The Appprentices is an ongoing robotic sound installation that materialises machine learning through vibration.
Three soft, identical objects move on a circular stage, producing material sounds amplified by the vibratory space of the stage. Progressively, their sounds and movements synchronize, as if they were learning to help each other. Then, they separate and move off-beat, as if they were learning to compete against each other. Finally, when the audience comes and softly hit the stage, they learn to reproduce the rhythm of the vibration, as if they grew attached to starting a communication.
The Appprentices result from a diffractive practice of machine learning, entangling design, music, art, engineering, and anthropology. We elicited social concepts to guide the design of behaviours for our object collective. We composed rhythmic structures as temporal forms embodying these concepts. We prototyped multi-agent reinforcement learning, reinforcement learning, and supervised learning to generate rhythmic variations within each behaviour. We designed a vibrating drone as a learning sonification to reinforce the feeling that objects tune into their environment—suggesting embodied ways for humans to be attentive to their non-human counterparts.
The project is being developed with Samuel Bianchini and Elena Tosi-Brandi, with the support of Joffrey Becker, Franck Weens, Victor Audouze, Corentin Loubet, and Joséphine Mas, in collaboration with the Reflective Interaction group of EnsadLab, and the Art Direction Nods team of Orange Xdlab, in the context of a postdoctoral fellowship program at École des Arts Décoratifs, Paris.