The Idea
AI systems learn by finding patterns in large amounts of data. Before a child can understand machine learning, they need a strong intuition for what pattern recognition actually means in practice.
This activity takes that idea and makes it physical. Children gather objects, identify what makes things similar or different, sort them into groups, and eventually try to predict which “group” a new object belongs to — which is exactly what a trained AI classifier does.
Part 1: The Sorting Challenge (10–15 minutes)
Step 1: Gather your objects
Go around the house and collect 15–20 small objects. Variety is good: a coin, a spoon, a LEGO brick, a battery, a pen cap, a button, a small toy, etc. Dump them in a pile in the middle of the table.
Step 2: Sort them — but don’t say how
Tell the child: “Sort these objects into groups. You decide how.”
Don’t tell them what feature to sort by. Let them decide. Common first sorts:
- By color
- By size (big/small)
- By material (metal, plastic, fabric)
- By whether it’s useful or decorative
Ask: “How did you decide which group each one goes in?”
Step 3: Reveal the rules
Once they’ve sorted, ask: “What’s the rule for each group?” Have them write it down on a sticky note. This is their “model” — the set of rules it learned from sorting.
Step 4: The test
Now bring out a new object they haven’t sorted yet. Can they predict which group it belongs in, using their rules?
Ask: “How sure are you? What if it could fit in two groups?”
Part 2: Pattern Sequences (10–15 minutes)
Step 5: Make a sequence
Create a simple pattern with some objects:
- Red button, blue button, red button, blue button, ___
- Spoon, fork, spoon, fork, ___
- Big coin, small coin, small coin, big coin, small coin, small coin, ___
Ask the child to predict what comes next. Then make it harder:
- Two attributes: big red, small blue, big red, small blue, ___
- Three: coin, button, bottle cap, coin, button, ___
Step 6: Make it ambiguous
Create a sequence where more than one answer is reasonable:
- Red, blue, red, blue, red, ___ (blue? Or is it about to change?)
Ask: “What would you guess? How confident are you? What information would make you more sure?”
This is a key moment: real AI systems also make predictions with varying degrees of confidence. When they’re uncertain, they might be wrong.
The AI Connection
Step 7: Make the connection explicit
Talk through what just happened:
“When you sorted those objects, you were acting like a machine learning system. Here’s how:
- The objects are data
- The features you noticed (color, size, material) are what AI calls features or attributes
- The sorting rules you figured out are like a trained model
- When you predicted where the new object belonged, you were making a prediction — just like AI does”
“Now here’s what makes AI different from you: it can look at millions of objects and find patterns you’d never have time to spot. But here’s what makes you different from AI: you understand why the pattern matters. AI just knows the pattern exists.”
Step 8: The tricky part — bias
For children aged 9+, extend the conversation:
“What if all the objects you practiced with were plastic? Would your rules work as well for metal objects? What if most of the red objects in your pile were also small — would your system think ‘red’ and ‘small’ always go together, even if that’s just a coincidence?”
This is the essence of training data bias. If the training data isn’t representative, the model learns the wrong things.
Variations
For younger children (6–7): Focus only on Part 1. Sorting one attribute at a time is enough. Skip the sequences.
For older children (9–10): Add a “test set” they set aside before sorting. After building their sorting rules, apply them to the test set and see how accurate they are. What percentage did they get right?
Competitive version: Two children each sort the same objects using different rules. Then swap — can you figure out the other person’s sorting rules just by looking at how they arranged things? This mirrors how AI interprets patterns without being told the rule.