Back to the mission

AI for Kids · AI Detective Academy · Mission 1

Picture Clue Patrol: How Does AI See?

30–38 minutes · film 3 min

Before you start — the hook

Show one clear animal card and one cropped or shadowed version. Ask: “What clues stayed the same?” Do not define image classification yet.

Do not define anything yet. Let them guess first.

The big idea

It places a picture into a category.

Words to know: image · feature · label · classification · prediction · unsure

  • Define image classification in child-friendly language.
  • Use at least two visible features to justify a category.
  • Choose “Unsure” when evidence is incomplete and explain why.

Activity — Mystery Picture Sort

You will need: paper, something to write with, and the cards described in the steps

  1. Place Cat, Dog and Unsure category cards on a table or screen.
  2. Sort six clear animal cards and name one visible clue for each.
  3. Sort six tricky cards containing shadows, costumes, partial views or unusual angles.
  4. Move weak-evidence cases to Unsure rather than forcing a guess.
  5. Write one rule that would help Pip avoid a repeated mistake.

Your own mission

Find a picture of an animal — a book, a magazine, or one you draw yourself. Show it to someone and finish this out loud: “I chose ___ because I noticed ___ and ___.” Two clues, not one!

Adapting it

Ages 6–8: use only picture cards and oral explanations.

Ages 13–16: compare a hand-written rule with a real classifier and discuss confidence scores.

Quiz and answer key

Pass mark 4 of 5. Always give the explanation, right or wrong.

  1. 1. What does image classification do?

    • It places a picture into a category. ✓
    • It makes the picture bigger and clearer.
    • It understands the picture the way you do.

    Why: This is the central concept of the lesson.

  2. 2. Why might a cat wearing a costume confuse a classifier?

    • The costume may create misleading visual features. ✓
    • The classifier thinks costumes are funny.
    • A costume turns the animal into something else.

    Why: AI can focus on the wrong pattern.

  3. 3. When is “Unsure” the best answer?

    • When the picture does not provide enough reliable evidence. ✓
    • When you want to finish quickly.
    • Never — you should always pick an answer.

    Why: Responsible systems should not invent certainty.

  4. 4. Is one feature always enough?

    • No. Several relevant clues are often needed. ✓
    • Yes, if it is a really good clue.
    • Yes, as long as the picture is clear.

    Why: Single-feature rules are fragile.

  5. 5. What should we do after an incorrect prediction?

    • Check the clues, question the rule and improve it. ✓
    • Decide the AI is broken and stop using it.
    • Ignore it — one mistake does not matter.

    Why: Errors are evidence for improvement.

Misconception to watch for: The learner practises evidence-based sorting and learns that computer vision can be useful without being perfect.