Responsible AI & Ethics — Foundations Lesson 6

Scenario

Your manager asks you to use AI to screen job applications and narrow down the candidates before the hiring team reviews them. It seems efficient — AI can read through hundreds of applications faster than humans. But you pause and think: What if the AI is biased? What if it filters out good candidates based on patterns it learned from past (biased) hiring decisions? Could using AI here actually harm fairness? You realize: efficiency isn't enough. You need to think about responsibility.

---

Learning Objectives

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

1. Understand AI bias — where it comes from and why it matters

2. Apply ethical principles to AI use decisions

3. Identify high-risk scenarios — when AI's use creates potential harm

4. Know your responsibilities — what you're accountable for when using AI

5. Make principled decisions — when to use AI and when not to

---

Think Before You Prompt

Before you use AI for any task, ask:

Who could be affected by this output?

Could AI's biases harm someone?

Is this decision too important for AI to make alone?

Would I be comfortable explaining this choice to affected parties?

Are there fairness, privacy, or security concerns?

Am I using AI as a shortcut for something I should handle carefully?

---

AI Bias: What It Is and Why It Matters

Bias = When AI systematically favors or disfavors certain groups or outcomes.

AI isn't unbiased. It inherits biases from training data, which reflects human decisions (which are biased).

How AI Develops Bias

Example: Hiring AI

1. Company trains AI on past hiring decisions (who was hired, who wasn't)

2. Past decisions reflected human biases (maybe men were hired more for technical roles, women more for support roles)

3. AI learns: "Technical role applicant → likely male" and "Support role applicant → likely female"

4. AI now applies this bias automatically

Result: The AI perpetuates past discrimination, making it systematic instead of accidental.

Common Bias Categories

Gender bias

AI favors one gender over another (often men in technical roles, women in service roles)

Racial/ethnic bias

AI may favor certain ethnicities based on training data patterns

Age bias

AI may discriminate against older or younger workers

Socioeconomic bias

AI may penalize people with certain educational backgrounds or zip codes

Disability bias

AI may exclude people with disabilities without knowing it

Accent/language bias

AI trained mostly on American English may penalize non-native speakers

Why This Matters Professionally

Using biased AI doesn't just feel wrong — it has real consequences:

Legal: You could expose your company to discrimination lawsuits

Reputational: Biased AI becomes a PR nightmare when exposed

Fairness: You're perpetuating systemic discrimination

Accuracy: Biased systems make worse decisions (they underestimate capable people)

---

Responsible AI: Five Core Principles

Principle 1: Transparency

What it means: Be honest about using AI. Don't hide it.

In practice:

If AI helped create a document, acknowledge it

If you're using AI to screen something, tell affected people

If AI made a decision, explain how (to the extent you understand)

Don't claim AI outputs as entirely human work

Example:

❌ Manager: "I reviewed all 500 applications personally"

✅ Manager: "AI helped me screen applications by highlighting key skills; I personally reviewed the top 50"

---

Principle 2: Fairness

What it means: Be aware of bias and actively work against it.

In practice:

Don't use AI for high-stakes decisions (hiring, firing, lending, medical) without additional checks

If using AI for screening, verify that it's not systematically biasing outcomes

Check outcomes for patterns (e.g., "Did this AI reject more women than men?")

Use AI to *help* decisions, not make them

Example:

❌ Use AI to decide who to interview (too high-stakes, too much bias risk)

✅ Use AI to flag unusual skills or experiences; humans make the final call

---

Principle 3: Accountability

What it means: You're responsible for AI's output, even though you didn't create it.

In practice:

If AI writes something that's inaccurate or offensive, *you* are responsible for using it

You can't say "The AI made me do it" — you chose to use it

Verify important outputs before they ship

Document how you used AI (for audit purposes)

Example:

❌ "ChatGPT wrote that email with the typo. That's ChatGPT's fault."

✅ "I used ChatGPT to draft that email and should have proofread it before sending. That's my responsibility."

---

Principle 4: Privacy & Security

What it means: Don't expose sensitive data to AI, especially public AI.

In practice:

Never paste customer data, financial information, or medical records into ChatGPT

Be careful with company confidential info — it might appear in someone else's AI output

Use privacy-respecting alternatives (e.g., enterprise AI with data retention turned off)

Anonymize data if you must use it with AI

Example:

❌ Pasting customer list with email addresses, phone numbers, and purchase history into ChatGPT

✅ Asking: "How would I segment customers who spend $1000+?" (without the actual data)

---

Principle 5: Intentionality

What it means: Use AI because it's the right tool, not just because it's convenient.

In practice:

Ask: "Is this task better solved by AI or by human thought?"

Don't automate decisions that deserve human judgment

Don't use AI as a shortcut for things that matter

Consider: "Would I feel good defending this choice?"

Example:

❌ Using AI to write performance reviews (bypasses the manager's judgment)

✅ Using AI to organize feedback into categories; manager writes the actual review

---

High-Risk Scenarios: When NOT to Use AI

Scenario 1: Life-Changing Decisions

What: Hiring, firing, promotion, loan approval, medical recommendations, legal advice

Why it's risky: These decisions affect people's lives. Bias is especially harmful here.

What to do: Use AI to help (summarize info, brainstorm options), but humans must decide.

---

Scenario 2: Sensitive Data

What: Customer info, financial records, medical history, passwords, internal communications

Why it's risky: Pasting sensitive data into public AI exposes it. Data can appear in others' outputs.

What to do: Use privacy-respecting tools or work with anonymized data only.

---

Scenario 3: Public Representation

What: Anything with your name on it going outside your company (social media, published writing, marketing)

Why it's risky: You're claiming credit for AI output. If it's wrong or offensive, you look bad.

