Evaluation of Musical Creativity and Musical Metacreation Systems

Abstract

The field of computational creativity, including musical metacreation, strives to develop artificial systems that are capable of demonstrating creative behavior or producing creative artefacts. But the claim of creativity is often assessed, subjectively only on the part of the researcher, and not objectively at all. This paper provides theoretical motivation for more systematic evaluation of musical metacreation and computationally creative systems, and presents an overview of current methods used to assess human and machine creativity that may be adapted for this purpose. In order to highlight the need for a varied set of evaluation tools, a distinction is drawn between three types of creative systems: those which are purely generative; those which contain internal or external feedback; and those which are capable of reflection and self-reflection. To address the evaluation of each of these aspects, concrete examples of methods and techniques are suggested to help researchers 1) evaluate their systems' creative process and generated artefacts, and test their impact on the perceptual, cognitive, and affective states of the audience, and 2) build mechanisms for reflection into the creative system, including models of human perception and cognition, to endow creative systems with internal evaluative mechanisms to drive self-reflective processes. The first type of evaluation can be considered external to the creative system, and may be employed by the researcher to both better understand the efficacy of their system and its impact, and to incorporate feedback into the system. Here, we take the stance that understanding human creativity can lend insight to computational approaches, and knowledge of how humans perceive creative systems and their output can be incorporated into artificial agents as feedback to provide a sense of how a creation will impact the audience. The second type centers around internal evaluation, in which the system is able to reason about its own behavior and generated output. We argue that creative behavior cannot occur without feedback and reflection by the creative/metacreative system itself. More rigorous empirical testing will allow computational and metacreative systems to become more creative by definition, and can be used to demonstrate the impact and novelty of particular approaches.

MorpheuS: Generating Structured Music with Constrained Patterns and Tension

“…A second expansion would be to allow for more flexible pattern detection, such as recognition of inverted patterns, augmentations, and diminutions and other variations. It would equally be interesting to evaluate if the generated music elicits the same emotion responses as expected given a tension profile, by measuring physiological responses or by recording listener judgments of tension as described in [92,93]. The tension model could also be expanded to capture other characteristics of tension such as timbre and cadence.…”

A Functional Taxonomy of Music Generation Systems

“…While human feedback may arguably be the most sound approach for evaluating post-hoc if the generated pieces sound good [Pearce and Wiggins 2001;Agres et al 2017], requiring people to rate the output at each step of the process can take an excessive amount of time. This is often referred to as the human fitness bottleneck [Biles 2001].…”

“…For example, music based on fragments of an already existing composition, as in the case with high-order Markov models, run the risk of crossing the fine line between stylistic similarity and plagiarism [Papadopoulos et al 2014]. Evaluating the creativity, which is sometimes equated to novelty, of the generated music is a complex topic treated in greater length in Agres et al [2017].…”

Designing Computationally Creative Musical Performance Systems

“…Within Computational Creativity (CC) systems deemed to display creative behavior are capable of reflection. 1 Without this element of reflection, a point raised by Agres et al [4], Bundy [9] and others, systems are only generative. As, in general, CSEMPs do not employ a full reflection loop or selfreasoning of their output, they are not considered to be creative systems from a CC standpoint.…”