Freaky: performing hybrid human-machine emotion

This paper explores the possibility of using statistical classification of physiological signals into emotion categories as a resource for open-ended human interpretation of emotion. Typically, design studies for affect assume either that it is possible for computers to objectively identify users' emotions, or that emotion is completely subjective and thus rely solely on human interpretation. By drawing on the feminist concept of performativity, we explain how to conceive of computational representations and human actors as co-constructing emotions. Through a case study of Freaky, a system that uses such models of emotion to sup-port human interpretation, we demonstrate how machine learning models of affect can be constructed and incorporated in systems designed for open-ended user interpretation of affect. Qualitative results from a user deployment show that a performative approach to modeling emotion is possible. We thus demonstrate the potential of performative theories to be generative of new computational and design practices that support hybrid human-machine enactments of emotion.