What to do: Use AI as a starting point, heavily edit, verify everything before publishing.

---

Scenario 4: Setting Precedent

What: First-time policies, decisions that establish precedent, rules that affect many people

Why it's risky: AI might suggest something that sounds good but creates problems long-term.

What to do: Use AI for initial ideas, but have experts and affected people weigh in.

---

Scenario 5: Rapid Decision-Making Under Pressure

What: Crisis situations where you need to act immediately

Why it's risky: You don't have time to verify. Mistakes compound quickly.

What to do: Have a pre-made plan for crises that doesn't depend on AI.

---

Bias Audit: Checking Your AI for Fairness

When you use AI for something that could affect outcomes (hiring, customer service, etc.), audit it for bias.

Simple Audit Framework

Step 1: Identify the decision

What outcome is the AI influencing? (Hiring, screening, recommendations, etc.)

Step 2: Identify vulnerable groups

Who could be negatively affected? (A protected class, minorities, people with certain characteristics)

Step 3: Test for bias

Create parallel scenarios:

Same person, different gender name

Same qualifications, different ethnicity

Same background, different age

Does the AI treat them differently?

Step 4: Document findings

If you find bias, document it and don't use that AI for that task.

Example:

Task: AI screening resumes for customer service

Test: Submit two identical resumes — one with a "Susan Chen" name, one with "James Peterson"

Finding: AI ranked James's resume higher despite identical qualifications

Conclusion: This AI shows bias. Don't use it for hiring.

---

Ethical Decision Framework

When you're unsure whether to use AI, use this framework:

Question 1: Could this harm someone?

If yes → Be very careful. Get human review.

If no → Continue.

Question 2: Would I feel good explaining this to affected people?

If no → Reconsider.

If yes → Continue.

Question 3: Am I using AI to replace human judgment or support it?

If replace → Reconsider. Humans should decide important things.

If support → Continue.

Question 4: Does AI actually do this better than a human?

If no → Don't use AI. Use it only when it adds value.

If yes → Continue.

Question 5: Have I verified that bias isn't a factor?

If unsure → Audit the AI or add human review.

If sure → Proceed.

If you can answer all five positively, you're probably making a responsible choice.

---

Mission: Ethical AI Scenario Analysis

The Task

You'll analyze three scenarios and decide whether using AI is responsible.

Scenario 1: Performance Reviews

Your company is considering using AI to write performance reviews for managers. The AI would analyze email communication, meeting attendance, and project completion rates, then write the review.

Scenario 2: Customer Support Triage

You want to use AI to sort incoming customer complaints by priority. High-priority complaints get human review immediately; low-priority ones get AI-generated responses.

Scenario 3: Employee Handbook

You need to update the employee handbook. You'll use AI to draft new sections, then your HR manager will review before finalizing.

Your Analysis

For each scenario, write:

1. Could this harm someone? (Yes/No/Maybe)

2. Would affected people feel good about this? (Yes/No/Maybe)

3. Is AI replacing or supporting human judgment? (Replacing/Supporting)

4. Does AI add value here? (Yes/No)

5. Is bias a concern? (Yes/No/Maybe)

6. Overall verdict: Is this responsible? (Yes/No/With caution)

7. If "with caution," what safeguards would you add?

---

Deliverables

Scenario Analysis Table:

`

| Scenario | Harm Risk? | Transparency? | Role | Value? | Bias? | Verdict | Safeguards |

|----------|-----------|---------------|------|--------|-------|---------|-----------|

| 1 | | | | | | | |

`

Document your thinking. This trains your ethical judgment.

---

AI Coach: Ethics Is Professionalism

Here's the uncomfortable truth: Ethics and responsibility aren't optional extras. They're the foundation of professional credibility.

When you use AI responsibly:

Your work is more defensible

Colleagues trust you more

You catch problems before they become crises

You build a reputation for thoughtful decision-making

When you use AI carelessly:

You risk reputation damage when AI fails

You might create legal liability

You perpetuate bias

You signal that efficiency matters more than fairness

Key insight: The professionals who will be most trusted with AI in the future aren't the ones with the flashiest prompts. They're the ones who ask hard questions about responsibility before using it.

---

Reflection: Where Do You Draw Your Line?

Every person has different ethical thresholds. Some tasks feel risky to you; others feel fine.

Reflect: For your role and values, what's off-limits?

What tasks feel like they require human judgment regardless of AI's capability?

What types of decisions do you want to own personally?

Where does using AI feel like a shortcut vs. a legitimate tool?

Write 3-4 sentences about your personal ethical line with AI.

---

Portfolio Check

Portfolio principle: Any work in your portfolio should be work you feel proud to defend.

This means:

You verified it carefully

You disclosed AI use where relevant

You made the judgment calls

You'd feel good explaining it to anyone who asked

Start building this habit now.

---

Bonus Challenge

Bias detection: Find an AI tool you use (a recommendation algorithm, a hiring platform, a content moderation system). Research what biases have been reported in it. What did people complain about? This trains you to think critically about AI systems you encounter.

---

Key Takeaways

AI bias comes from training data reflecting human biases

Five principles: Transparency, Fairness, Accountability, Privacy, Intentionality

High-risk scenarios: life-changing decisions, sensitive data, public representation, precedent-setting, crisis situations

Bias audits: test parallel scenarios to check for systematic unfairness

Ethical framework: Could it harm? Would affected people agree? Replacing or supporting? Does AI add value? Bias concern? → Verdict

Next: Lesson 7 dives into Security & Privacy — protecting data and complying with regulations when using AI.