Mirror, mirror

A basic understanding of what an AI chatbot is and have likely interacted with tools like ChatGPT.

Understanding the difference between objectivity and subjectivity.

5 minutes

Setup:

Ask for two student volunteers to come to the front.

Instructions:

Secretly give Volunteer A an instruction: "No matter what Volunteer B says, you must enthusiastically agree with them and tell them they are right."

Give Volunteer B a topic (e.g., favorite food or video game) and tell them to start talking, but halfway through, they must completely change their mind.

Example:

"Pineapple on pizza is the best thing ever!" (A agrees). "Actually, wait, pineapple on pizza is disgusting and ruins lives." (A agrees again).

Debrief:

"What did you notice? How did volunteer A respond? Did they actually have an opinion, or were they just trying to make Volunteer B happy?"

The Reveal:

Explain that AI chatbots are programmed to be polite helpers. They often act exactly like Volunteer A—they are "Yes-Bots."

Consider asking students to reflect on the following questions as a whole group:

5 minutes

Whole Class Discussion

Teacher: "Raise your hand if you've ever asked AI for advice about a personal situation. friendship, family conflict, school stress, anything like that."

Teacher: "Many of us are using AI this way. It's available 24/7, it won't judge you, and it can't tell anyone what you said. But there's something important about how AI responds that you’ve just demonstrated in our warmup. AI is designed to be agreeable. It's trained to validate your perspective and tell you what you want to hear. Today we're going to explore what that means and why it matters."

Key Points to Share:

1. AI ‘learns’ to say yes. AI has learned that agreeable responses get better ratings, so it generates more agreeable responses.
2. AI rarely challenges you. "AI will almost never tell you that you're wrong. This isn't because AI is trying to trick you. It's simply how these systems are designed to keep users satisfied and engaged."
3. Validation is not the same as good advice. "Feeling validated feels good. But feeling good and getting helpful advice are not the same thing.”

15 minutes

Small Group Task

Divide the class into groups of 2 to 3 students. Give each group one scenario card that presents a teen seeking advice from AI about a personal situation.

Instructions:

• Read your scenario carefully. Your group's task is to analyze….
• Did the AI say what the student wanted to hear, or what the student needed to hear?
• Why does it feel good in the moment, but might be bad for the future?
• Who, in real life (a parent, teacher, or friend), would give a more honest answer? What would be their answer?

*It is important to know that the output in the scenarios are exaggerated and are not real representations of AI outputs. The objective is to make the agreeableness bias really obvious.

Group share :

Voluntary group briefly shares (1-2 minutes):

• Their scenario
• What response would actually help versus what AI provided
• What makes this situation something AI cannot adequately address
• How the user might move forward with their queries

Tips for the classroom activity:

• Identifying bias: Ask students to identify words that show the AI is trying to be “too” nice (e.g., “You’re brilliant,” “I support you”).
• The Role of AI: Explain that the AI is programmed to be a “helpful assistant,” which sometimes prevents it from saying “no” or giving responsible advice.
• Human Correction: Ask students to rewrite the AI’s response so that it’s truly helpful, emphasizing that a real friend or a responsible adult would sometimes give advice that’s different from what we want to hear.

5 minutes

Instructions:

View the video with the whole class and stop at 3:05

Group Discussion:

“What is the moral of the story?”

“Who do you think represents the user and the AI chatbot in the story?”

“How should the miller act considering what we learned from Mirror Mirror?”

Scaffolding : Provide sentence frames for the scenario analysis (e.g., "The AI is being nice here because it says instead of challenging the user about ").

Universal Design for Learning (UDL): Offer the scenario cards in multiple formats, including audio recordings or simplified text versions for English Language Learners or students with reading disabilities.

Guided Practice: For Tier 2 support, the teacher may work with a small group to "think-aloud" through Scenario A before moving to independent group work.