ExplainabilityTransparencyTrustworthinesscs.SD
An Autoethnographic Exploration of XAI in Algorithmic Composition
Ashley Noel-Hirst, Nick Bryan-Kinns
August 11, 2023
arXiv.org
3 citations
Key Takeaways
-
Explainability matters for creative tools: Current AI music generation models are "difficult to understand and control in meaningful ways," suggesting that for musicians to effectively use these tools, they need transparency into how the AI makes decisions—not just access to the final output.
-
Understanding the training data is crucial: The study found that when musicians used the explainable MeasureVAE model, they became more aware of "musical features of the training dataset" itself. This highlights that responsible AI deployment requires transparency about what data trained the system, as it directly shapes what the AI can create.
-
XAI tools work best in real workflows: Rather than being used in isolation as originally designed, the explainable AI model proved most valuable when integrated into musicians' actual creative processes. This suggests that responsible AI design should consider how tools will actually be used in practice, not just in controlled settings.
-
Interpretability enables creative appropriation: The study demonstrates that when AI systems are designed to be interpretable, users can adapt and repurpose them in unexpected ways that go beyond the developers' original intentions—potentially expanding the tool's value and responsible use cases.
Abstract
Machine Learning models are capable of generating complex music across a range of genres from folk to classical music. However, current generative music AI models are typically difficult to understand and control in meaningful ways. Whilst research has started to explore how explainable AI (XAI) generative models might be created for music, no generative XAI models have been studied in music making practice. This paper introduces an autoethnographic study of the use of the MeasureVAE generative music XAI model with interpretable latent dimensions trained on Irish folk music. Findings suggest that the exploratory nature of the music-making workflow foregrounds musical features of the training dataset rather than features of the generative model itself. The appropriation of an XAI model within an iterative workflow highlights the potential of XAI models to form part of a richer and more complex workflow than they were initially designed for.