AI for young mindsBeginnerLesson 783 min read

How a computer learns from examples

AI doesn't learn like you do from a teacher's rules. It finds patterns in lots of examples. Try being the AI yourself!

✦ Ask the tutor 🧪 Lab · 15 min 📝 Lesson test

Speed

Lesson in motion

In 60 seconds

How a computer learns from examples

AI doesn't learn like you do from a teacher's rules. It finds patterns in lots of examples. Try being the AI yourself!

1/4

In simple words

How did you learn which mangoes are ripe? Nobody gave you a rulebook. You saw lots of mangoes, touched them, smelled them, and your brain found the pattern. AI learns in a similar way, from examples.

Old computer programs followed rules that people wrote, step by step. Many AI systems work differently: people give them lots of examples with the right answers, and the computer finds the pattern by itself. This is called machine learning.

Two stages

Stage	What happens	Mango example
Training (learning)	The computer sees many labelled examples and finds a pattern	Thousands of mango photos marked "ripe" or "not ripe"
Using	It uses the pattern on something new	You show a new mango photo, it guesses "ripe"

Your email app may do the same with spam: it learned from many messages marked "spam" or "not spam", and now it guesses for new messages.

Why variety matters

If the computer only saw yellow Alphonso mangoes, it might think every green mango is unripe, even if some green kinds are ripe! More varied examples (different kinds, colours, lighting, angles) help it learn the real pattern, not a shortcut.

Weak examples

Only one kind of mango
All photos in bright sunlight
Just a few examples

Strong examples

Many kinds of mangoes
Photos in different light and places
Lots of examples, checked carefully

Note

AI finds patterns, but it doesn't "understand" mangoes like you do. It can't taste or smell them. It can be tricked by things it never saw in training.

Be the AI! In the lab, you will learn a secret rule only from examples, just like a computer does.

Watch and read more

Lab

An unplugged "Be the AI" game where you learn a secret rule from picture cards.

~15 min

The problem

Play with a parent, sibling, friend or class. No screen needed.

Teacher: draw 10 simple picture cards (fruits, animals, shapes). Secretly choose a rule, like "things that are round" or "things that can fly".
Show the cards one by one to the "AI" player, saying "YES" or "NO" for each. This is training.
Now show 5 new cards without saying anything. The AI player guesses YES or NO. This is using.
Count correct guesses, then reveal the rule.
Play again with a trickier rule. Did more varied cards help?

You are done when

You played both stages: training with answers, then guessing on new cards
You counted how many guesses were right
You noticed if the AI player learned a wrong shortcut
You tried a second round and compared

Hard questions

Try to answer before you reveal. If you can answer these, you understood the lesson.

Q1In the game, the AI player guessed "yes" for everything red. What went wrong?Reveal

Maybe many of the training "yes" cards happened to be red, so the player learned a shortcut: "red means yes". The real rule was something else. Real AI can make the same mistake when examples aren't varied enough.

Q2Why test the AI on new cards, not the same ones used in training?Reveal

Getting the training cards right might just mean remembering them. Testing on new cards shows whether the player really learned the pattern. Real AI builders do the same: they test on examples the AI has never seen.

Please sign in to continue.

Questions people ask

Who gives AI its examples?

People and companies collect them: photos, text, sounds and more. Sometimes people label them by hand, like marking photos "ripe" or "not ripe".

Does AI keep learning while I use it?

Some systems are updated later with new examples, but many don't change while you use them. Learning usually happens in the training stage.

Is this how chatbots learn too?

In a similar way. Chatbots learn patterns from huge amounts of text, which helps them guess good next words. That's also why they can sound right and still be wrong.

Lesson test

5 questions. Get 3 right (60%) to pass and complete this lesson.