Explainability Paths for Sustained Artistic Practice with AI

Human control over training data is essential: The paper emphasizes that artists and creators need agency not just when using AI tools, but throughout the entire pipeline—from selecting and curating training materials through preprocessing and model training. This contrasts with typical AI tools where users only interact with the final product.

Smaller datasets can be viable alternatives: Rather than requiring massive datasets, the research demonstrates that small-scale, carefully curated datasets can support effective generative audio models, making AI tools more accessible to individual artists and smaller organizations without requiring enormous computational resources.

Interactive machine learning enables better creative control: By integrating interactive machine learning as a "mapping tool," creators can iteratively refine how their inputs translate to outputs, allowing for a more transparent and controllable creative process rather than treating the AI as a "black box."

Explainability supports sustained artistic practice: The paper argues that current AI tools prioritize ease of use over transparency, which actually undermines long-term creative work. When artists understand how and why their AI systems make decisions, they can develop deeper, more intentional artistic practices rather than treating the technology as a novelty.

Transparency requires rethinking AI development priorities: The research suggests that responsible AI for creative fields means shifting away from the industry standard of "immediate accessibility" toward systems designed with explainability built in from the start, benefiting both artists and the broader goal of trustworthy AI.