Play Me Something Icy: Practical Challenges, Explainability and the Semantic Gap in Generative AI Music

The "semantic gap" is a fundamental challenge: Users struggle to translate abstract musical concepts into text prompts, creating a disconnect between what people want to hear and what they can describe in words. This gap isn't currently addressed by existing text-to-music tools, making it difficult for users to get desired results even when they have clear creative intentions.

Explainability and usability are in tension: The paper identifies a trade-off between making AI processes transparent to users and keeping tools simple and accessible. Tools that explain their decision-making may become too complex for practical creative use, while simplified interfaces leave users confused about how their inputs produce outputs.

Users need better control without losing creative spontaneity: There's a mismatch between how generative AI music tools work (often producing unpredictable results) and how human creativity actually functions. Users want meaningful input and control over outputs, but current tools don't adequately balance user direction with the exploratory, generative nature of the technology.

Current tools lack transparency about their creative process: The paper emphasizes that users cannot easily understand or predict why the AI makes specific musical choices, limiting their ability to learn from the tool or troubleshoot when results disappoint—a key concern for responsible AI deployment in creative fields.

Improvements should prioritize user understanding: The authors suggest that generative music tools need redesign to better help users understand both how to communicate musical ideas through text and how the AI interprets and executes those ideas.