Behind the Chatbot: Interactive AI Mechanics

This interactive flipped classroom or workshop activity explores generative AI by seamlessly blending a self-guided AI conversation with an in-depth workshop connecting information literacy frameworks, LLM mechanics, ethics, and prompt engineering.

Try it yourself: Use the link below to launch Behind the Chatbot and work through the same asynchronous, self-paced conversation your students will experience — no setup required, just click through it on your own.

Part 1: Pre-Class Activity — Behind the Chatbot (20–40 Minutes)

Students step through a one-on-one, browser-based conversation with an AI persona built specifically for this activity — no prompt to copy and paste, no chatbot account to set up. Framed as a "backstage pass," the activity walks students through 13 sequential steps, one concept at a time, building toward a live demonstration:

• Next-Word Prediction & Probability: A role-reversal opener where students act as the AI to feel out statistical token likelihood before it's ever named.
• Tokenization & Vector Embeddings: Students test the model's blind spots directly (e.g., counting letters it can't "see") and solve spatial word-analogy puzzles that reveal how meaning is mapped geometrically.
• Attention Mechanisms & Context Windows: Introduced through the "moving highlighter" metaphor for resolving ambiguous words, with a note on where memory runs out.
• Temperature & RLHF: Students watch the same prompt produce an "Accountant" vs. a "Poet" response, then sit in the trainer's chair to see how human feedback shapes AI behavior — and its dark side, reward hacking.
• The Black Box & Hallucinations: The activity closes with its emotional peak — students invent a fake fact, watch the AI confidently fabricate details about it, and see the reveal that nothing in its tone gave it away.

Every student's conversation autosaves. Instructors get a passcode-protected dashboard to see who's completed the activity, track progress in real time, and read (or copy) full transcripts — so this is no longer just an assigned reading, it's something you can verify was done.

Part 2: In-Class Workshop (1+ Hour)

The workshop builds directly on the pre-class conversation, grounding discussion in core ACRL Framework concepts — "Information Creation as a Process" (comparing peer review with probabilistic text generation) and "Authority Is Constructed and Contextual" (contrasting academic authority with SEO "Domain Authority"). Students practice vertical and lateral reading, revisit LLM mechanics (tokenization, embeddings, attention, temperature, RLHF) through worked visual examples, work through a structured set of ethical concerns spanning bias, job displacement, privacy, environmental impact, and misinformation, and close by learning a four-part framework for building effective research prompts: Role/context, Task, Constraints, Format.

Materials Included

• Behind the Chatbot (link below): An interactive, self-scoring AI activity that guides students through the pre-class conversation and gives instructors a completion dashboard.
• Presentation Slides: Slide deck and visual outlines for the in-class workshop.
• Activity Source Code: The full HTML/JavaScript code for Behind the Chatbot is too large to share via the ACRL Sandbox, but I'm more than happy to share if you contact me directly.
• Original Script: The original, non-code script used to build the artifact.

This resource is ideal for multiple-session library instruction, embedded course partnerships, or standalone AI literacy workshops.

Disclaimer

Behind the Chatbot guides AI behavior through a structured, guardrailed script, but AI outputs are probabilistic and cannot be fully controlled. Responses will vary across sessions. The author is not responsible for unexpected, inaccurate, or inappropriate content generated during the activity — in fact, this variability serves as a core teachable moment for the lesson. Because the activity saves student names and responses to run the instructor dashboard, please review it against your institution's student-data policies before assigning it.

Feedback & Collaboration

I am interested in any experiences you have with this activity, whether through personal testing or in a classroom setting. Please feel free to reach out to share your findings or discuss the activity further via my personal email, ALA Connect, or LinkedIn.

Updated 8/13/2026