In M. Fan, M. Horn, & M. Roussou (Eds.), Proceedings of the annual meeting of ACM SIG Interaction Design and Children (IDC 2025). ACM.
ABSTRACT: Artificial intelligence (AI) offers promising opportunities in providing educational feedback. However, AI agents often lack the nuanced, multimodal interactions of human teachers. Twenty-seven elementary school students participated in a 45-minute individual activity targeting integer arithmetic (e.g., “-2 + -3 = _”) to assess the comparative effectiveness of feedback provided by human teachers (n=15) and AI agents (n=12). Audio, video, and eye-tracking data were collected. Pre- to post-assessment scores revealed double the gains in the human condition. We draw on embodiment theory, ethnomethodological conversation analysis, and multimodal analytics to identify students’ interpretive solution reenactments in the human condition as staging dialogic affordances via (a) co-speech indexical gesture as disambiguating speech utterance; and (b) rapid iterative attunement to their interlocutor’s multimodal responses. The study contributes to a growing area of research around AI pedagogical agents by complexifying its technical and theoretical discourse.