AI and Fairness in Decisions

Computing · Artificial Intelligence · ages 9-11

Name ______________________ Date ____________

AI is sometimes used to help decide who gets approved for a loan.
Circle one: True False
Which of these is an example of an important decision AI might help make about a person?
- Choosing who gets a job interview
- Picking a song for a playlist
- Choosing a wallpaper color
Why can a biased AI hiring program be risky?
- It might unfairly reject good candidates
- It always hires the fastest typer
- It never looks at any applications
A hospital's AI flags a patient as high risk. Who should check that flag before treatment decisions are made?
- No one, the AI is always right
- The patient's neighbor
- A doctor or nurse
This chart shows how many out of 100 applicants from two equally skilled groups got invited to interview. What does this gap suggest?
- The AI might be treating one group unfairly
- Group B applied on a later day
- Group A used nicer paper
If an AI system makes an unfair loan decision, no human is responsible because the computer decided.
Circle one: True False
Out of 100 loan applicants, Group A had 80 approved and Group B had 40 approved. How many more Group A applicants were approved, out of every 100?
Answer: ______________
A sentencing AI recommends longer sentences for people from one neighborhood, because that area was policed more, not because more crime happened there. What is the real problem?
- The tool needs a faster computer
- The AI learned an unfair pattern from biased data
- The judge typed the request too fast

Answer key

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

True · Banks use computer programs to help sort loan applications, which makes loan approval a high-stakes AI decision.
Choosing who gets a job interview · Job interviews affect someone's real life and future, so that counts as a high-stakes decision, unlike picking a song.
It might unfairly reject good candidates · A biased program can reject strong candidates because of an unfair pattern, not because of their real skill.
A doctor or nurse · A trained person should always review an AI's flag, since the AI itself cannot be held responsible for mistakes.
The AI might be treating one group unfairly · When equally qualified groups get very different results, it's a warning sign that the AI learned an unfair pattern.
False · People chose to build and use the AI, so they stay responsible for checking and fixing its unfair decisions.
40 · 80 minus 40 is 40. A gap that large between equally qualified groups suggests the AI may be unfair.
The AI learned an unfair pattern from biased data · The AI copied a pattern caused by unequal policing, not by actual behavior, so its recommendation is unfair.