United Humans Technology

Vector Voyage: RAG Card Quest – Facilitator’s Guide for Educators

This guide is designed specifically for educators, teachers, or facilitators leading Vector Voyage: RAG Card Quest in classrooms, workshops, or after-school programs. The game introduces key AI concepts like Large Language Models (LLMs), vector spaces, hallucinations, and Retrieval-Augmented Generation (RAG) in an interactive, low-tech format. It’s ideal for middle and high school students (ages 12+), promoting critical thinking, collaboration, and digital literacy without requiring computers or advanced tech.

As a facilitator, your role is to guide the quest, spark discussions, and connect gameplay to real-world AI. The game is flexible for 20-60 minutes, scalable for group sizes, and aligns with curricula in computer science, ethics, or STEM.

Educational Objectives

By the end of the session, participants will:

• Understand how LLMs “learn” from data (via vector spaces and training).
• Recognize AI limitations, such as hallucinations (inaccurate outputs from incomplete data).
• Explore how RAG improves AI by combining external knowledge without retraining models.
• Discuss ethical implications, like biases in data and responsible AI use.
• Develop skills in teamwork, logical reasoning, and creative problem-solving.

Preparation (10-15 min Before Session)

1. Print and Prepare Materials:
   • Use the printables from the main instructions: 40-60 base sentence cards, 20-30 RAG cards, 10 query cards.
   • Prepare grids: One 5×5 “Vector Map” per group (on paper or whiteboard), and 3×3 “Reinforcement Maps” for RAG.
   • Customize cards if needed: Add icons or simplify language for younger players.
   • Total setup: <30 min; materials cost < $5.
2. Group Setup:
   • Divide into teams of 2-6 (ideal: 4 per group for discussion).
   • Assign roles optionally: “Explorer” (draws cards), “Mapper” (places words on grid), “Guardian” (spots hallucinations), “Navigator” (traces paths). Rotate roles for inclusivity.
3. Session Structure:
   • Total Time: 45-60 min (adjustable).
   • Intro (5 min) + Gameplay (30-40 min) + Debrief (10 min).

Facilitation Tips During Gameplay

1. Introduction (5 min):
   • Welcome players: “Today, we’re on a Vector Voyage quest to explore how AI like chatbots ‘thinks’! We’ll build maps of knowledge, face funny AI mistakes, and use RAG to fix them.”
   • Quick AI Primer: Explain key terms simply (e.g., “Vector space is like a map where similar ideas are close together”). Use examples: “If AI hallucinates that cows eat fish, how do we correct it?”
   • Set Ground Rules: Encourage teamwork, no wrong answers in discussions, and fun over perfection.
2. Guiding Phase 1: Training – Build the Base Vector Space (15-20 min):
   • Monitor: Ensure players place words logically (e.g., group related terms) and draw lines for connections. If stuck, prompt: “How might ‘cow’ connect to ‘herbivore’?”
   • Intervene: If paths seem too easy/hard, suggest skipping stopwords or adding a random element (e.g., dice roll for card draws).
   • Educational Tie-In: Pause midway: “This is like AI training on data – more sentences mean better maps, but limited data leads to gaps.”
3. Guiding Demo Phase: Test the Base (5 min):
   • Facilitate Discussion: After tracing a path, ask: “Is this logical? Why might the AI ‘hallucinate’ here?” Highlight absurd examples (e.g., “Cow eats fish via water link?”).
   • Build Excitement: “Our map has flaws – time for RAG to save the quest!”
4. Guiding Phase 2: RAG – Combine with External Knowledge (10 min):
   • Encourage Creativity: Let teams decide which RAG connections to add: “Vote on the best reinforcement!”
   • Highlight Change: Compare before/after paths: “See how RAG fixed the hallucination without changing the base map? That’s efficient AI!”
   • Adapt for Time: If short, focus on 1-2 queries.
5. Debrief and Reflection (10 min):
   • Group Share: “What was your favorite hallucination? How did RAG change it?”
   • Discussion Prompts:
     • “How do vector spaces help AI understand words?” (Relate to real LLMs like ChatGPT.)
     • “Why do hallucinations happen in AI, and how can we prevent them ethically?”
     • “In real life, how might RAG help with homework or news searches?”
     • “What biases could sneak into our maps? (E.g., if sentences only feature certain animals.)”
   • Extension Ideas: Assign homework – “Run the Python sim at home and add your own sentence.” Or follow-up: “Design a new query card.”

Adaptations for Different Groups

• Younger Players (12-14): Simplify grids (4×4), use more icons, focus on fun hallucinations.
• Older Players (15+): Add complexity – track “parameters” (number of connections), discuss real AI models (e.g., “Grok uses RAG for accurate info”).
• Large Classes: Run in stations; rotate groups.
• Inclusivity: For diverse needs, allow verbal placement (no writing), or pair stronger readers with others.
• Extensions: Integrate with tech – use the Python sim for demos if devices available. Link to ethics: “How can we ensure AI data is fair?”

Potential Challenges and Solutions

• Confusion on Rules: Demo one round as facilitator.
• Uneven Participation: Rotate roles; praise contributions.
• Time Overruns: Set timers per phase.
• Deepening Learning: Provide handouts with AI terms and real examples (e.g., “LLMs like Grok 4 have billions of parameters!”).

This game fosters curiosity about AI while being accessible and fun.