AILevel 1 · AI ExplorerLesson 11

L1 · 11

Labels & Categories

Before an AI can sort the world, we have to decide the boxes. Clear labels and clean categories are what let a model tell things apart.

⏱ 1.5 hours🤖 Concept lesson · no coding📚 After AI-L1-10💬 Discussion + worksheet
01

Learning Goals

5 min

By the end of this lesson you can:

  • Explain that a label names the category of an example.
  • Design clear, separate categories (classes) for a model.
  • Explain why messy categories make a worse AI.
02

Warm-Up · How Many Groups?

8 min

Last lesson Quick, Draw! sorted doodles into words. Those words were its categories.

How would you sort these into groups: 🍎 🍌 🐱 🐶 🍇 🐰 ? How many groups, and what would you call them?

Reveal one answer

Two groups: fruit (🍎🍌🍇) and animals (🐱🐶🐰). "Fruit" and "animals" are the labels for your categories.

03

New Concept · Boxes for the World

18 min

Labels and categories

A category (also called a class) is a group, like "cat" or "dog". A label is the category name we attach to one example.

Think of sorting your room into boxes. The boxes are the categories; writing "LEGO" on a box is the label.

🍎🐶🍌🐱fruitanimals🍎🍌🐶🐱
The categories you choose decide what the AI can tell apart.

What makes a category good?

  • Clear — everyone agrees what belongs in it.
  • Separate — categories don't overlap.
  • Balanced — roughly the same number of examples each.
Why it matters

The categories you pick are the only things the AI can ever say. Choose "cat" and "dog", and it can never say "rabbit".

04

Worked Example · A Fruit Classifier

18 min

Lakshmi wants an AI that names local fruit from a photo. She chooses three clear classes:

  • rambutan
  • durian
  • mangosteen

These are good categories — clear, separate, and easy to collect balanced examples for.

A bad set of categories

Compare these poor choices and why they fail:

Bad categoryProblem
"fruit" and "rambutan"Overlapping — a rambutan is a fruit.
"nice fruit" and "yucky fruit"Not clear — people disagree.
"round things"Too broad — a ball isn't fruit.
The takeaway

Spend time choosing clean categories before collecting data. Good boxes make an AI that's easy to train and trust.

05

Try It Yourself

20 min

Use your worksheet.

01 🟢 Label the items

Sort these into two clear categories and name them: teh tarik, kopi-O, nasi lemak, roti canai, milo ais, satay.

Hint

Think "drinks" and "food".

02 🟡 Fix the categories

An AI uses the classes "animal", "cat", and "pet". Explain what's wrong, then rewrite them into three clean, separate classes.

Hint

These overlap — a cat is an animal and a pet. Pick categories that don't contain each other.

06

Mini-Challenge · Design Three Clean Classes

12 min

Design a classifier for the contents of a school bag. Choose three clear, separate categories and give two example items for each.

It works if no item could belong to two of your categories, and a classmate agrees with your sorting.

Show one good design

Classes: "books" (textbook, notebook), "stationery" (pencil, ruler), "food" (water bottle, kuih). Clear and non-overlapping.

07

Recap

5 min

A category is a group; a label is the category name on one example. Good categories are clear, separate and balanced. The classes you choose decide everything the AI can tell apart.

Vocabulary Card

category (class)
A group an AI can sort examples into, like "cat" or "dog".
label
The category name attached to one example.
overlap
When two categories share members — a sign the categories need fixing.
08

Homework · Categorise a Collection

≤ 20 min

Pick a collection at home — clothes, snacks, toys, coins. Design three clear, separate categories for it, name them, and list three example items in each.