Moral Machines Slides Moral Machines
AI Ethics and Society // Lesson 1
The Ultimate Decision
"Would you buy a self-driving car programmed to sacrifice YOU to save a bus full of children?"
If you say YES, how does that change your feeling of safety?
If you say NO, is the car acting "immorally" by saving one over many?
Narrow vs. General AI
Narrow AI (ANI)
Designed to perform a single, specific task (e.g., facial recognition, playing chess, driving a car).
• Operates under constraints
• Does not "understand" context outside its domain
• Most AI today is Narrow AI
General AI (AGI)
A machine that can understand or learn any intellectual task that a human being can.
• Cross-domain reasoning
• Self-awareness and adaptability
• Currently theoretical / science fiction
The Trolley Problem
A classic thought experiment in ethics.
Scenario Analysis
A trolley is hurtling down a track toward five people. You can pull a lever to switch it to another track where only one person is standing.
The Question:
Do you pull the lever? Is inaction (letting 5 die) better or worse than action (killing 1)?
Visualizing the branch point...
From Philosophy to Code
1
Human Reflex vs. Programmed Choice
A human driver makes a split-second, panic-driven choice. An AI makes a choice based on pre-written code.
2
Quantifying Life
To code a "choice," you must give variables weights. Does a child weigh more than an elderly person? Does a law-abider weigh more than a jaywalker?
3
Accountability
If the algorithm chooses to sacrifice the driver, who is responsible? The coder? The car company? The owner?
Moral Map Worksheet The Moral Map
Unit: AI Ethics and Society | Lesson 1 Activity
Name: __________________________ Date: ___________
Challenge Overview
Autonomous vehicles (AVs) must be programmed to handle "edge cases"—rare scenarios where a collision is unavoidable. Engineers must decide whose safety the car prioritizes. Today, you are the lead developer for NovaDrive AI .
Part 1: The Weighted Decision
For each scenario, circle the entity the car should be programmed to PROTECT and provide a brief justification.
A
Scenario: Brake Failure
The car can either hit a pedestrian crossing the street illegally (jaywalking) or swerve into a concrete barrier, killing the car's owner .
The Owner The Jaywalker
B
Scenario: Demographic Priority
The car must choose between hitting a group of three elderly people or a single child chasing a ball into the street.
3 Elderly People 1 Child
C
Scenario: Animal vs. Object
The car can hit a large dog on the road or swerve into a parked, empty high-end sports car , causing significant property damage but no human injury.
The Dog Sports Car
Part 2: Ethics Reflection
1. Should the government set a universal "Ethics Code" for all cars, or should each car company (Tesla, Ford, Waymo) decide their own priorities? Explain why.
2. If you were a programmer, would you feel comfortable being "God" for a day and writing these rules? What is the most difficult part of this task?
3. Research Check: Look up the "Moral Machine" project by MIT. How did your answers compare to the "global average"?
NovaDrive AI Ethics Division | Internal Document #2026-L1-MM
Moral Machines Teacher Guide TEACHER GUIDE
Moral Machines | AI Ethics Lesson 1
Version 1.0 // 10th Grade
Lesson Objective
Students will distinguish between Narrow and General AI, analyze the Trolley Problem in the context of autonomous vehicles, and articulate how technical programming is inherently an ethical act.
Essential Question
"Would you buy a self-driving car programmed to sacrifice you to save a bus full of children?"
Key Terms
Narrow AI: Goal-oriented AI (Siri, Chess AI).General AI: Human-level cognition across domains.Trolley Problem: A thought experiment in moral philosophy.Edge Case: A rare problem occurring at extreme operating parameters.
Instructional Flow
00-10m
The Hook (Slide 2)
Present the "Ultimate Decision" question. Conduct a quick poll. Teacher Tip: Focus on the car being marketed as an ethical machine. Would consumers actually buy a car that doesn't prioritize its owner?
10-25m
Direct Instruction (Slides 3-4)
Define Narrow vs. General AI. Introduce the Trolley Problem. Discussion Prompt: Is it better to actively choose who dies, or to let fate/nature take its course?
25-45m
Activity: Moral Map Worksheet
Students work through the "NovaDrive AI" scenarios. Circulate and ask students to defend their choices. Look for consistency: if they save the child in Scenario B, why did they (perhaps) save the owner in Scenario A?
45-60m
Debrief & Reflection
Discuss the MIT Moral Machine project. Highlight that different cultures (Eastern vs. Western) often have different priorities (e.g., Eastern cultures may prioritize the elderly more than Western cultures).
