🛡️
← Owens AI Institute

AI Safety, Alignment & Ethics

Interpretability, red-teaming, alignment, and doing AI responsibly.

Elementary

Using AI Safely

Using AI safely starts with simple habits: AI is a tool, keep your private information private, and ask a trusted grown-up when something feels off.

  1. 1What Does Safe Mean?
  2. 2AI Is a Tool
  3. 3Keeping Your Secrets Secret
  4. 4Ask a Grown-Up
  5. 5Good Questions and Tricky Questions
  6. 6If Something Feels Wrong
  7. 7AI Time and Other Time
  8. 8Real Friends and AI Helpers
  9. 9Safety Detective
  10. 10A Safe AI Helper
  11. 11Lab: Safe Choices

AI Makes Mistakes

AI is powerful but not perfect. Kids learn that AI makes mistakes, can even make things up, and why it is smart to always check its answers.

  1. 1Nobody Is Perfect, Not Even AI
  2. 2Why AI Gets Things Wrong
  3. 3Checking AI's Answers
  4. 4AI Can Make Things Up
  5. 5Asking How Do You Know?
  6. 6Two Answers That Disagree
  7. 7Trust, But Check
  8. 8When to Believe AI
  9. 9Mistake Hunter
  10. 10Smart Checkers

Being Fair with AI

AI learns from examples — and if the examples are unfair, the AI can be unfair too. Kids learn what fairness means and how to make sure everyone is included.

  1. 1What Is Fair?
  2. 2AI Learns from Examples
  3. 3Unfair Examples, Unfair AI
  4. 4Everyone Should Be Included
  5. 5When AI Leaves Someone Out
  6. 6Different People, Different Needs
  7. 7Speaking Up for Fairness
  8. 8Fair AI Helps Everyone
  9. 9Fairness Checkers
  10. 10Fair and Kind

Honest and Kind with AI

Being honest and kind matters with AI too. Kids learn to do their own thinking, tell the truth about AI's help, and never use AI to be unkind.

  1. 1Being Honest with AI
  2. 2AI and Schoolwork
  3. 3Your Own Ideas Matter
  4. 4Telling the Truth About AI's Help
  5. 5Being Kind in What You Ask
  6. 6Not Using AI to Be Mean
  7. 7AI-Made Pictures and Stories
  8. 8Respecting Other People's Work
  9. 9Honesty Heroes
  10. 10Honest, Kind, and Proud

You're in Charge

People are always in charge of AI. Kids learn to think for themselves, ask good questions, and grow up as smart, safe, kind AI users.

  1. 1People Decide, AI Helps
  2. 2Your Brain Is Amazing
  3. 3Asking Good Questions
  4. 4Saying No to AI
  5. 5Telling Right from Wrong
  6. 6Helping Make AI Better
  7. 7AI Rules That Keep Us Safe
  8. 8Growing Up with AI
  9. 9Be the Boss
  10. 10A Smart, Safe, Kind AI User

Capstone Project: Make an AI Safety Poster

Create a poster teaching others the rules of safe AI use.

  1. 1Project: Make an AI Safety Poster

Middle

Why AI Safety Matters

AI safety is the work of making powerful AI helpful and not harmful. Students learn why it matters, the ways AI can go wrong, and who keeps AI safe.

  1. 1What Is AI Safety?
  2. 2Powerful Tools Need Care
  3. 3Ways AI Can Go Wrong
  4. 4Mistakes vs. Misuse
  5. 5Small Stakes and Big Stakes
  6. 6Who Keeps AI Safe?
  7. 7Safety Is a Habit
  8. 8AI Safety in Your Life
  9. 9Safety Stakes Sort
  10. 10Module Check: Why Safety Matters

Bias and Fairness in AI

AI can be biased when it learns from biased data. Students examine how bias enters AI, real-world examples, and how to make AI fairer.

  1. 1What Is Bias?
  2. 2How Bias Gets Into AI
  3. 3Biased Data, Biased AI
  4. 4Real Examples of AI Bias
  5. 5Who Gets Left Out
  6. 6Measuring Fairness
  7. 7Making AI Fairer
  8. 8Fairness Is Everyone''s Job
  9. 9Bias Detective
  10. 10Module Check: Bias and Fairness

Truth, Trust, and Misinformation

AI can produce convincing falsehoods. Students learn about hallucination, deepfakes, misinformation, and how to evaluate what they see.

  1. 1Can You Trust an AI?
  2. 2Hallucination: When AI Makes Things Up
  3. 3Deepfakes and Synthetic Media
  4. 4Spotting AI-Generated Content
  5. 5Misinformation and How It Spreads
  6. 6Checking Sources and Facts
  7. 7Healthy Skepticism
  8. 8Being a Responsible Sharer
  9. 9Fact-Check Challenge
  10. 10Module Check: Truth and Trust
  11. 11Lab: Ai Answer Checker

The Alignment Problem

The alignment problem is getting AI to do what we actually want. Students explore intent, specification gaming, human values, and keeping humans in control.

