musical production or perception abilities.By tailoring the selective
forces of the artificial environment and behavioral endowments of the
artificial creatures,we can set up evolutionary simulations to answer a
variety of questions about the evolution of musical behavior.Further-
more,we can listen in on the process of artificial evolution in a way that
we could never do in nature.This is akin to using a cheap electronic
synthesizer to replay an orchestral symphony,perhaps,but it can still
give us an evocative impression of the piece’s outline,and may be one
of the only options available if the natural orchestra can no longer be
assembled.
The kinds of questions that evolutionary computer simulations can
address fall into two main categories.First,simulation models can act as
proofs of concept,demonstrating that a certain behavior could evolve
from some initial state through a series of cumulative stages.For
instance,they could help us explore whether and how a particular kind
of proposed mental mechanism,say,a neural network with certain learn-
ing capacities,could evolve into a system capable of learning and pro-
ducing hierarchically structured musical sequences.Second,simulations
are one of the best tools for elucidating the dynamics of an evolutionary
process,showing what the course of evolution of a certain behavior could
have looked like over time.For instance,one could show how a popula-
tion of singing creatures with some memory ability may build up a shared
culture of songs over time.More generally,one could test hypotheses as
“runnable models,”instantiated,dynamic thought experiments that can
be put in motion within the computer.This provides a means of discov-
ering the implications of ideas that may be too complex to explore purely
verbally.Simulations also can be used to generate hypotheses about the
evolution of real behaviors or about reasons that certain behaviors might
not have evolved.
Evolutionary simulations have several advantages for exploring
behavioral questions (Todd 1996).Perhaps the most obvious and impor-
tant is that they can proceed much more rapidly than natural evolution.
This allows observation of many generations of behavioral adaptation,
and,combined with precise parametric control of simulations,makes it
possible to replay the evolutionary movie under different experimental
conditions.To make simulations run quickly,the evolutionary models
they instantiate must be relatively simple and clear,and to run at all,they
must be coherent and complete (as in any computer program).This is
also an advantage,since it requires that models be carefully thought
out by the researcher and understandable by others.Simulations can
also include a degree of complexity much greater than that allowed,
for instance,by mathematical modeling.Numerous levels of adaptive
processes,including information processing, learning, development,362 Peter Todd