Family Activity: Build a Mini “AI Detective” Game to Teach Bias & Fairness (9–13)

Why an “AI Detective” game is the easiest way to explain AI fairness to kids

Kids (especially ages 9–13) understand fairness instinctively: “That’s not fair!” is basically a universal language. The tricky part is helping them see how an AI system—something that sounds super smart—can still make unfair decisions.

This mini “AI Detective” game turns your family into investigators. Your child plays the detective, you play the “AI,” and together you run a case: deciding who gets approved for a made-up club membership. The twist? The AI makes decisions based on a few clues (data), and some clues are missing or skewed, creating bias.

This is one of those rare family activities about technology that is:

• Hands-on (no special devices needed)
• Quick to set up
• A real-world way to teach AI literacy
• Perfect for answering: “How do you explain AI fairness to kids?”

By the end, your child won’t just know that “AI can be biased.” They’ll know why it happens and what fairer decision-making looks like.

What you’ll build: a mini “AI Detective” decision game (20–40 minutes)

You’ll create a simple game with “case cards” (people applying to a club) and an “AI rule sheet” (how the AI decides yes/no). Then you’ll intentionally introduce bias and let your detective uncover it.

Materials

Keep it simple—use what you already have:

• Paper or index cards
• Pencil/marker
• Optional: sticky notes, a coin, a timer

Step 1: Create 10 case cards (your dataset)

On each card, write:

• Name (any)
• Age (9–13 range is fine)
• Experience (0–3 points)
• Recommendation (Yes/No)
• Availability (Weekdays/Weekends)
• Neighborhood (North/South) — this is your “sensitive” feature for the lesson

Tip: Let your child help invent the applicants. It makes the lesson feel like a real mystery.

Here’s a ready-to-copy sample dataset you can use as-is:

Applicant	Experience (0–3)	Recommendation	Availability	Neighborhood	Should be accepted? (Human guess)
Asha	3	Yes	Weekends	North	Yes
Ben	2	Yes	Weekdays	South	Yes
Cam	1	No	Weekends	North	Maybe
Dalia	0	Yes	Weekends	South	Maybe
Ezra	3	No	Weekdays	South	Maybe
Finn	2	No	Weekends	North	Maybe
Gia	1	Yes	Weekdays	South	Yes
Hugo	0	No	Weekdays	North	No
Iris	2	Yes	Weekends	North	Yes
Jae	1	Yes	Weekends	South	Yes

The last column is important: ask your child to circle who they think should be accepted before the AI makes any decisions. This gives you a “fairness baseline.”

Step 2: Make the “AI rule sheet” (the model)

Tell your child: “This is how our pretend AI decides.” Then start with a reasonable rule:

Fair-ish starting rule (Version 1):

• Accept if Experience ≥ 2 OR Recommendation = Yes

Run through all 10 cards and mark the AI’s decisions. Your child should feel like this is mostly sensible.

Step 3: Add the twist—introduce bias on purpose (Version 2)

Now you’ll quietly update the AI rule sheet:

Biased rule (Version 2):

• Accept if Experience ≥ 2 OR Recommendation = Yes
• BUT if Neighborhood = South, require Experience ≥ 2 AND Recommendation = Yes

Don’t call it “biased” yet. Just run the same 10 cards again and record accept/reject.

Step 4: Your child becomes the detective

Give your child the results and ask them to investigate:

• “Do you notice any pattern in who gets rejected?”
• “Is the AI treating everyone the same?”
• “Which clue seems to matter too much?”

This is where kids usually spot it fast: the Neighborhood clue is changing outcomes.

The fairness lesson: bias isn’t just ‘mean’—it’s often hidden in rules or data

Now connect the game to real AI in a way that’s clear and not scary.

Explain bias in kid-friendly language

Try this script:

• Bias means the AI’s decisions “lean” in a direction that isn’t fair.
• The AI isn’t mad at anyone—it’s following rules or learning from patterns.
• If the patterns are unfair (or the rules are unfair), the results will be unfair.