  1. 1What Do We Want AI to Do?
  2. 2The Alignment Problem
  3. 3When AI Takes a Goal Too Literally
  4. 4Reward and What Gets Measured
  5. 5Human Values Are Complicated
  6. 6Teaching AI What We Value
  7. 7Keeping Humans in Control
  8. 8Why Alignment Gets Harder as AI Gets Smarter
  9. 9Align the Robot
  10. 10Module Check: The Alignment Problem

Living Ethically with AI

Living ethically with AI means privacy, honesty, consent, and respect. Students build a practical ethic for using AI as good digital citizens.

  1. 1Ethics: Doing the Right Thing
  2. 2Privacy and Your Data
  3. 3AI and Schoolwork: Honest Use
  4. 4Consent and Other People's Data
  5. 5AI, Art, and Credit
  6. 6Kindness and Respect with AI
  7. 7Your Rights with AI
  8. 8Being a Good Digital Citizen
  9. 9Ethics Case Studies
  10. 10Module Check: Living Ethically with AI

Capstone Project: Run a Bias Audit

Investigate an AI system or scenario for unfairness.

  1. 1Project: Run a Bias Audit

High School

The Landscape of AI Risk

AI risk is a serious field of study. Students build a taxonomy of misuse, accident, and structural risks, and reason rigorously about uncertainty and stakes.

  1. 1Why AI Risk Is Worth Studying
  2. 2A Taxonomy of AI Risks
  3. 3Misuse Risks
  4. 4Accident Risks
  5. 5Structural and Systemic Risks
  6. 6Near-Term vs. Long-Term Risk
  7. 7Reasoning About Uncertainty and Stakes
  8. 8Risk, Benefit, and Tradeoffs
  9. 9Build a Risk Assessment
  10. 10Module Check: The Risk Landscape

The Alignment Problem in Depth

The alignment problem, in depth. Students study outer and inner alignment, reward hacking, goal misgeneralization, scalable oversight, RLHF, and corrigibility.

  1. 1What Alignment Really Means
  2. 2Outer Alignment: Specifying Goals
  3. 3Reward Hacking and Specification Gaming
  4. 4Inner Alignment and Mesa-Optimization
  5. 5Goal Misgeneralization
  6. 6Scalable Oversight
  7. 7RLHF and Its Limits
  8. 8Corrigibility and Control
  9. 9Design an Alignment Approach
  10. 10Module Check: The Alignment Problem
  11. 11Lab: Reward Hacking

Bias, Fairness, and Justice

Algorithmic fairness, rigorously. Students study formal fairness definitions, the impossibility results, real harms, justice and power, auditing, and mitigation.

  1. 1Algorithmic Bias, Rigorously
  2. 2Sources of Bias in the ML Pipeline
  3. 3Formal Fairness Definitions
  4. 4The Impossibility of Satisfying All Fairness Criteria
  5. 5Real-World Cases and Harms
  6. 6Fairness, Justice, and Power
  7. 7Mitigation and Its Limits
  8. 8Auditing AI Systems
  9. 9Audit a System for Fairness
  10. 10Module Check: Bias, Fairness, Justice

Interpretability, Robustness, and Control

Trustworthy AI demands interpretability and control. Students study the black-box problem, adversarial robustness, evaluation, red-teaming, and containment.

  1. 1The Black Box Problem
  2. 2Interpretability: Opening the Box
  3. 3Adversarial Examples and Attacks
  4. 4Robustness and Distribution Shift
  5. 5Evaluating AI Systems for Safety
  6. 6Red-Teaming and Stress-Testing
  7. 7Monitoring Deployed AI
  8. 8Control, Shutdown, and Containment
  9. 9Red-Team an AI System
  10. 10Module Check: Interpretability and Control

Governing AI

AI must be governed. Students study laws, standards, accountability, the AI safety research field, global coordination, and their own role.

  1. 1Why AI Needs Governance
  2. 2Who Governs AI?
  3. 3Laws and Regulations
  4. 4Standards, Audits, and Accountability
  5. 5The AI Safety Research Field
  6. 6Global Coordination and Competition
  7. 7Ethics Frameworks for AI
  8. 8Your Role in AI Safety
  9. 9Draft an AI Policy
  10. 10Module Check: Governing AI

Capstone Project: Write an AI Safety Case

Analyse a system's risks and argue for safeguards.

  1. 1Project: Write an AI Safety Case