En anglais

AI and Fairness in Decisions

Computing · Artificial Intelligence · ages 9-11

Name ______________________ Date ____________

AI is sometimes used to help assess loan applications.
Circle one: True False
Which decision is high-stakes because it can seriously affect a person's life?
- Choosing a playlist's background color.
- Choosing which applicants get invited to a job interview.
- Choosing a song for a short music break.
Why can a biased AI hiring tool be risky?
- It can reject suitable applicants for unfair reasons.
- It can give applicants more chances to explain their skills.
- It can help interviewers organize the questions they want to ask.
A hospital AI flags a possible patient risk. Who should review the warning before it guides treatment?
- The software supplier's support worker who handles equipment faults.
- The person who installed the hospital's AI software.
- A health professional qualified to assess this patient's case.
Each equally qualified group has 100 applicants in this made-up test. The chart shows interview invitations. What does the gap suggest?
- Group B had fewer applicants.
- The chart identifies the training records that caused the gap.
- The unequal results need checking for unfair treatment.
An organization can avoid responsibility for an unfair loan decision by saying 'the computer decided'.
Circle one: True False
Group A and Group B each have 100 loan applicants. There are 80 approvals in A and 40 in B. How many more applicants were approved in A?
Answer: ______________
In a made-up case, two areas have equal crime levels. One has more arrests because it has more policing. An AI treats those arrests as proof of more crime. What is the problem?
- It uses a difference in policing as evidence of a difference in crime.
- It treats the two areas as having the same crime level.
- It leaves the arrest records out of its recommendation.

Answer key

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

AI and Fairness in Decisions W1-mt_cVp_nop-5L-s1

True · Loan assessment is one high-stakes task AI can help with.
Choosing which applicants get invited to a job interview. · An interview can affect access to work. Ordinary playlist choices usually have much smaller consequences.
It can reject suitable applicants for unfair reasons. · Unfair rejection is a possible harm from bias. The other options describe possible benefits rather than the risk.
A health professional qualified to assess this patient's case. · The warning needs qualified assessment of the patient's case. Installing software or handling equipment faults does not establish that qualification.
The unequal results need checking for unfair treatment. · The groups have equal totals and qualifications. Their unequal results are a warning sign, but the chart does not identify a cause.
False · The organization using the tool still has a responsibility to check and address unfair results.
40 · 80 minus 40 is 40 more approvals. These counts alone do not tell us whether the decisions were fair.
It uses a difference in policing as evidence of a difference in crime. · The arrest gap comes from different policing in this case. The tool incorrectly uses that gap as evidence of different crime levels.