Gather a tiny collection
Put ten safe everyday objects on a table: buttons, blocks, pencils, or paper shapes. Avoid small objects for children who may put them in their mouths. You also need two sheets of paper labeled Group A and Group B. An adult secretly decides a sorting rule, such as round versus not round.
Place a few examples in each group. Ask your child to inspect them and guess the hidden rule. Explain that this is a simplified model of learning from examples; actual machine-learning systems are more complex and this game does not reproduce how every AI works.
Choose examples that can mislead
If every round example is red and every non-round example is blue, a learner may infer that color is the rule. Introduce a blue circle. Does the prediction change? Now add a red square. Talk about how a pattern that looked convincing can fail on a different example.
Do not mark an early wrong guess as a failure. Ask what information was available and what was missing. This is a useful way to discuss why the choice of examples matters and why testing only familiar cases can hide problems.
Separate practice from the test
Let your child use six shapes to work out the rule. Hold back four shapes until they are ready to test it. Record the guesses before revealing the intended groups. If the rule changes after the test, create a fresh test set rather than repeating only the examples already seen.
Take turns choosing the rule. A child may invent one that needs information the other player cannot see, such as ‘objects I like.’ That is a good moment to ask whether the available information is enough for a fair prediction.
Connect it to everyday AI
Explain that some AI systems learn patterns from examples and use those patterns to make predictions about new inputs. Predictions can be wrong, particularly when new examples differ from the ones used in learning. Sorting buttons does not teach a machine to care, understand a family, or decide what is fair.
Finish by drawing an imaginary machine that sorts something useful. Label its examples, its prediction, and one situation where a person should check its work. The learning goal is a thoughtful explanation, not memorizing a definition.
