Practice, then take a test
AI learns with practice data, and then takes a test with data it has never seen to check how well it learned.
Story
Mina is practicing for a spelling test. She got every word she practiced right. But the test had a word she had never seen. “Huh? I didn’t practice this one…” You only know if you really learned when you try a problem you have never seen.
Let’s learn
When learns, it is called . The used for training is called .
After it has finished learning, it takes a test. The test uses data it has never been shown, not even once. If you test it with things it has already seen, you cannot tell whether it just memorized them or really knows.
- Train: It practices with 1000 labeled photos.
- Test: It has to guess 100 photos it has never seen.
- Score: Count how many of the 100 it got right.
- Train again: If it got many wrong, show it more examples and let it practice again.
AI gets better little by little by making mistakes. It is just like a person who learns by trying again the problems they got wrong.
At a glance
-
Practice
-
Take a test
-
Check the score
Many wrong? Practice again
In simple words
One friend memorized the answers in the workbook. Another friend learned how to solve the problems. Give them a new problem and you will see who really learned.
Try it
Play ‘Teacher and AI.’
- Draw ten triangles and ten squares on paper and cut them out.
- Show twelve of them to a friend and say “triangle” or “square” for each one. This is training.
- Show the other eight one at a time and let your friend guess. This is the test.
- Count how many your friend got right. Did your friend get a lopsided triangle right, too?
Quiz
If the sentence is right, choose [True]. If it is wrong, choose [False].
-
1To find out if AI learned well, you have to test it with data it has never seen.
That’s right!
Not quite. Read the reason.
If you test it with things it has already seen, you cannot tell whether it memorized them or really knows.
-
2Once AI has learned something, it never gets it wrong.
That’s right!
Not quite. Read the reason.
AI makes mistakes, too. It looks at what it got wrong, trains again and gets better little by little.
-
3The data AI uses to learn is called training data.
That’s right!
Not quite. Read the reason.
Training data is AI’s practice workbook.
Correct answers / 3
Remember
- AI practices with training data.
- It takes a test with data it has never seen.
- When it gets things wrong, it trains again and gets better.
Today’s words
Tap a word to see what it means.
Go deeper
These pages are for older students and grown-ups. Read them together with a grown-up.