Effects of Visual Explanation on Perceived Creative Autonomy in an AI-Based Generative Music System
Jason Brent Smith, Jason Freeman · 2023
This work aims to explore how a user’s understanding of a creative AI’s decision-making affects their experience when collaborating with it, through the inclusion of Explainable AI features in an interactive generative music system. We have created multiple versions of a generative music system that use different visualization tools to show how the user input motion, captured through a video camera, is being received and interpreted as the manipulation of audio effects. We have designed and conducted a study to measure participants’ perceptions of the system’s autonomy in the creative process, their sense of the system’s ability to support their creativity, and their trust and satisfaction in the visualizations as explanations of the system’s behavior. In this study, we found a range of musical expertise and knowledge of machine learning concepts among subjects, which correlated with their preferences and expectations for autonomy in AI behavior. Increased understanding of the system’s behavior, achieved through interacting with its visualizations, allowed users to perceive it as acting closer to their expectations.