Explainable AI for the Arts

The Arts offer a complex and demanding domain in which to undertake AI and Explainable AI (XAI) research with substantially lower life-critical risks than domains such as health, mental health, medicine, and automotive contexts. However, despite the growth in XAI research there is very little research on XAI for the Arts (XAIxArts) with most XAI research examining task-oriented explanations of AI decisions. This paper outlines a framework for characterising XAI for the Arts and illustrates its use by reviewing 87 AI music generation systems in terms of their explainability. A demo XAI generative music system is introduced which offers meaningful real-time interaction with latent spaces for music generation. The paper concludes by reflecting on the conundrum of how much explanation we should strive for in XAI for the Arts.