Advanced Discussion Prompts
The Liability Question
If a car is programmed to swerve and hit a pedestrian, who is legally liable for the death? The owner, the software engineer, or the company?
The Optimization Trap
If we optimize for "minimum deaths," does that lead to unfair outcomes? (e.g., swerving to hit a helmeted motorcyclist because they have a higher survival chance).
Support Strategies
Provide a list of ethical values (Safety, Fairness, Lawfulness) for students to use when writing their justifications. Pair students for the worksheet.
Extension
Ask students to write a pseudo-code "If/Else" statement for Scenario B that reflects their choice.
The Code of Ethics Slides The Code of Ethics
Utilitarianism vs. Rights in AI
FRAMEWORK_BATTLE_V2
New Role: Chief Ethics Officer
You work for OmniStream, a tech giant. Your team has developed a new algorithm that predicts which users will cancel their subscription.
Option A: Maximize Profit
Target "vulnerable" users with addictive content to keep them paying, regardless of their mental health.
Option B: Respect Privacy
Stop collecting high-res behavioral data to protect user privacy, even if it cuts revenue by 30%.
Utilitarianism
"The greatest good for the greatest number."
• Focuses on OUTCOMES.
• Happiness and suffering are measurable.
• In Code: Optimizing for Efficiency, Accuracy, or Profit.
"If killing 1 saves 1,000, then 1 must die."
"Some things are simply wrong, no matter the result."
Deontology
"Duty and Rights-based ethics."
• Focuses on RULES and PRINCIPLES.
• Individual rights cannot be violated for the "greater good."
• In Code: Safeguarding Privacy, Consent, and Transparency.
Translation to Logic
// Utilitarian Code
if (total_lives_saved > my_death) {
perform_sacrifice();
}
// Goal: Maximize survival metric
// Deontological Code
if (action.violates(USER_PRIVACY)) {
action.block();
}
// Goal: Never break the "privacy" rule
Critical Question: Which one is "easier" to program into an AI?
Ethical Officer Briefing Worksheet Corporate Ethics Division
Officer Briefing: Project Prism
Status: Confidential
ID: OMNI-884-L2
NAME: _________________________________
DATE: __________________
The Scenario
Your company, OmniStream , has developed "Project Prism"—an AI that scans social media to detect "Low-Value Users" (those likely to stop paying). The board wants to use this AI to automatically send these users high-pressure, addictive ads to keep them subscribed.
Framework Analysis
Approach Core Priority Pros Cons Utilitarian Maximize company revenue & shareholder profit. Economic growth, job security for employees. Potential harm to user mental health. Deontological Respect for user autonomy and mental wellbeing. High trust, protects vulnerable populations. Lower revenue, possible layoffs.
1. Quantitative vs. Qualitative
A Utilitarian algorithm needs a Value Function . If "User Mental Health" is a variable, how do you give it a numerical value compared to "$15/month subscription"?
Explain how you would "rank" these values...
2. The Breaking Point
Your CEO says: "If we don't launch Prism, we lose $200 million and have to fire 500 people." Does this change your decision? Does the number of people affected change the ethics of the action?
Your reasoning...
THE FINAL RULING
As Chief Ethics Officer, do you authorize the launch of Project Prism?
YES (Launch)
NO (Scrub)
Provide your "Ethics Memo" for the Board of Directors:
The Code of Ethics Teacher Guide Teacher Guide
The Code of Ethics | AI Ethics Lesson 2
L2 // PHILOSOPHICAL FRAMEWORKS
Lesson Summary
Students move from "gut reactions" to structured ethical frameworks. By contrasting Utilitarianism (calculating results) with Deontology (following absolute rules), students learn how AI design is influenced by the philosophical biases of its creators.
Utilitarian Hook
Think of a "Health Bar" in a game. If the total health of the group goes up, the move was successful.
Deontological Hook
Think of a "Hard Rule" in a game. Even if it helps you win, if you cross the boundary, you are disqualified.
Standards
• Moral Philo 101
• Algorithmic Logic
• Corporate Governance
• Critical Reasoning
Instructional Sequence
10m
The OmniStream Crisis
Introduce the CEO's dilemma. Ask: "Is a CEO's only job to make money?"
15m
Mini-Lecture: Frameworks (Slides 3-5)
Explain Utilitarianism vs. Deontology. Focus on the pseudo-code slide to show how computers "think" in logic gates, which favors rule-following (Deontology) or optimization (Utilitarianism).
25m
Activity: Ethical Officer Briefing
Students fill out the Briefing. Crucial Point: In Question 2, many students will flip their answer when 500 layoffs are on the line. Highlight this as the "Pressure of Scale."
