Detective Case Brief

Debrief questions for after playing Kai's Decision Tree Detective

🕵️

🔍 Case File · Module A · Decision Trees & Training Data

Operation: Train the Detective

Use these questions to guide a discussion after your child plays the Decision Tree Detective game. Encourage them to explain their reasoning — not just give short answers.

👋 Note for the grown-up: These questions work best with the game fresh in mind. Grades 3–5 students can handle "why" and "what if" thinking. Push for explanations, not just yes/no answers. Approximate discussion time: 10–15 minutes.

🌳 "Kai's tree got Fox right — but got confused by Penguin. What was different about how Penguin was trained versus how Fox was trained?"

What to listen for: Fox was in the training data; Penguin wasn't. The key idea is that an AI can only make good decisions about things it has seen before.

🔎 Go deeper: "What would you add to Kai's training set so Penguin would be classified correctly next time?"

🗂️ "In the game, you chose which animals to add to Kai's training data in Phase 3. Why did adding more data help — and why isn't just adding MORE data always enough?"

What to listen for: More data helps only if it is the right kind of data. Adding 100 more dogs won't help Kai classify a Penguin better.

🔎 Go deeper: "What kind of training data would you add to make Kai smarter — more dogs, or completely different kinds of animals?"

🤔 "Imagine someone built a real AI doctor trained only on data from adult patients. A child comes in for diagnosis. What might go wrong?"

What to listen for: Children's bodies, symptoms, and diseases are different from adults'. An AI trained only on adult data might diagnose incorrectly or miss something important for children.

🔎 Go deeper: "Why do you think it might be hard to collect training data from children compared to adults?"

🌍 "Who is responsible for making sure an AI has good training data — the programmer, the company, or everyone who uses it? Could it be all three?"

What to listen for: There's no single right answer — this is a values discussion. Programmers choose what to include; companies decide on timelines and budgets; users notice when it fails. All three share responsibility.

🤖

Kai says: "Every time you added a new animal to my training data, I could make better decisions. But here's the tricky part — I don't know what I'm missing. Only you can look at my training data and spot the gaps. That's why humans need to stay involved."

💡 Core Insight — Module A AI learns from examples. Incomplete training data = confident but wrong AI. The more diverse and complete the training data, the more reliable the AI.

🔗 Connection to Module B When training data is not just incomplete but systematically skewed toward one group, AI doesn't just make mistakes — it makes unfair mistakes. That's data bias.