(Anti)-Intentional Harms: The Conceptual Pitfalls of Emotion AI in Education

Abstract

'Emotion AI' is a subset of artificial intelligence (AI) technologies that claim to be able to detect the inner emotional states of individuals by collecting biometric information such as face scans, voice recordings, and traces of physical movement. Despite their growing popularity in education, these systems have the potential to produce serious harm. In this paper, we argue that a major concern with emotion AI technologies has to do with the theories of emotion that undergird them. Most emotion AI technologies are built on the foundations of anti-intentionalist theories of human emotion, which claim that emotions can be understood as discrete, universal states that arise as automatic physiological responses. Anti-intentionalists suggest that emotions are not directed at any object, or subject to cognitive reasons. In our work, we focus on the increasing use of these technologies in education to illustrate the ways in which these anti-intentionalist systems are problematic, as they dissolve the space for pushback against the judgements they make. We argue that their use thereby contributes to harms towards children broadly centered around student disempowerment, surveillance, and classification. We then consider three alternative policy approaches to emotion AI use in schools in light of their role with this political agenda of emotion commodification, assessing each of these options - interpretability, technical reform, and non-use - for their desirability and feasibility. In doing so, we underscore the conceptual harms produced by emotion AI systems in the context of education, and the criteria by which these technologies should be judged by educators and policymakers.