In what is AI, explained for kids, we looked at the big picture. This article zooms in on one important idea behind many modern AI systems: machine learning.
Rules versus examples
Imagine you want a computer to tell cats from dogs in photos.
Approach one: write the rules. "If it has pointy ears and whiskers, it is a cat." But some dogs have pointy ears, and some cats have floppy ones. The rules get longer and longer and still fail.
Approach two: show examples. Show the computer thousands of photos, each labelled "cat" or "dog," and let it work out for itself what tends to separate them. This is machine learning: instead of being told the rules, the computer finds patterns in examples.
That is also how young children learn. Nobody gives a toddler a definition of "dog." They meet dogs of all sizes and shapes until the idea clicks.
The three ingredients
- Data: the examples (photos, sentences, numbers).
- Training: the computer adjusts itself again and again to make fewer mistakes on those examples.
- A model: the result of training, which can then make guesses about new things it has never seen.
An important word is guess. A trained model makes predictions, and predictions can be wrong.
Why the examples matter so much
A model can only learn from what it is shown. If you train a cat-and-dog detector using only photos of black dogs, it may struggle with golden ones. If the examples are unfair or incomplete, the model's guesses will be too. This is a valuable lesson even for young children: good AI depends on good, varied data, and people choose that data.
It also explains why it is wise to check what an AI tells you. Our guide on using AI chatbots safely with children covers habits like checking answers and not sharing personal information.
An activity: be the machine
You can act out machine learning with paper.
- Cut out about 20 cards. On each, draw a simple creature: some with wings, some with legs, some tall, some small.
- Secretly decide on a rule, for example "creatures with wings are zorps."
- Show a friend cards one at a time, saying "zorp" or "not a zorp" for each.
- After ten cards, show new cards and ask them to guess.
Your friend just learned from examples, without ever hearing the rule. Then try with a bad set of examples (where all zorps also happen to be blue) and see how they get fooled. That is exactly how a model can go wrong.
What machine learning can and cannot do
- It is excellent at finding patterns in lots of data: recognising speech, sorting photos, suggesting words as you type.
- It does not understand things the way a person does. It does not have feelings or intentions.
- It is not magic and it is not always right.
Where to go next
For an even gentler introduction to patterns, read pattern recognition for kids. Children curious about how all this fits together can explore the AI course, which introduces these ideas with simple, hands-on examples.