Reflection in Creativity Support Tool Interaction: Characterisations for AI-based Music Composition

Creativity Support Tool (CST) evaluations in Human-Computer Interaction predominantly apply mixed-methods to assess user engagement. For example, CST evaluations include questionnaires based on the theory of flow: an optimal state where people feel in control and lose self-awareness. Reflection is also crucial in CST interaction. However, reflection is marked by ambiguity and self-awareness, which contrasts with engagement. Few CST evaluations address reflection, and even fewer use reflection questionnaires in systematic mixed-methods evaluations. This thesis challenges the dominance of engagement-based evaluation in CSTs. It argues that a systematic evaluation of reflection in CST interaction is needed to characterise and support the user’s creative process.

The thesis thus develops the Reflection in Creative Experience (RiCE) questionnaire to systematically evaluate reflection in CST interaction. RiCE is applied across user studies in the case study domain of composing music with Artificial Intelligence Generated Content (AIGC). The domain represents the state-of-the-art in CST development and provides a rich and specific context for characterising reflection. The focus is on generative AI: models built from datasets that produce new data with similar properties.

A user study with RiCE characterised reflection in artist-researchers’ use of different AI tools; they reflected on their process when curating AIGC in realtime, and reflected on themselves when organising their curated AIGC. A new AI-based musical CST for reflection was also evaluated, characterising reflection’s interplay with moments of focused engagement; during focus, reflection occurs when users are uninterrupted and learn from AIGC.

The new knowledge contributions are the RiCE questionnaire and novel characterisations of reflection in AI-based music composition. RiCE enables the systematic assessment of reflection in different CSTs and study conditionsan advance on existing engagement-focused tools. The new characterisations of reflection provide value to AI-based musical CST users by showing how to invoke different types of reflection in their practice.