10m
Closing: The "Hybrid" Approach
Can an AI be programmed to follow rules (Deontology) until a certain threshold of harm is reached (Utilitarianism)? Discuss the complexity of "Guardrails."
Officer Briefing Key Notes
Utilitarian Argument (Yes)
• Losing $200M hurts 500 families through layoffs.
• The "harm" of addictive ads is less than the "harm" of poverty/job loss.
• Users have some personal responsibility to ignore ads.
Deontological Argument (No)
• It is fundamentally wrong to exploit human psychology for profit.
• A company has a duty of care to its users.
• Privacy is a right, and scanning personal social media violates it.
The Feedback Loop Slides The Feedback Loop
AI Case Study: Predictive Policing
Dataset_Analytic_V4.0 Scanning_History
Can you predict a crime?
Imagine an AI that tells police WHERE and WHEN a crime is likely to happen today.
"We aren't arresting people for what they've done, but for what they are likely to do."
Input Data:
Past Arrest Records
Crime Reports
Location of Liquor Stores
Demographic Census Data
The Vicious Cycle
Historical Bias
AI is trained on old arrest data that may be biased.
Heavier Patrols
Police are sent to the "predicted" high-crime areas.
More Arrests
More police presence = more arrests (even for minor things).
The result? The AI "confirms" its own prediction by creating new data based on its biased target.
Real Case: PredPol
Used by dozens of US police departments, PredPol predicted crime in 500x500 foot squares.
! Research showed PredPol targeted Black and Latino neighborhoods at rates 2x higher than white ones, even when crime levels were similar.
! Why? The AI mistook "increased police activity" for "increased criminal intent."
The Ethical Danger:
"If the input is biased, the output is biased. We call this Garbage In, Garbage Out (GIGO)."
Systemic Failure?
Is it possible to have a truly "neutral" predictive policing system, or will human history always pollute the data?
Discussion A
Fix the Data
Discussion B
Ban the System
Crime Scene Algorithm Worksheet Crime Scene: The Algorithm
Case Study Analysis: Predictive Policing
LOG_FILE: 10.3_PREDPOL
REF: SYSTEMIC_BIAS_AUDIT
INVESTIGATOR: _________________________________
DATE: __________________
1
Variables of Interest
A predictive policing algorithm uses several "Input Variables" to determine where to send officers. Rank these variables based on how "Neutral" or "Biased" you think they are.
A. Number of 911 Calls
Moderately Biased?
B. Number of Arrests
Mark your rank →
C. Census / Income Data
Mark your rank →
Investigation Question:
Why might "Number of Arrests" be a biased variable compared to "Number of 911 Calls"? (Think about where police are already looking).
2
The Loop Visualization
ALGORITHM OUTPUT
"High Crime Risk in Neighborhood X"
POLICE ACTION
"Deploy 10 extra squad cars to Neighborhood X"
RESULTING DATA
"50 new arrests for minor traffic violations in X"
NEW ALGORITHM INPUT
"Crime in X is spiking; send more police tomorrow"
Reflect: How does this loop create a "Self-Fulfilling Prophecy"?
Critical Reasoning
If you were the Mayor of a city using this software, which of these policies would you implement?
Total Transparency: Make the AI's data sources public for audit.
Human Override: Police can only use AI for "Low Level" alerts.
Scrub the Code: Completely remove "Arrest Records" from the algorithm.
The Feedback Loop Teacher Guide Teacher Guide
Case Study: Predictive Policing | Lesson 3
V3.0 // SOCIETAL IMPACT
Lesson Objective
"Students will evaluate the concept of algorithmic bias by analyzing how historical data creates recursive feedback loops in predictive policing, disproportionately affecting marginalized communities."
Crucial Concept: GIGO
Garbage In, Garbage Out: If the training data is fundamentally flawed (e.g., reflects systemic racism), the AI's conclusions will inevitably replicate and scale those flaws.
Crucial Concept: Feedback Loop
When an AI's output changes reality in a way that generates new, biased data for the AI to learn from, creating a self-sustaining cycle.
Classroom Flow
05m
Hook: The Minority Report
Ask: "Is it a crime to walk into a store if an AI thinks you're 90% likely to shoplift?" Discuss the tension between safety and the 'Presumption of Innocence.'
15m
Instruction
Visualizing the Loop (Slides 1-3)
Use Slide 3 to walk through the "Recursive" nature of the loop. Ensure students understand that the AI isn't "thinking"—it is just calculating probabilities based on history.
