Experiments in Modular Design for the Creative Composition of Live Algorithms

Musical Autonomy and Live Algorithms

Live algorithms for music (truncated to “live algorithms” or abbreviated as LAMs) were proposed as a conceptual framework by Blackwell and Young (2004, 2005) as a way of analyzing the broad range of live music performance systems that exhibit autonomy or agency in some form or another. They are intended for real-time performance, usually with instrumental musicians, but also on their own or with other live algorithms.

The coining of the term “live algorithms” pulls focus away from the ideal of a virtual musician, towards an eclectic mix of dynamic and interactive behaviors, more generally categorized as open dynamical systems. Although designers of some of the most successful live algorithms were specifically concerned with the simulation of human musical behavior through virtual musicians, they were also involved, actively or inadvertently, in revealing novel ways in which a computer can act as an improvising partner (Rowe 1993, 2001; Biles 1994; Winkler 1998; Lewis 2000). Thus, although “virtual musician” has certain connotations and perhaps undesirable or presently unrealistic expectations, the term “live algorithms” leaves open a wide variety of behaviors, ranging from weakly to strongly musician-like, with modest implications of agency. (The algorithm is itself “live” rather than being “for live use” by a performer.)

Blackwell and Young proposed that researchers and composers interested in live algorithms could collaborate easily by considering the most elementary conceptual breakdown of an entire live algorithm system into basic high-level components: analysis (P) and synthesis (Q) elements with a patterning process (f) in between. As a physically embodied agent, a live algorithm is a computational system connected to the world, minimally, via sensors and actuators relevant to music. Typically this would be a microphone (via an analog-to-digital converter) and loudspeakers (via a digital-to-analog converter). Such embodiment pinpoints the potential for modularity, a mechanism that facilitates substitution: One live algorithm can be substituted for another in a performance context, and live algorithms can be exchanged with musicians in an improvising ensemble. Modularity is a crucial feature of modern design, particularly collaborative design, and facilitates creativity by providing the basis for combinatoric search through a space of possibilities. It is compelling to consider what modular approaches can do to address the technical, collaborative, and creative challenges of developing autonomous musical agents.

Blackwell and Young’s PfQ model suggests an additional layer in which a modular strategy can facilitate a functional mechanism for substitution: the system f that is nested between audio analysis and audio synthesis components (which are themselves modular; see Figure 1). The f component can be thought of as an abstract behavioral system: the brain of the live algorithm. However, the interface between P and f, and between f and Q, is not specified, and there is no reason to believe that there exists a suitable interface for universal collaboration between developers of different modules. A LAM workshop and concert in 2009, organized by the author at Goldsmiths College and Café OTO in London, had this question of intercomponent modularity in mind, but only settled on the broadest notion of communication between elements, where different programmers could specify relevant network messages, to which their collaborators could respond by designing behaviors that conformed to these specific impromptu messages. A number of interesting dynamical systems models were presented at the workshop, such as swarms, recurrent neural networks, and simulations of self-organizing criticality systems such as sand piles and forest fires. Each of these dynamical system behaviors seemed to require different forms for input and output to be creatively embedded in a real-time musical agent, and the predominant tendency was for bricoleur-style (McLean and Wiggins 2010) creative design solutions. This tendency resulted in idiosyncratic systems lacking clear modularity and a clear understanding of interaction between elements.

This article presents examples of “composing” LAMs using a mixture of creative computing techniques. No attempt is made to find a general-purpose modular system as a viable solution for LAM design. Nevertheless, Blackwell and Young’s PfQ model is a relevant way to approach the design of live algorithms, dividing a...