Explainable AI and music

The field of eXplainable Artificial Intelligence (XAI) has become a hot topic examining how machine learning models such as neural nets and deep learning techniques can be made more understandable to humans. However, there is very little research on XAI for the arts. This chapter explores what XAI might mean for AI and art creation by exploring the potential of XAI for music generation. One hundred AI and music papers are reviewed to illustrate how AI models are being explained, or more often, not explained, and to suggest ways in which we might design XAI systems to better help humans understand what an AI model does when it generates music. The chapter then demonstrates how a latent space model for music generation can be made more explainable by extending the MeasureVAE architecture to include explainable attributes in combination with offering real-time music generation. The chapter concludes with four key challenges for XAI for music and the arts more generally: i) the nature of explanation; ii) the effect of AI models, features, and training sets on explanation; iii) user-centered design of XAI; iv) Interaction Design of explainable interfaces.