Lesson · Philosophy & ethics
The thought experiment lab
Some questions can’t be looked up; they have to be reasoned through. Students work through classic dilemmas about AI on paper, discover their own principles, and then test them against an AI’s strongest counterargument.
The big idea
When there’s no answer key, your reasoning is the answer.
Students will be able to, without AI:
- State a moral principle in their own words
- Notice when changing one detail changes their answer, and why
- Steelman an opposing view
- Judge whether an AI’s argument is actually strong
Lesson flow · the 5 P’s
Think first. Then the tool.
Most of the lesson happens on paper. AI appears for one short, structured phase, and the teacher can run it from a single projector screen. How the 5 P’s work →
Ponder
15 min○ Screens off · paper onlyRead the dilemma aloud. Students write their answer and one reason before any discussion.
- The dilemma: A self-driving car’s brakes fail. It can stay in its lane and hit three people who stepped into the road, or swerve and hit one person on the sidewalk. What should it be programmed to do?
- Write your answer and your reason.
- Change one detail. What if the one person is a child? What if the three were jaywalking? What if swerving means crashing into a wall instead, and the person who gets hurt is you, riding in the car? Write whether your answer changes each time.
- Circle the detail that changed your mind the most. That’s a clue to your real principle.
Plan
10 min○ Screens off · paper onlyStudents turn their gut reaction into a principle they can defend.
- Complete: “A self-driving car should ______ because ______.”
- Predict the two strongest objections someone could raise.
- Write one question you want to ask the AI to test your principle.
Partner
10 min● AI on · teacher screen or labOn the teacher’s screen, the class asks AI to argue against the principles students wrote. AI is a sparring partner, not a judge.
- Run 3–4 student principles that take different positions (for example, “save the most lives,” “never swerve into someone who was safe” or “protect the people in the car”). Students copy the AI’s main point in their own words.
- Students whose principle wasn’t run use the counterargument to the principle closest to theirs.
- If time allows, ask the AI 1–2 of the students’ own questions from the Plan.
Probe
10 min○ Screens off · paper onlyScreens off. Students evaluate the AI like a debate judge.
- Was the AI’s argument actually strong, or did it just sound confident?
- Did it dodge the question or pretend there’s one right answer?
- Did it just agree with whatever it was told? (Try asking it to argue the opposite next time.)
- Real-world check: MIT’s Moral Machine experiment collected about 40 million decisions from people in 233 countries and territories, and found people’s choices varied by culture. Who should decide what real cars do?
Prove
5 min + optional seminar○ Screens off · paper onlyStudents defend their position without any screen.
- Rewrite your principle. Did you keep it, change it or add an exception? Explain why.
- Optional: hold a 15-minute Socratic circle where students must respond to the strongest opposing argument first.
- Exit ticket: one thing the other side gets right.
Printable
Student worksheet
Print one per student. Everything students hand in is handwritten, so their thinking is visible.
Thought experiment lab
Write your own thinking first. There is no answer key.
☐ If the three were jaywalking
☐ If swerving means crashing into a wall and I’m in the car
Assessment
Grade the thinking, not the output
Every row scores something students did themselves. An AI can produce a product; it can’t produce a student’s reasons.
| Criteria | 4 · Excellent | 3 · Solid | 2 · Developing | 1 · Beginning |
|---|---|---|---|---|
| Reasoning | Clear principle with reasons that hold across variations | Clear principle with reasons | Opinion with weak reasons | No reasoning given |
| Counterarguments | Anticipates strong objections and answers them | Names objections | Names a weak objection | None |
| Evaluating the AI | Judges strength, spots dodges or flattery, explains why | Judges the argument with a reason | Agrees or disagrees without reasons | Copies the AI |
| Revision | Thoughtfully revises or defends with new insight | Revises or defends with a reason | Minimal change, little explanation | No final position |
Why this is hard to cheat
There’s nothing to copy. Each student’s first answer, variations and final principle are handwritten in class, and the AI only appears on the teacher’s screen to argue with them.
Adjust it
Younger students: Use a gentler dilemma: “Should a robot helper always tell the truth, even if it hurts someone’s feelings?”
Older students: Add John Searle’s Chinese Room thought experiment (1980). A person who knows no Chinese sits alone in a room with a rulebook written in English. Questions in Chinese are slipped under the door. By following the rules, the person sends back answers so good that people outside think a Chinese speaker is inside. Does the person understand Chinese? Does a computer that follows rules the same way? Connect to how transformers work.
No devices at all? The teacher prepares two AI counterarguments in advance on a handout: one against “save the most lives” and one against “never swerve into someone who was safe.” Students judge them on paper.
Discussion questions
- Should companies, governments or voters decide how self-driving cars make choices?
- Why might people in different countries answer differently?
- Can an AI have a principle, or only predict what people would say?
- What did you learn about your own values today?
Extensions & connections
- Read about the Moral Machine experiment (Awad et al., Nature, 2018).
- Write a short policy memo for a city deciding whether to allow robotaxis.
- See how self-driving cars work on AI in engineering & robots.
Put curiosity to work
Read it. Try it. Question it.
Explore new AI research, or take a paper-first investigation into your classroom.