When I collaborate with artificial agents to explore ways of interacting with sound through body movement, I deliberately curate the set of body motion features that are “visible” to them. I hide what I believe is not useful for my purposes, and design their perception of their environment to my own liking. However, some deep learning algorithms are notoriously “black boxes”: it is possible to observe incoming data and outgoing data, but their internal operations are often difficult to explain in ways that are easily understandable by humans. Something is visible, yet something is concealed, or incomprehensible. This lack of (mutual?) “understanding” can be at times fun or frustrating, a source of anxiety as well as of inspiration.
Federico Visi, GEMM))) Gesture Embodiment and Machines in Music – Luleå University of Technology, Sweden