strosity containing thirty-five notes of minuscule duration (mostly
triplets) jumping over three octaves.The authors noted (1995:45):“In
retrospect it is clear that the network had far too small a training set to
learn about many of these kinds of errors....”
The general problem here is that evolutionary search processes are
highly adept at exploiting weaknesses and quirks of fitness functions:the
evolving creators are constantly looking for the easy way to higher
fitness,and jumping on it when they discover it.This means that in prac-
tice,a researcher may have to modify a particular fitness function a few
times before it is specific enough to lead to the evolution of desired
musical behaviors and to avoid being tricked by shortcut solutions.In
nature,this kind of fitness-function evolution often happens automati-
cally,for instance,when a species of predator discovers a new way of sur-
prising its prey,and the prey must adapt a new defensive strategy in turn.
This kind of back-and-forth reciprocal modification of selectee and selec-
tor can also be captured in evolutionary computer simulations,where it
can be used to study another class of phenomena:coevolution of musical
production behavior and perceptual preferences.Evolution of Musical Diversity through Coevolving Creators and Critics
Evolutionary simulation tools developed by musicians looking for new
ways to generate creative compositions can be adopted to explore spe-
cific scientific questions.We modified some of these tools,for instance,
to investigate ways that musical diversity can be generated within and
across generations,seeking to answer the question,“why are there so
many love songs?”Some aspects of this project illustrate the way that
evolutionary computer simulations can be put to scientific use (for more
details,see Werner and Todd 1997).
Species with highly evolved,elaborate communication systems often
have a great diversity of signals within a given population and between
populations (including successive generations and recently diverged
species) over time.Humans of course have an unmatched capacity to
generate novel signals,both linguistic and musical.Many songbirds have
repertoires of dozens of distinct song types,a few species can sing hun-
dreds of different songs,and the brown thrasher checks in with a remark-
able repertoire size of over 2,000 (Catchpole and Slater 1995).Moreover,
any one male of a given songbird species typically sings a different reper-
toire from other conspecific males.Moving from air to ocean,humpback
whales each sing a unique song (Payne and McVay 1971;Payne,this
volume),and even cephalopods (particularly cuttlefish,octopuses,and
squid) have a surprising variety of signal types,with some species using373 Simulating the Evolution of Musical Behavior