Preconfigured and GPT NPC Conversations

The NPCConvo (Non-Player Character Conversation) task invites learners to observe a scripted exchange between two or more AI-driven characters.

In collaborative AI-to-AI interactions, avatars engage each other while actively prompting the player to weigh in—encouraging critical thinking, role-taking, and participation in unfolding dialogue scenarios.

Pedagogical Rationale

NPCConvo tasks are grounded in social learning theory (Bandura, 1977), which posits that learners acquire knowledge by observing others. By watching AI characters interact, learners witness processes such as:

• Problem-solving strategies

• Conflict resolution

• Linguistic variation and tone

• Cultural or clinical reasoning

• Decision pathways and ethical choices

These tasks also draw on vicarious learning principles, where learners interpret and internalize lessons from second-hand experiences. In VR, this is particularly powerful because learners feel physically "present" during the interaction, enhancing emotional resonance and attention (Makransky et al., 2020).

This GPT-powered NPC interaction models workplace conflict and provides the learner with embedded hints and emotional cues. These dialogues scaffold upcoming decision-making tasks by prompting observation and contextual reasoning.

Classlet’s NPCConvo format immerses learners in AI-driven character dialogues as a way to observe, interpret, and analyze interactions. These tasks are designed to support vicarious learning, communication analysis, and scenario modeling — particularly powerful in immersive environments. Classlet supports two distinct modes of NPCConvo:

---

🟧 1. Preconfigured Dialogue (ConvoP)

Fixed-script interactions for controlled learning moments

ConvoP tasks are fully scripted and authored by the instructor, offering a high degree of control over the learning content. The learner passively observes a conversation between two or more AI avatars, designed to demonstrate a target concept or social dynamic. These scenes often serve as modeling tools to illustrate:

• Proper vs. improper communication

• Decision-making under constraints

• Emotional tone, power dynamics, or clinical clarity

Pedagogical Strengths:

• Supports explicit teaching goals with predictable learning outcomes

• Enables scaffolded reflection — e.g., learners are prompted to analyze errors, summarize positions, or evaluate outcomes

• Ideal for compliance, ethics, or language training

Typical Flow: Scene launch → Dialogue play → Reflective prompt → Learner response → Feedback

This preconfigured NPC dialogue in Lingnan University, introduces a realistic learning dilemma, prompting learners to consider strategies for information literacy. By modeling uncertainty and emphasizing effective research skills, the scene supports scaffolded reflection before task engagement.

---

🟨 2. GPT-Generated Dialogue (ConvoG)

Dynamic AI roleplay with unpredictable variation

ConvoG uses prompt-configured GPT avatars to simulate real-time conversations, either between multiple AI agents or between an AI and the learner. Rather than a fixed script, the dialogue unfolds based on:

• Character background and response configuration

• Learner input or scenario variation

• Optional RAG (Retrieval-Augmented Generation) support for factual grounding

Pedagogical Strengths:

• Enables adaptive, generative simulations

• Useful for open-ended reasoning, argumentation, and exploration

• Models human–AI collaboration and situated improvisation

Configuration Options Include:

• Role design (e.g., doctor, CEO, citizen)

• Response tone and constraints (ask questions, avoid high-level abstraction, challenge assumptions)

• RAG integration to support deep knowledge alignment (e.g., from policy documents, medical manuals, or case studies)

Typical Flow: Prompt configuration → Avatar dialogue unfolds dynamically → Learner may observe or join → Reflect/respond → Feedback loop

This scene at University of Hong Kong, uses fish characters to simulate a bullying scenario, where one expresses insecurity and confusion about its abilities. Such dialogues help learners develop empathy and critical thinking by exploring social-emotional cues in a low-stakes, immersive context.

---

🧩 Instructional Design Implications

Feature	ConvoP	ConvoG
Scripted?	Yes	No – generated dynamically
Learner role	Observer	Observer or participant
Control	High (curated path)	Low-medium (generative)
Use cases	Modeling, analysis, assessment	Exploration, reasoning, engagement
Best for	Ethics, compliance, language pragmatics	Open-ended dialogue, hypothesis testing, adaptive training

🔄 NPCConvo Task Flow

Step	Description
1. Scenario Setup	Instructor defines the scene, topic, and characters (e.g., doctor–patient, historical figures, coworkers in a workplace).
2. Dialogue Playback	Learner observes a fully scripted NPC dialogue presented as voice, text, or both. Speech bubbles or voice actors enhance realism.
3. Prompt Delivery	A reflection question appears after the exchange: e.g., “What misunderstanding occurred?”, “Which response was inappropriate?”, or “Summarize the advice given.”
4. Learner Response	Response can be in the form of MCQ, open text, or audio. The task may require interpretation, inference, or rephrasing.
5. Feedback	The system or instructor provides clarification or further insight, reinforcing critical analysis and deeper comprehension.

🧠 Instructional Uses

NPCConvo tasks are especially useful for:

• Language and pragmatics: Learners observe social cues, politeness strategies, or speech acts in action.

• Medical communication: View example consultations to analyze tone, empathy, or diagnostic questioning.

• Ethics or civic reasoning: Observe dilemmas unfold through scripted dialogue.

• Workplace training: Simulate team meetings, negotiations, or peer discussions.

• Cultural understanding: Highlight values, customs, or linguistic diversity in interactions.