Abstract:

Preparing future STEM teachers requires more than strong subject knowledge; it also involves developing the ability to navigate classroom dialogue, respond to students’ ideas, and make decisions in real time. However, opportunities to practice these skills in safe, low-stakes environments are often limited. This study explores how AI-powered simulations can support this aspect of teacher education by providing accessible and flexible opportunities for practice.

A 3D AI-driven character conversation simulation was developed to enable users to engage in role-playing scenarios based on realistic classroom situations. To support a gradual introduction, participants first interacted with the system through a 2D text-based interface in a web browser. This allowed them to experiment with personalised prompts and observe how the AI responds, helping them understand how to guide interactions. They then progressed to a more immersive 3D environment with a 3D pupil character as the avatar, where they engaged in voice-to-voice conversations with an AI-powered character, simulating more natural classroom exchanges and responding to dynamically generated student dialogue based on predefined scenarios and user input. This transition from text-based interaction to immersive communication reflects a scaffolded learning design, supporting users as they move from controlled experimentation to more authentic practice.

The participants were master’s students enrolled in Natural Science: Authentic Scientific Research in the Classroom, preparing to become educators, with some also aiming for future roles as teacher trainers. This provided valuable insight into how individuals with strong disciplinary backgrounds and developing pedagogical awareness engage with AI-supported character simulations. Following the experience, participants completed a survey evaluating realism, interaction quality, usefulness of feedback, support for reflection, usability, and ethical considerations. The findings indicate that while the perceived realism of AI interactions was moderate, the simulation was highly valued as a practice tool. In particular, participants appreciated the relevance and usefulness of the feedback and highlighted AI-based dialogue as an effective way to explore and rehearse different teaching scenarios. These findings align with prior work on simulation-based and experiential learning (e.g., Kolb, 1984; Schön, 1983), emphasizing the importance of iterative practice and reflection in teacher development. Participants also raised considerations related to trust and ethical use of AI, underlining the need for responsible design.

Overall, the study suggests that AI-powered simulations can serve as a meaningful addition to teacher education by offering structured, low-stakes, and scalable opportunities for practice and reflection. At the same time, the gap between feedback quality and interaction realism highlights an important area for future development. Enhancing the naturalness and believability of AI interactions may further strengthen user engagement and learning outcomes. This work contributes to ongoing discussions on simulation and immersive technologies in education, highlighting how AI can support more practice-oriented, flexible, and scalable approaches to preparing future STEM educators.

Keywords:

AI in Education, Simulation-based Learning, Pre-service Teacher Education, STEM Education, Immersive Learning, 3D Virtual Environments, Conversational AI, Role-playing.