Controlling emergence in an interactive multi-agent musical system

2017-03-20T21:59:32Z (GMT) by Daniel Waghorn
Swarm intelligence has quickly become a popular technique for designing problem-solving systems, imitating the collective behaviours of social animals to address problems with very large solution spaces. Swarm systems specifically try to replicate and harness two properties of swarms in nature – self-organisation, where group behaviour works without central controllers; and emergence, where interaction between agents in the swarm produces higher-level patterns and structures. <br>     <br>    The tendency of swarms to produce very complex behaviour even when the agents they comprise are very simple makes them immediately promising for use in generative music; however, while music is often highly structured, swarm behaviour is generally unpredictable. A small number of musical systems have begun to experiment with swarming techniques, simulating the behaviours of groups of birds, ants or fish, and translating their movements into music. While these existing systems are useful for discovering the kind of music that popular swarming algorithms can produce, there has been no focus so far on how to control these swarms, and therefore it has remained unclear whether they are suited to use as musical tools. <br>     <br>    This thesis aims to address controllability of musical swarms, by examining how control mechanisms added to a musical swarm might influence the swarm to create higher-level musical features as demanded by a user. As part of this goal, a new swarm-based musical tool, called Chirashi, was developed and is demonstrated within. Chirashi implements a common flocking swarm algorithm, and turns user input into influence upon the behaviour of agents within the swarm, to guide groups of agents to produce music with desired aesthetic features, while retaining the swarm’s ability to act autonomously to produce emergent behaviour. In addition to the tool itself, the thesis considers the musical effects of control mechanisms upon the swarm, examining its behaviour and capacity to comply with musical demands while in use in actual music making. Above all, this thesis intends to expand the repertoire of tools available to practitioners of real time, generative music, by establishing control over a technique that is unpredictable enough to surprise its user, to make it predictable enough to allow the user to learn and master it.