Rethinking “In-The-Loop” Approaches in Educational AI: A Learning Sciences Perspective

Abstract

In educational AI, “In-the-Loop” framings (e.g., human-in-the-loop) are presented as systems that secure human-technology balance in enabling timely feedback and personalization. This learning sciences perspective-piece argues that the critical issue is not whether humans participate sufficiently in educational processes, but how the educational loop is rendered actionable. When educational activity is organized around cycles of data capture, computational interpretation/inference, and probable responses, pedagogical action is reconfigured around what computational systems can detect, compare, and act upon. In higher education, these operational requirements can create tensions between promises of meaningful human participation and the conditions under which participation occurs. Actual enactments of educational AI often rely on epistemically simplified proxies that shape what counts as evidence, constrain interpretation, redistribute agency, and impose system-defined temporality on natural educational cycles. In the context of assessment, this poses risks where predictive indicators of learning drift toward a kind of deterministic practice that normalizes data overreliance and trajectory-prescription. Such tensions persist even with increased human oversight/automation, because they arise from the loop’s operational commitments to closure and actionability. Through learning sciences perspectives, I caution against treating loop operability as compatible with how learning, as a process, happens, and as a sufficient pedagogical condition. Finally, I outline implications for learning-theoretically grounded approaches to educational AI.