Labels and Emotions Machine Learning's Scientific Objects

Abstract

Abstract: One strategy for challenging the use of automated emotion recognition systems has been to contest the validity of the psychological research programs on which they are based, by critically examining their history. This article takes a related but distinctive approach, examining instead how machine learning and emotion recognition became connected in the recent past. It argues that machine learning transforms the theoretical categories that it operationalizes, the judgments that it purports to automate, and show more broadly how technical and scientific objects can interact. A shared emphasis on facial images and emotion labels made the psychology of emotional expression highly compatible with machine learning techniques. Emotional expressions, in turn, were interpreted within a machine learning worldview in which visible expressions can point to deeper underlying signals. Future normative work on emotion computing might draw on such accounts of how machine learning transforms its scientific objects.