Course 2, lesson 14 of 100, Ages 6+

Labels and names

Telling AI what’s what

Like I’m 5

A label is a name tag. If a picture shows a dog, its label says 'dog'. Labels help AI learn which picture is which.

The big idea

When we teach AI with examples, each example often comes with the right answer, called a label: a photo plus the label 'cat', or a sound clip plus the label 'hello'.

People do a lot of this labelling, and it takes care. If labels are wrong, the AI learns wrong. That's why teams double-check labels and ask several different people to label the same examples.

Examples

  • Photo tags: Tagging friends in photos created labels that helped face recognition.
  • Email: Every time you press 'mark as spam', you add a label.
  • Picture puzzles: Clicking every square with a traffic light helped label street photos.

How it works

  1. Collect examples, like photos or sounds.
  2. Add the correct label to each one.
  3. Train the AI on the examples and their labels.

Check your understanding

What is a label?
Options: The right answer attached to an example; The price of a computer; A sticker on a laptop.
Answer: The right answer attached to an example. A label tells the AI what the example shows.
What happens if lots of labels are wrong?
Options: The AI learns the wrong thing; The AI gets faster; Nothing at all.
Answer: The AI learns the wrong thing. AI learns from labels, so wrong labels teach wrong lessons.

Remember

Labels are the right answers attached to examples. Correct labels mean better learning.

Talk about it

If you labelled photos for an AI, what could make it tricky?

Go deeper

Learning from labelled examples is called supervised learning. Label quality is often measured by how often different labellers agree with each other.