to find this level, though, as dis-
cussed by Cope (
2005
). But the more minimal the implementations are—while still
functioning—the more general conclusions we can draw.
There are two hard problems involved. Firstly, how do we implement suitable
conceptual representations? Secondly, there is the related problem of how to imple-
ment re-conceptualisation from material form into new conceptual models. I have
stressed the importance of misunderstandings in the parsing process, since they help
form a personal expression. Then a rather simple re-conceptualisation model might
suffice to start with, or a combination of simple models running in parallel, to widen
the repertoire of recognised material. Each model interprets the given material in
a particular way, and the choice of models will contribute to the “personality” of
the algorithm, in the same way as the characteristic shortcomings of a human artist
contribute to his personal style.
8.4.2 Conceptual Representations
Knowledge and concept representation has always been a problem in computing.
The conceptual representations in this model need to be flexible and open-ended,
but we want to avoid the symbolic approach of traditional AI, for reasons explained
earlier. While Pearce and Wiggins (
2002
) mention the ability to represent musical
material as a hierarchical structure (see also Dahlstedt
2004
;
2005
), McCormack
(
2005
) states that representation and generative mechanism should be on the same
level as the material resulting from the process, hence a collapse of hierarchies. This
is an important point, and I believe the iterated process between material form and
conceptual form bridge this gap between levels, and provides a path between them
in both directions.
A conceptual representation has two components: a description of what we want
to achieve (e.g. desired properties, list of constraints), and a description of what we
want to do (a generative procedure or list of tool actions). Let us call them
descrip-
tion
and
instruction
. In a goal-driven creative process, with a clear vision of the fi-
nal form of the work, the description component is more important. But the clearest
vision may be revised if something unexpected but interesting is found. Also, a de-
termined idea about the description may still lack sufficient detail to form the basis
of a full artwork. Hence, flexibility is still needed. On the other hand, a work based
primarily on a generative idea may lack an description component, and instead give
more weight to procedural instructions. In free improvisation, description may ini-
tially be empty, and both are open for change, according to how the process unfolds.
So, both components are needed, in a weighted combination, to cover a wide range
of processes.
