Training Data Inspector

Two AIs trained on different data. Only one is ready for the real world.

🔍 The Case: A school wants to use an AI to recommend healthy food choices. Two AI systems were built — but they were trained on very different datasets. Your job as a Training Data Detective: examine both datasets, identify the problems, and predict which AI will fail — and who it will fail for.

⚠️ AI System A

The "QuickBot"

Trained on 500 food examples. Ready in 2 weeks.

🍔 Burger (unhealthy)
🥗 Caesar salad (healthy)
🍕 Pizza (unhealthy)
🥪 Sandwich (healthy)
🍟 French fries (unhealthy)
🍎 Apple (healthy)
❌ No African foods
❌ No Asian foods
❌ No Latin American foods
❌ No traditional dishes

✅ AI System B

The "GlobalBot"

Trained on 2,000 food examples. Ready in 3 months.

🍔 Burger (unhealthy)
🥗 Caesar salad (healthy)
🍚 Rice and dal (healthy)
🫔 Injera with lentils (healthy)
🍜 Miso soup (healthy)
🫘 Black beans (healthy)
🥙 Falafel wrap (healthy)
🍱 Bento box (mixed)

A student from Nigeria brings jollof rice and fried plantain for lunch. Which AI is more likely to classify it correctly? Why?

The school uses AI-A. A student from Mexico brings tamales — which AI-A has never seen. Predict what might happen:

⚠️ AI-A will probably:

✅ AI-B would probably:

The school is in a hurry and picks AI-A because it's faster and cheaper. Write one consequence this could have for students:

🔧 Your Fix: List 3 food types you would add to AI-A's training data to make it fairer for ALL students in your school: