When AI Gets It Wrong

Computing · Artificial Intelligence · ages 7-9

Name ______________________ Date ____________

Why is AI not magic?
- It knows everything without being taught
- It follows maths rules using data it has seen
- It can never make a mistake
Which of these is an example of an AI making a mistake?
- A voice assistant hears 'call mom' as 'call Tom'
- A calculator adds two plus two to get four
- A ruler measuring a pencil correctly
AI can make mistakes if its training data has gaps or errors.
Circle one: True False
An AI trained mostly on pictures of golden retrievers gets confused by a picture of a poodle. Why?
- It was not shown many poodle pictures during training
- Poodles are not real dogs
- The AI has no camera
Zara typed 'I will meet you soon' but her phone changed 'soon' to 'moon'. What kind of mistake is this?
- A maths problem with no answer
- A broken screen
- An autocorrect error
An AI recommends a maths app right after a child watches a maths video. Why did it pick that?
- It read the child's mind
- It matched patterns in what the child looked at before
- It guessed using magic
Which sentence best explains why AI sometimes gets things wrong?
- AI follows maths rules learned from examples, and those examples were not perfect
- AI mistakes are always the user's fault
- AI is alive and sometimes feels lazy
An AI mixed up a photo of a muffin with a photo of a chihuahua. What is the most likely reason?
- The AI wanted to trick people
- Muffins and chihuahuas are the same animal
- The training photos looked similar, and the AI had not seen enough different examples

Answer key

For grown-ups. Fold this page away before handing over the rest.

It follows maths rules using data it has seen · AI does not know things by magic. It uses maths rules on data it has already seen.
A voice assistant hears 'call mom' as 'call Tom' · The assistant guessed the words from sound patterns and got it wrong. That mix-up is a real AI mistake.
True · Gaps or errors in what the AI was shown lead to bad guesses later.
It was not shown many poodle pictures during training · The AI mostly saw golden retrievers, so a poodle's different shape confused its guess.
An autocorrect error · Autocorrect guesses words from patterns it has learned, and sometimes that guess is wrong.
It matched patterns in what the child looked at before · The AI is matching patterns in past choices with maths, not truly reading minds.
AI follows maths rules learned from examples, and those examples were not perfect · AI follows maths on data, and when that data is incomplete, its answers can be wrong.
The training photos looked similar, and the AI had not seen enough different examples · Muffins and chihuahua puppies can look alike in photos. Without enough varied examples, the AI's pattern match failed.