Reimagining Machine Learning's Role in Assistive Technology by Co-Designing Exergames with Children Using a Participatory Machine Learning Design Probe

The paramount measure of success for a machine learning model has historically been predictive power and accuracy, but even a gold-standard accuracy benchmark fails when it inappropriately misrepresents a disabled or minority body. In this work, we reframe the role of machine learning as a provocation through a case study of participatory work co-creating exergames by employing machine learning and its training as a source of play and motivation rather than an accurate diagnostic tool for children with and without Sensory Based Motor Disorder. We created a design probe, Cirkus, that supports nearly any aminal locomotion exergame while collecting movement data for training a bespoke machine learning model. During 5 participatory workshops with a total of 30 children using Cirkus, we co-created a catalog of 17 exergames and a resulting machine-learning model. We discuss the potential implications of reframing machine learning's role in Assistive Technology for values other than accuracy, share the challenges of using "messy"movement data from children with disabilities in an ever-changing co-creation context for training machine learning, and present broader implications of using machine learning in therapy games.