Two big ways AI bias shows up (and how your game demonstrated them)

• Biased rule (design bias): You literally told the AI to be stricter for “South.” In real life, this can happen when someone chooses a shortcut rule that ends up hurting a group.
• Biased data (training bias): Even if you remove Neighborhood, bias can sneak in if another clue acts like a “stand-in.” For example, if Availability or Recommendation is systematically different for reasons outside a kid’s control.

A simple fairness check kids can actually do

You don’t need advanced math. Use counts.

1. Count how many North applicants were accepted.
2. Count how many South applicants were accepted.
3. Compare.

If one group is accepted much less often, ask: “Is that difference explained by the goals of the club (skills, interest), or is it coming from something unrelated?”

To make it even more concrete, use this mini scorecard after Version 2:

Group	# Applicants	# Accepted	Acceptance Rate
North	5	___	___%
South	5	___	___%

Your child fills in the blanks like a real investigator.

Talk about fairness without making it overwhelming

Keep it grounded:

• Fair doesn’t always mean “everyone gets the same outcome.”
• Fair usually means “everyone gets the same chance based on relevant reasons.”
• If the AI uses a reason that shouldn’t matter (like Neighborhood in this game), it can become unfair.

This is the heart of activities to teach kids about AI bias: show them that “smart” systems can still make flawed choices—and we can measure and fix it.

Make it better: 3 kid-friendly upgrades that teach real AI thinking

Once your detective finds the bias, the best part is letting them improve the system.

Upgrade 1: Remove the sensitive clue (and see what changes)

Create Version 3 of the rule:

• Accept if Experience ≥ 2 OR Recommendation = Yes
• Do not look at Neighborhood

Ask:

• “Did the acceptance rates get closer?”
• “Do the results match our ‘human guess’ column more often?”

Upgrade 2: Add a “reason code” for each decision

Real systems often need explanations. Have your child write one reason next to each accept/reject:

• “Accepted: Experience ≥ 2”
• “Accepted: Recommendation = Yes”
• “Rejected: Not enough evidence”

Then discuss:

• “Is the AI using good reasons?”
• “Are the reasons consistent across people?”

Upgrade 3: Fix the data (more complete information)

Tell your child: “Sometimes AI is unfair because it doesn’t have the right info.” Add one more clue to every card:

• Club Tryout Score (0–5)

New rule idea:

• Accept if Tryout Score ≥ 4 OR (Recommendation = Yes AND Experience ≥ 1)

This shows a powerful lesson: better, more relevant data can reduce unfairness.

Quick reflection questions (keep it conversational)

Use any 3–4 of these:

• “What did the AI assume about people from the same neighborhood?”
• “What information would make this decision more fair?”
• “If you were rejected, what explanation would you want?”
• “Should an AI be allowed to use every clue it can find?”

These questions naturally lead into AI lessons for kids at home—without needing a lecture.

Next Steps: turn this into an ongoing at-home AI fairness habit

If you want this to stick (and not feel like a one-time activity), here’s a simple plan.

• Play again with a new ‘case.’ Switch the setting to something your child cares about:
  • Picking a team captain
  • Choosing who gets a limited number of event tickets
  • Recommending books or videos
• Try a “mystery bias” round. Don’t tell your child what changed in the rules; let them discover it.
• Add one fairness goal each time. Examples:
  • “Similar applicants should get similar outcomes.”
  • “The AI must give a reason for every decision.”
  • “No decisions based on personal background clues.”
• Connect it to real life gently. Ask when you see recommendation systems:
  • “What do you think this app is using as clues?”
  • “Could those clues accidentally leave someone out?”

If your child enjoyed the detective role, that’s your signal to keep going. You’re not just teaching them about technology—you’re teaching them how to question it thoughtfully.

And that’s the core of how to explain AI fairness to kids: make it hands-on, make it measurable, and show them they can improve the system—not just accept it.