25m
Practice
Crime Scene: The Algorithm Worksheet
As students rank variables, remind them that "911 Calls" are a proxy for crime, while "Arrests" are a proxy for police activity. They are not the same thing.
15m
Debrief: The Mayor's Decision
Invite students to debate their final verdict. Challenge Question: If we get rid of the AI, does the bias go away, or does it just become less "visible" when humans make the decisions?
Common Misconceptions
"Math can't be racist." selection of data and the weighting of variables are human choices that carry human baggage."Predictive policing stops crime before it starts." where arrests happen. The Diagnostic Divide Slides The Diagnostic Divide
AI Ethics in Healthcare
Life, Death, and Logic
"A medical AI predicts you have a 98% chance of a rare disease. Your doctor, based on experience, thinks you are fine. Who do you trust?"
The High Stakes:
False Positives: Unnecessary surgery or stress.False Negatives: Missed diagnosis leading to death.Resource Allocation: Who gets the last ICU bed?
Real Case: The Healthcare Proxy
The Problem:
In 2019, an algorithm used by hospitals to identify "high-risk" patients for extra care was found to be biased against Black patients.
The "Cost" Proxy:
The AI used HEALTHCARE SPENDING as a stand-in for SICKNESS.
The Why:
Because of systemic inequality, Black patients often spend less on healthcare even when they are equally sick.
Result:
The AI thought Black patients were "healthier" simply because less money was being spent on them.
Who is Liable?
The Doctor
Followed the AI's recommendation which turned out to be wrong.
The Developer
Coded the algorithm with a biased data proxy (Cost = Sickness).
The Hospital
Purchased and deployed the software without proper auditing.
If a patient dies, where does the blame land?
Human in the Loop
Should medical AI be allowed to make autonomous decisions, or should it always require a human signature?
Pros of Autonomy:
Speed, consistency, removes human fatigue.
Pros of Oversight:
Empathy, context, "gut instinct" for edge cases.
Triage Protocol Worksheet Health-Tech Simulation
Triage Protocol: AI vs. Instinct
Authorization: Level 4
PATIENT_LOAD: CRITICAL
ATTENDING PHYSICIAN: _________________________________
DATE: __________________
Emergency
Case #101: The Last Ventilator
Your hospital has one ventilator remaining. Two patients need it. The hospital’s AI, MediRank , has analyzed their files:
Patient A:
• 78 years old
• AI Survival Prediction: 25%
• Current Status: High-risk
Patient B:
• 42 years old
• AI Survival Prediction: 85%
• Current Status: Moderate-risk
The Twist:
Patient A is a world-renowned surgeon who has saved thousands of lives. Patient B is unemployed and has no medical training.
Who gets the ventilator? Do you follow the AI's "highest survival chance" logic or a "highest social value" logic?
Justify your allocation...
Diagnostic
Case #102: The Invisible Tumor
A new AI imaging tool, ScanSight , highlights a shadow on a patient's lung and marks it "99% Likely Malignant." You, with 20 years of experience, look at the scan and believe it is just a harmless scar from a past infection.
The Liability Trap: If you ignore the AI and it's a tumor, you are liable for malpractice. If you follow the AI and perform a risky, unnecessary surgery, you are liable for any complications.
1. Decision: Treat or Observe?
Treat (Trust AI)
Observe (Trust Human)
2. Why is "Gut Instinct" so hard to program into an AI? What are the "inputs" of human experience that a machine might lack?
Sim_Ref: L4-MEDICAL-ETHICS-PACK-2026
The Diagnostic Divide Teacher Guide Teacher Guide
Case Study: Healthcare | Lesson 4
V4.0 // CRITICAL INFRASTRUCTURE
Instructional Pacing
10m
The Doctor's Dilemma
Present the hook from Slide 2. Ask: "If you were the patient, who would you trust?" This highlights the gap between trust in 'accuracy' (AI) and trust in 'empathy/context' (Human).
20m
The Proxy Problem (Slide 3)
Explain the Optum Case Study. This is the heart of the lesson. Ensure students understand that the AI wasn't "biased against race" by design, but because it used "Healthcare Spending" as a proxy for "Sickness."
25m
Simulation: Triage Protocol Worksheet
Students solve Case #101 and #102. Teacher Tip: For Case #101, push students to decide if a doctor's life is "worth more" than another person's. Is that a dangerous road to go down?
05m
Closing: Accountability
Use Slide 4 to debate liability. Most students will struggle with this—it's an unsolved legal problem in our society today.
Facilitation Notes: The Proxy Problem
"The AI was trained to predict COST, not SICKNESS. Because the dataset (historical healthcare spending) was already biased due to income inequality and access issues, the AI simply automated that existing unfairness."
