Embodied Exploration of Latent Spaces and Explainable AI

Embodied interaction can make AI systems more understandable: The paper demonstrates that performers using body movements to interact with an AI audio synthesis model could explore and understand how the model works, suggesting that physical, intuitive engagement may help demystify complex AI systems for users.

Art and performance offer practical pathways to explainable AI: By integrating artistic practices with AI transparency principles, the research shows that creative domains can contribute meaningfully to making AI systems more interpretable—not just through technical documentation, but through experiential understanding.

E-textiles enable new forms of human-AI communication: The use of sensor-equipped textiles to translate body movements into AI model interactions represents an innovative interface that could make AI exploration more accessible and intuitive than traditional technical interfaces.

Embodied practices reveal AI behavior intuitively: The paper suggests that allowing people to physically "feel out" how an AI model responds through movement and sound creates a more natural understanding of the model's latent space than abstract visualizations or code alone might provide.

Interdisciplinary approaches strengthen AI responsibility: By bringing together performance, embodiment, and AI explainability, the research illustrates that responsible AI development benefits from diverse perspectives beyond computer science and engineering.