Skip to content

Teacher's Resources

Student

Video Overview

Launch Lesson Plan

Lesson Plan: The AI Safety Inspector

Topic: Responsible AI & Ethics

Duration: 60 Minutes

Target Group: Years 7–13 (Adaptable)

Learning Objectives

By the end of this lesson, students will be able to:

  1. Break down any AI system into 4 stages: Data, Weights, Input, and Output.
  2. Identify legal and ethical risks at each of these stages.
  3. Apply this "Safety Check" to their own Schools Challenge project.

Resources Needed

  • Whiteboard/Projector.
  • Printed copies of the "4 Stages of Responsible AI" one-pager (from the Student Pack).
  • Sticky notes.

1. The Hook: The "Magic" Machine (10 Mins)

Goal: Demystify AI and introduce the concept of "Hidden Costs."

Teacher Action: Draw a box on the board labeled "MAGIC BOX."

Discussion: Ask the class: "If I had a box that could paint any picture you asked for in 3 seconds, how does it do it?"

  • Student Answers: "It searches Google," "It draws it," "Magic."

The Reveal: Explain that AI isn't magic; it's a factory line. It doesn't "know" anything; it mimics what it has seen.

  • Ask: "If the box draws a picture of Mickey Mouse, who owns the picture? Disney? The Box? Or you?" (Let them debate; there is no single right answer, which highlights the problem).

2. Core Concept: The "Chef" Analogy (15 Mins)

Goal: Explain the 4 Stages using a non-tech metaphor.

Draw the following flow on the board. Use the Chef Analogy to explain the technical terms.

AI Stage The Chef Analogy The Risk (Ask Class)
1. Training Data The Library of Cookbooks. The chef reads millions of recipes to learn what "food" is. Did the Chef steal the recipes? Did they only read books about cake (Bias)?
2. Model Weights The Chef’s Brain. The chef doesn't remember every book, but they remember the patterns (e.g., "Sugar makes things sweet"). This is the "Product." Who owns the Chef's skills? (Intellectual Property).
3. Input Your Order. You walk in and say, "Make me a sandwich with my name written in ham." Did you give the Chef your personal info? Is it safe?
4. Output The Meal. The sandwich appears. What if the sandwich is poisonous? (Hallucinations). Who pays for the hospital?

3. Activity: The "Safety Inspector" (25 Mins)

Goal: Practical application of the 4-part model.

Instructions:

Split the class into small groups. Give each group one Scenario Card (below). They must act as "Safety Inspectors" and find the failure in the 4 stages.

Scenario A: The "Copycat" Art Generator

An app allows users to type "Paint me a hero," and it generates art that looks exactly like the style of a famous living artist, "Artist X." The app developers scraped Artist X's website to train the model.

  • The Flaw: Training Data. They used copyrighted work without permission/license.

Scenario B: The "Medical" Chatbot

A school installs a mental health chatbot. To use it, students must type in their full name and list their symptoms. The chatbot stores this data on a public server to "learn" for next time.

  • The Flaw: Input. Serious privacy violation. Students are inputting sensitive medical data that is not being protected.

Scenario C: The "Hiring" Manager

A company uses AI to scan CVs. It was trained on 10 years of past hiring data. It keeps rejecting female applicants because, in the past, the company mostly hired men.

  • The Flaw: Training Data / Model Weights. The data has historical bias, so the model "Weights" have learned that "Men = Good Employees."

Report Back: Ask each group to present their "Inspection Report" and how they would fix it.

4. Plenary: Applying it to the Challenge (10 Mins)

Goal: Link back to their project deliverables.

Teacher Action: Hand out the System Card Template (or point to it on screen).

Task: Look at Section 4 ("Risks & Limitations").

Closing Question:

"For your Council Challenge project, I want you to answer one question before you leave:"

  • If you are building an app for the Council, where does your Training Data come from?
  • Option A: We are training it ourselves (Safe, but hard).
  • Option B: We are using ChatGPT/Google (Easy, but you must admit you don't know the sources).

Homework: Teams must draft the "Responsible AI" section of their System Card using the 4-part model.

Teacher’s Cheat Sheet: Key Definitions

  • Scraping: Downloading data from the web automatically (often a legal grey area).
  • Hallucination: When an AI confidently states a fact that is completely false.
  • Black Box: When we don't know how the AI made a decision (common in "Closed" models).

TVAI Support

TVAI Volunteers may be available to run a 30 minute workshop with your teams but this is dependant on the availability and timing, reach out using the email address for this.