This practice contribution focuses on interactive learning with artificial intelligence (AI), specifically learning environments in which learners communicate with AI avatars through natural spoken-language discourse. These scenarios differ from earlier approaches, such as intelligent tutoring systems or pedagogical agents, through their high interactivity, natural language flow, and the capacity of current Large Language Models (LLMs) to support open communication about almost any content. This enables educators to configure and deploy diverse learning scenarios at different difficulty levels, including algebra tutoring, virtual patient simulations in medical education, peer-feedback role-plays in special education, and interactive teachers in STEM education. However, it remains unclear how such scenarios should be configured regarding interaction and appearance. Regarding interaction, the broad affordances of LLMs must be narrowed to specific learning situations with defined goals. Prompting must therefore constrain the LLM so that it provides authentic and motivating dialogue without deviating from the intended learning objectives. Strong responsiveness to learners may increase authenticity and motivation but risk losing instructional focus; strict goal adherence may make the scenario rigid and demotivating. Regarding appearance, the avatar’s look, sex, and voice may shape perceptions of personality, credibility, and competence. This contribution first discusses the benefits and characteristics of avatar-based learning scenarios. It then introduces a peer-feedback role-play and an interactive teacher scenario, conceptualizes each educationally, and discusses their AI-avatar implementation, including prompt-specific peculiarities and pitfalls. Finally, it examines effects of avatar appearance on users’ perceptions and derives practical conclusions.
Zellner et al. (Mon,) studied this question.