Humanised avatars, dialogic spaces, and the future of online learning

Abstract

This paper examines how humanised AI avatars powered by large language models are reshaping online learning and asks whether they can act as genuine participants in educational dialogue rather than mere imitators of human conversation. It situates this question in the wider context of recent demonstrations that conversational AI can pass behavioural versions of the Turing Test, arguing that the key issue for educators is not imitation but authenticity within dialogic spaces. Drawing on a multi-iteration design-based research study in an online Executive MBA programme, the study combines dialogic theory with quantitative and qualitative analysis, using Tech-SEDA coding, the SOLO taxonomy and a Naïve Bayes classifier to link features of asynchronous discussion to levels of cognitive engagement and summative performance. The findings show that AI-mediated interventions can significantly increase higher-order dialogic moves, deepen students’ conceptual understanding and predict learning outcomes from dialogic quality with high accuracy, while also highlighting important limitations around authenticity, agency and the absence of machine consciousness. These results imply that humanised AI avatars can be productively used as scalable simulators and mediators of dialogue when they are transparently framed as non-conscious tools, carefully orchestrated by teachers, and embedded within ethical design principles that protect student autonomy, promote critical AI literacy, and cultivate genuinely dialogic, polyphonic learning communities.