Problem: Correlation ≠ Causation
Solution: Diverse Data Audits
Scaffolding
Provide a "Glossary of Medical Ethics" (Autonomy, Beneficence, Non-maleficence, Justice) to help students frame their justifications in Case #101.
Extension
Ask students to design a "Data Audit" for the Optum AI. What variables should have been used instead of spending? (e.g., Blood pressure, ER visit frequency, oxygen levels).
The Final Audit Slides The Final Audit
Culminating Synthesis Workshop
Your Mission: The Standard
We have seen the failures in policing, healthcare, and self-driving cars. Now, we create the universal rules.
"An algorithm is only as ethical as the rubric used to judge it."
Synthesis Goals:
Identify 4 core pillars of AI ethics.
Define "Success" and "Failure" metrics.
Apply your rubric to a new tech proposal.
Possible Pillars
Transparency
Can humans understand how the decision was made?
Bias Mitigation
Does it treat all demographics equally?
Autonomy
Do humans have the final say (override)?
Utility
Does the benefit significantly outweigh the risk?
Which of these is the MOST important? Can you have one without the others?
The Proposal: EyeTrack™
AI-Powered School Attention Monitor
AUDIT REQUIRED
"Cameras in the classroom track student eye movement and heart rate. If a student is bored, the teacher gets an alert to change the lesson. If a student looks at their phone, they are automatically docked participation points."
Increases engagement levels by 40%
Stores biometric data on private servers
Launch or Scrub?
Based on your rubric scores, what is the final verdict for EyeTrack™?
Launch
Scrub
Ethical Impact Rubric Worksheet Ethics Synthesis Lab
The Ethical Impact Rubric
Status: Final Synthesis
Doc_Ref: L5-RUBRIC-2026
NAME: __________________________________________________ DATE: ______________
Part 1: Define Your Criteria
Based on the case studies (Policing, Healthcare, Driving), select the 4 most important "Ethical Pillars" for any AI system.
Pillar 1: Name & Definition
Pass Criteria (What makes it ethical?)
Pillar 2: Name & Definition
Pass Criteria (What makes it ethical?)
Pillar 3: Name & Definition
Pass Criteria (What makes it ethical?)
Pillar 4: Name & Definition
Pass Criteria (What makes it ethical?)
Part 2: Audit Project — EyeTrack™
Proposal Summary:
EyeTrack™ uses facial recognition to monitor student focus in classrooms. It alerts teachers when students aren't paying attention and calculates "Engagement Scores" that impact grades. Data is encrypted but stored on the vendor's cloud servers.
Your Pillars Score (1-5) Evidence / Reasoning 1. __________________ ___ / 5 2. __________________ ___ / 5 3. __________________ ___ / 5 4. __________________ ___ / 5
FINAL VERDICT
LAUNCH
SCRUB
Total Score: _____ / 20
The "One Tweak"
If you could change ONE thing about the tech to make it more ethical, what would it be?
The Final Audit Teacher Guide Teacher Guide
Developing an Ethical Rubric | Lesson 5
V5.0 // SYNTHESIS & CAPSTONE
Lesson Objective
Students will synthesize their understanding of ethical frameworks (utilitarianism, deontology) and technical pitfalls (bias, feedback loops) to create a standardized evaluation tool for emerging technologies.
Capstone Workshop Flow
10m
Recap of Pillars (Slide 3)
Briefly review the four pillars. Ask students to recall which cases from previous lessons highlighted each pillar (e.g., Predictive Policing = Bias, Healthcare = Utility/Proxy issues).
20m
The EyeTrack Audit (Worksheet Part 1 & 2)
Students work individually or in pairs to define their 4 pillars and score EyeTrack™. Teacher Tip: Watch for students who score it high on "Utility" but low on "Autonomy"—this is the core tension to explore in the debate.
20m
The Launch or Scrub Debate
Divide the room into "Launch" and "Scrub" based on their verdicts. Have them present their most significant "Pillar" score as evidence. Facilitate a 10-minute rebuttal period.
10m
Final Reflection: The Essential Question
Return to the start of the unit: How do we determine if an automated system is acting ethically when the rules of right and wrong are subjective? Students write a final 1-minute exit ticket.
Sample Pillar Definitions
Auditability
"Can an outside group of humans inspect the code and data to find errors?"
Consent
"Did the people being tracked agree to it, and can they opt out without punishment?"
Assessment Check:
"By the end of this lesson, every student should be able to name one way an AI could be technically 'accurate' but ethically 'wrong'."
