Reducing Barriers to the Use of Marginalised Music Genres in AI

AI music systems currently favor dominant genres: The paper identifies that high-quality AI music generation relies on extremely large datasets, which means Western Classical and pop music are overrepresented while marginalised music genres are systematically excluded from AI training data.

Small datasets can reduce bias when properly used: Researchers found that fine-tuning large AI models with smaller datasets of underrepresented music genres can actually help address bias in deep learning models, challenging the assumption that bigger datasets are always better.

Transparency and control are critical for cultural fairness: The study highlights that explainable AI (XAI) tools—which make AI decision-making more transparent and controllable—are essential for ensuring marginalised music genres are fairly represented and that users understand how AI systems work with their cultural music.

Including underrepresented cultures strengthens AI systems: Participants emphasized that working with marginalised music genres, despite technical challenges, is important because it improves cultural representation and helps address broader bias issues in AI development.

This is an emerging area requiring community collaboration: The researchers are building an international network for responsible AI in music, indicating that addressing these barriers requires ongoing collaboration between technologists, musicians, and communities from underrepresented cultures.