A Short Review of Responsible AI Music Generation

Artificial intelligence continues to become increasingly embedded in musical practice and yet there is little evaluation of how transparent and ethical these systems are. Surveys of AI models to date focus on the technical features of AI models and there is a lack of surveys of the practical and ethical application of AI models for musicians, producers, and composers. This paper surveys 27 contemporary AI models for music generation in terms of creative input and output (symbolic music, audio, text, or image), musical task (symbolic composition or audio generation), and Responsible AI properties of transparency & explainability, fairness, accountability, and ethical AI. Analysis of these facets of AI model use in creative practice highlights the trade-offs and challenges in designing equitable and ethical AI tools for music-making. The survey highlights a lack of transparency and control of AI model training and fine-tuning, a lack of openness of licensing and source code, a lack of ethical reporting of training datasets, and a focus on AI models for audio generation at the expense of real-time music generation for use in composition, performance, and improvisation. Our analysis offers insights for researchers, developers, and musicians seeking to navigate this fast evolving landscape of musical AI. We suggest that research is needed to develop clearer frameworks for evaluating AI models in creative domains, focussing especially on user journeys that help users understand the mechanics, limitations, and ethical considerations of these systems in music making practice.