Emergent Sound Repertoires in Virtual Societies

Computer simulations wherein musical forms may originate and evolve in artificially created worlds can be an effective way to study the origins and evolution of music. This article presents a simulation in which a society of distributed and autonomous but cooperative agents evolve sound repertoires from scratch by interacting with one another. We demonstrate by means of a concrete example the important role of mimetic interactions for music evolution in a virtual society. The article begins with a succinct commentary on the motivation for this research, its objectives, and the methodology for its realization. Then we state the objective of the particular simulation introduced in this article, followed by a detailed explanation of its design and functioning, along with a critical assessment of its results.

Background

Motivations and Objectives

Why should one study the origins and evolution of music by means of computer simulations? There are theoretical and pragmatic motives behind this research. The quest for the origins of music is not new; philosophers of all times have addressed this problem. The book Music and the Origins of Language by Downing Thomas (1995) offers an excellent review of the theories purported by philosophers of the French Enlightenment, for example. More recently, The Origins of Music, edited by Nils Wallin, Bjorn Merker, and Steven Brown (2000), collates a series of chapters written by leading contemporary musicologists. With the exception of one chapter (Todd 2000), however, none of these thinkers seek theoretical validation through computer simulation. Although we are well aware that musicology does not necessarily need such support, we do think, however, that computer simulation can be useful in developing and demonstrating specific theories of music.

Our research attempts to describe a theoretical framework for studying the origins and evolution of music based upon the notion that music is an adaptive complex dynamic system. In simple terms, adaptive complex dynamic systems emerge from the overall behavior of interacting autonomous elements in a non-hierarchical manner; that is, there is no central control driving these interactions. In our case, these individual elements are modeled as agents geared to interact cooperatively with one another in a virtual society under specific environmental conditions. We are particularly interested in establishing the fundamental properties and mechanisms that these agents, the interactions, and the environment must possess in order to create music. The rationale here is that if we furnish these agents with proper cognitive and physical abilities, combined with appropriate interaction dynamics and adequate environmental conditions, then the virtual society should be able to evolve realistic musical cultures. This research, therefore, addresses three relevant questions: What are the appropriate agent interactions and what motivates them? What are the proper cognitive and physical abilities of these agents? What environmental constraints are necessary to foster musical evolution?

From a pragmatic perspective, we are interested in developing new technology for implementing autonomous generative music systems. The use of computers as generators of musical compositions was pioneered in the mid 1950s by such people as Lejaren Hiller and Leonard Isaacson, whose 1956 composition Illiac Suite for String Quartet is recognized as being the first computer-composed piece of music (Manning 1985). Since then, we have witnessed a number of impressive algorithmic composition systems (Miranda 2001). These systems can generally be classified into two categories: systems that learn to produce music from given examples (Cope 1987; Linster 1990; Todd 1989) and systems that embody some sort of mapping from algorithm to musical phenomenon (McAlpine et al. 1999; Supper 2001). Although the great majority of these algorithms were not designed for music in the first place, some of these systems can indeed generate interesting music. In general, however, the quality of this music is seldom satisfactory to the discerning audience. As for those systems that learn from given examples, some of them are able to satisfactorily mimic established musical styles such as medieval, baroque, and jazz. We are not, however, interested in mimicking particular musical styles here. Can we possibly discover what is needed to improve these systems? Would self-organizing grammars as opposed learned ones lead to interesting new compositions? Would musically biased generative algorithms as opposed to adapted...