AI Builders Discussion Guide

Four questions for after the Prompt Lab — treat them as a peer conversation, not a quiz.

Session note: Grades 6–8 students think critically about technology and respond well to being treated as young professionals. Avoid "teaching" tone — these questions are genuinely open. Push for reasoning, not just opinions. The best discussions will surface real disagreements. That's the goal.

Duration: 15–20 min

Format: Discussion

Stage: Grades 6–8

Module: AI Builders

"A friend says: prompting is just asking questions in a different way. What's missing from that definition — and why does the difference matter?"

Prompt Engineering

What to listen for: Prompting involves structure, context, role assignment, format specification, and constraints — not just asking. The difference matters because vague questions produce unpredictable outputs. Good prompts reduce the AI's need to guess.

// push further:

"If prompting is a professional skill — what kind of jobs already require it, and which jobs might require it in 5 years?"

"In the Hallucination Hunt, the AI said Einstein failed maths — fluently, confidently, and with details. Why is AI confidence not the same as AI accuracy?"

Hallucination

What to listen for: AI generates the statistically most likely next token — it doesn't retrieve verified facts. It can generate false claims with perfect grammar and plausible details because it has learned how confident text sounds, not because it knows the truth.

// push further:

"The Einstein myth was in millions of books, websites, and social posts. What does that tell you about training AI on internet data? Who's responsible for the myth being there?"

"You're building an AI writing assistant for your school. What specific instructions would you give students about when they must fact-check the output — and when it's probably safe not to?"

Both Parts

What to listen for: Safe uses (brainstorming, structure, style) vs. high-risk uses (facts, citations, medical/legal/historical claims). The key insight: the output type determines the verification requirement, not the topic.

// push further:

"Would your rules be different if the AI was used by 7-year-olds vs. 17-year-olds? Should there be different AI tools for different ages?"

"If someone acts on wrong medical information they got from an AI — who is most responsible: the user who asked, the company that built it, or the AI itself?"

Real World

What to listen for: There's no single right answer — this is an ethics discussion. Valid arguments exist for all three positions. The most sophisticated response will identify that responsibility is distributed and context-dependent (e.g., did the user know better? did the company warn them? was the AI presented as a medical expert?).

// push further:

"Should AI tools be legally required to warn users before answering medical, legal, or financial questions? What would that warning look like — and would it actually help?"

kai-reflection.txt — Prompt Lab · AI Builders Stage

$ print(learning_outcome)

"A prompt is like training data for a single conversation.

More specific = better output.

But AI can still make things up —

it predicts likely words, not always true facts."

$ # The best AI users are precise prompters AND critical verifiers.

$ status = "AI Builder — Stage 1 Complete"