AI & Digital Ethics Grade 7: AI in Healthcare Finance and Education

Study Guide: AI in Healthcare, Finance, and Education (Grade 7)

1. The Driving Question

"If a robot doctor can diagnose diseases faster than a human, should it make the final call? And if an AI tutor knows exactly how you learn best, is it fair that only some kids get access to it?" AI is already making decisions in hospitals, banks, and classrooms—but who’s responsible when it gets things wrong? And how do we make sure these tools help everyone, not just the people who can afford them?

2. The Core Idea — Built, Not Listed

Imagine you’re at a hospital where an AI scans X-rays and flags potential tumors. The AI is really good at spotting patterns humans might miss—like how your phone’s face ID recognizes you even if you’re wearing a hat. But here’s the catch: the AI was trained on thousands of X-rays from mostly wealthy hospitals in big cities. If you live in a rural area, the AI might not recognize rare conditions common in your community because it’s never seen them before. That’s a bias—when the AI’s "knowledge" is limited by the data it was given.

Now think about your school’s math app. It adjusts problems based on how fast you answer, like a personal coach. But what if the app’s algorithm decides some students are "too slow" and stops giving them challenging work? That’s reinforcement learning—where AI improves by repeating actions that get "rewards" (like correct answers), but it can accidentally reinforce unfair assumptions.

In finance, AI decides who gets a loan. If it’s trained on past loan data where certain neighborhoods were denied, it might keep rejecting people from those areas—even if they’re qualified. That’s algorithmic discrimination, where AI repeats old human biases at scale.

Key Vocabulary: - Bias (in AI): A flaw in an AI system that makes it unfair or inaccurate because of the data it was trained on. Example: A voice assistant that struggles to understand accents because it was mostly trained on voices from one region. Grade 7 Note: Bias isn’t always intentional—it’s often a "blind spot" in the data.

• Reinforcement Learning: A type of AI that learns by trial and error, getting "rewards" for correct actions (like a dog learning tricks for treats). Example: A chess-playing AI that starts by making random moves but gets better by learning which moves win games. Grade 7 Note: This is how AI like self-driving cars learn to avoid crashes—but it can also learn bad habits if the rewards are flawed.

• Algorithmic Discrimination: When AI systems treat people unfairly based on race, gender, income, or other factors because of biased data or design. Example: A hiring AI that favors resumes with names it associates with certain backgrounds, even if the candidates are equally qualified. Grade 7 Note: This is illegal in many places, but it’s hard to catch because AI decisions can be invisible.

• Transparency (in AI): How easy it is to understand why an AI made a decision. Example: A credit card AI that denies your application should explain whether it was because of your income, your address, or something else. Grade 7 Note: Some AI is a "black box"—even the people who built it can’t fully explain its choices.

3. Assessment Translation

How This Appears on State Tests (Grade 7): - Multiple Choice: Questions will ask you to identify bias, transparency, or fairness in a scenario (e.g., "Why might an AI loan-approval system reject applicants from certain neighborhoods?"). Distractors often include: - Confusing bias with accuracy (e.g., "The AI is wrong because it’s not smart enough"). - Ignoring the role of data (e.g., "The AI is just bad at math"). - Overgeneralizing (e.g., "All AI is unfair"). - Short Answer: You’ll analyze a case study (e.g., an AI grading essays or diagnosing diseases) and explain one ethical concern and one way to address it. A proficient response: - Names the specific problem (e.g., "The AI might be biased against non-native English speakers"). - Explains why it’s a problem (e.g., "This could lower grades for students who are still learning English"). - Suggests a fix (e.g., "Train the AI on more diverse writing samples"). - Evidence-Based Writing: You might be asked to argue whether AI should replace human teachers in certain subjects, using evidence from provided texts.

Model Proficient Response (Short Answer): Prompt: "An AI tool is used to predict which students might drop out of high school. It flags students based on attendance, grades, and behavior. What’s one ethical concern with this tool, and how could the school address it?" Response: "One concern is that the AI might unfairly target students who are already struggling, like those with learning disabilities or unstable home lives. If the AI is trained on data from mostly wealthy schools, it might not recognize that some students need extra support instead of punishment. The school could address this by including more diverse data in the AI’s training and having teachers review the AI’s predictions before taking action."

4. Mistake Taxonomy

Mistake 1: Ignoring the Data Prompt: "A hospital uses an AI to diagnose skin cancer. It’s 95% accurate for light-skinned patients but only 70% accurate for dark-skinned patients. Why is this a problem, and what’s one way to fix it?" Common Wrong Response: "The AI is bad at its job and should be turned off." Why It Loses Credit: Doesn’t explain why the accuracy differs (biased training data) or propose a solution. Correct Approach: - Identify the bias: The AI was trained mostly on light-skinned patients, so it doesn’t recognize cancer on darker skin. - Explain the harm: This could lead to misdiagnoses for people of color. - Fix it: Train the AI on more diverse data, including images of skin cancer on all skin tones.

Mistake 2: Assuming AI is Neutral Prompt: "A bank uses an AI to approve loans. It denies loans to people from certain ZIP codes more often. Is this fair? Explain." Common Wrong Response: "Yes, because the AI is just following the rules." Why It Loses Credit: Doesn’t recognize that the AI might be repeating past discrimination (e.g., redlining). Correct Approach: - Explain how AI can inherit bias: If the bank historically denied loans to certain neighborhoods, the AI might learn to do the same. - Question the data: Ask if the AI is using irrelevant factors (like ZIP code) to make decisions. - Suggest oversight: Banks should audit the AI’s decisions to check for discrimination.

Mistake 3: Overlooking Transparency Prompt: "Your school uses an AI to recommend classes. You’re placed in a lower-level math class, but you think you could handle a harder one. What should you do?" Common Wrong Response: "Nothing, because the AI knows best." Why It Loses Credit: Doesn’t consider that the AI’s decision might be flawed or unexplained. Correct Approach: - Ask for transparency: Request to see how the AI made its decision (e.g., what data it used). - Challenge the data: Maybe the AI only looked at your grades from one semester, not your improvement. - Advocate for a human review: Schools should let students appeal AI decisions.

5. Connection Layer

• Within Subject: AI in healthcare-AI in education — Both use reinforcement learning to personalize outcomes (e.g., AI tutors adjust lessons like AI doctors adjust treatments), but both can reinforce inequalities if the training data is limited.
• Across Subjects: Algorithmic discrimination-History (Jim Crow laws) — AI can repeat historical discrimination (like redlining) by using biased data, just like old laws enforced segregation. Understanding one helps you spot the other.
• Outside School: Transparency in AI-Food labels — Just like you check ingredients to avoid allergens, you should be able to "check" why an AI made a decision (e.g., why you were denied a loan). Some places, like the EU, now require this by law.

6. The Stretch Question

"If an AI therapist could listen to you 24/7 and give perfect advice, would you use it? Why or why not—and what’s one rule you’d want to make sure it follows?"

Pointer Toward an Answer: - Privacy: Would the AI share your conversations with parents, doctors, or advertisers? (Think about how social media apps track your mood.) - Empathy: Can an AI really understand emotions, or is it just predicting responses based on data? (Compare it to a friend who listens vs. one who just says, "That’s nice.") - Access: If only wealthy people can afford the AI, does it widen the gap between who gets help and who doesn’t? (Like how private tutors give some students an advantage.) - The rule you’d add might be something like: "The AI must explain when it’s unsure about your feelings and suggest talking to a human." Or: "No data can be sold to companies." The best answers will weigh both the benefits and the risks.