culture,and evolution,can all be incorporated into the models,adding
greatly to their realism and predictive power.But we must be careful to
avoid the attraction of building complex models for their own inherent
interesting behavior,and instead construct specific models to address
specific questions in an accurate and analyzable fashion.Finally,simu-
lated creatures can be dissected,probed,and prodded in ways that
normal animals would not withstand,and a battery of psychological and
“neurological”tests can be performed to assess their behavior for com-
parison with living experimental subjects.
In this chapter we consider ways in which evolutionary computer
simulations can be built to help investigate questions regarding the
evolution of musical behavior.In the next section we describe the main
approaches currently used to make computer models of individual
musical behavior:rule-based and learning systems.These are the behav-
ioral mechanisms that an evolutionary simulation will modify over time,
through selective forces acting on their musical output.In the third
section we explore tools available for simulating the evolutionary
process itself.This process requires the ability to generate new musically
behaving individuals and assess their fitness before reproducing the
most-fit individuals in the next generation.We list four main ways of
evaluating individual fitness:human critics,automated rule-based critics,
learning critics,and coevolving critics.The latter form of fitness evalua-
tor is used in the fourth section in a system designed to explore the gen-
eration of musical diversity through the coevolution of music creators
and critics,akin to male songbird singers and female listeners choosing
mates from among those singers.Finally,we conclude with further ques-
tions on the origins of music that can be pursued with evolutionary
simulations.What Should We Evolve in a Music System?
To investigate the evolution of musical behavior and cognition in a
computational system,we must first determine what exactly we want
to evolve.We are not in general interested in the evolution of specific
instances of music per se,but rather in how music-generating or
-processing mechanisms change over time.We therefore need computa-
tional models of music composition and comprehension.Two main types
of models have been developed for these tasks over the past few decades:
rule-based and learning mechanisms (see Loy 1989,for more details).
Generally,the former are given their rules by the designer,but it is pos-
sible to create a system that learns rules as well,combining the two
approaches.Both kinds of systems can be used for either musical363 Simulating the Evolution of Musical Behavior