Skip to main content
Queen Mary Academy
  1. Queen Mary Academy
  2. Educators
  3. Innovation and Scholarship
  4. Innovative Pedagogies
  5. Centre for Excellence in AI in Education
  6. Blog
  7. Items

When Generative AI Becomes a Learning Partner in Virtual Reality

Dr Lei Fang draws on a recent international study, reflecting on what it takes for Generative AI (GenAI) to act as a learning partner, rather than a pedagogical novelty.

Published:
Dr Lei Fang Heasdshot

Virtual reality (VR) is often promoted as an immersive way to engage students, while GenAI promises personalised support through conversation and feedback. But when the two are combined, an important question arises for educators: does embedding a GPT-powered avatar in a VR environment genuinely support learning — and under what design conditions?  

In a recent study, my international co-authors and I, working across the UK, Malaysia, and Hong Kong, examined how students interacted with a GenAI-powered avatar embedded within a VR learning environment. In this setting, students completed learning tasks inside a virtual space, while the GenAI avatar provided real-time prompts, guidance, and reflective questions (see Figure 1). This allowed us to explore not only engagement, but also students’ cognitive and motivational responses when GenAI is embedded directly into immersive learning activities. 

On initial observation, students responded positively. Many found the GenAI avatar engaging and helpful, and appreciated the immediacy of support within the VR environment. However, our findings revealed a more nuanced picture.  

Why “adding an AI avatar” is not enough 

There is a growing assumption that simply embedding AI into VR will automatically enhance learning. Our findings suggest otherwise. While students generally found the AI avatar engaging and helpful, learning outcomes depended heavily on how AI scaffolding was designed — not on the presence of AI or immersion alone. In other words, AI does not become a learning partner by default. 

We found that: 

  • Students’ expectations of usefulness and their confidence in using the system strongly shaped whether they wanted to continue learning with VR. 
  • Interacting with the GPT avatar increased students’ metacognitive activity (planning, monitoring, evaluating). 
  • However, greater reflection sometimes reduced students’ sense of immersion and perceived usefulness, revealing a trade-off between reflection and flow. 

This challenges a common assumption in immersive learning: more realism or more interaction does not automatically mean better learning. Instead, our findings show that the key lies in balancing immersion and reflection. Students often found the VR experience most engaging when they could move smoothly through tasks without interruption. When the AI avatar prompted them to slow down, justify their decisions, or evaluate alternatives, their learning clearly deepened—but the experience could feel less immersive. This highlights the importance of careful orchestration: designing moments of reflection that support deeper learning while preserving the overall sense of immersion. 

What does this mean for educators? 

For educators considering GenAI and immersive technologies, several practical lessons emerge: 

First, AI scaffolding must be intentional. AI prompts should be aligned with learning goals, not simply designed to respond to student queries. 

Second, personalisation matters. Students differ in confidence, prior knowledge, and readiness for reflection. Adaptive AI scaffolding is more effective than one-size-fits-all approaches. 

Third, reflection should be supported without breaking flow. Reflection is essential for deeper learning, but it needs to be well-timed, lightweight, and purposeful — especially in immersive environments. 

Finally, immersion is not the end goal. Effective immersive learning design balances reflective depth, interactivity, and cognitive load. 

Learning with AI, not just from AI 

Taken together, our findings suggest a shift in how we think about GenAI in education. Rather than treating AI as a tool for efficiency or content delivery, we can design it as a learning partner that supports questioning, sense-making, and judgement. As AI and immersive technologies continue to evolve, the challenge for educators is no longer whether to adopt them, but how to integrate them in ways that remain pedagogically meaningful. The real opportunity lies not in automation or realism, but in thoughtful orchestration of AI, reflection, and human judgement. 

 If you’d like to read more about the study or reference it in your own work, please cite the paper as: 

Fang, L., Zhou, X., Shen, D. J., Nurgissayeva, A., Techanamurthy, U., & Lopez-Ozieblo, R. (2025). Generative AI as a learning partner: Structural insights from a VR–GPT educational platform. In Artificial Intelligence XLII: Proceedings of the Forty-Fifth SGAI International Conference on Artificial Intelligence (AI-2025), pp.437-443. Cambridge, England: Springer. https://link.springer.com/book/10.1007/978-3-032-11442-6. 

Dr Lei Fang

Senior Lecturer, Deputy Head of Centre for Mathematical Education, Chair of GenAI Working Party of Institute and Faculty of Actuaries (IFoA). 

https://www.qmul.ac.uk/maths/profiles/leifang.html 

 

 

Back to top