ABSTRACT
Conversational intelligent tutoring systems have demonstrated utility in facilitating deep learning across a wide variety of subjects. This typically takes the form of a tutor agent playing the role of a human tutor or a peer agent working with the learner. Some systems have employed “trialogues” with both agents together, enabling an array of pedagogical strategies not possible with a single conversational partner. Newer systems allow for four (and potentially more) distinct “personalities,” affording the application of both novel strategies and those adapted from conventional schoolhouse instruction. This begins to step beyond the common paradigm of a virtual classroom (an online repository for learning content and communication modes)—it, in fact, constitutes a virtual class, complete with complex, programmable, adaptable social structures into which system designers place human learners. This phase shift toward a new educational paradigm involves a wide range of variables, both established and novel, that we begin to identify here. These include, notably, new and modified pedagogical strategies, design and impact of controlled social situations, causes and effects of increased and sustained engagement, and, of course, learning outcomes. Critically, the psychological impact on the learners is not well understood and deserves preliminary examination with respect to these factors.
