Social StudiesFreePrintable

Algorithmic Bias Case Studies

A free social studies worksheet ready for your classroom. Open in Kuraplan to grab the print-ready PDF, customize it for your students, or generate a fresh version in seconds.

Download now

Name:Date:

Algorithmic Bias Case Studies

Read the case file and examine the results table. Then investigate how bias can enter automated decisions, who is affected, and what responsible action could look like. Use evidence from the case in your answers.

Case file: Northstar’s screening tool

Northstar is a fictional Canadian company that uses software to rank applicants for entry-level jobs. The software was trained on applications and hiring decisions from the company’s previous ten years. In that period, most people hired into technical roles were men. The tool learned to give lower scores to some applications that included activities more common among women, even when applicants had similar qualifications. It also favoured uninterrupted work histories, which could disadvantage people who had taken time away from paid work. After an internal review, Northstar compared the tool’s recommendations for two groups of equally qualified applicants. The figures below are fictional and are provided for this exercise. A high recommendation means the applicant would be invited to interview.

Applicant group

Recommended for interview

Group A: applications with traditionally male-coded activities

60% (60 out of 100)

Group B: applications with traditionally female-coded activities

30% (30 out of 100)

Investigate the evidence

Answer in complete thoughts. When a question asks you to explain, support your reasoning with details from the case or table.

1.What kind of unfair pattern is suggested by the case: bias related to gender, age, or something else? Identify the evidence that supports your choice.

2.Calculate the percentage-point gap between the two groups’ recommendation rates. What does this number tell Northstar?

3.Which explanation best describes how the tool could have learned this pattern?

It learned from past hiring decisions that reflected a workforce dominated by men, and it treated patterns in applications as signals of likely success.
The computer independently decided that women should not be hired, without using any examples or rules.
The results prove that applicants in Group B had fewer qualifications.
The software’s use of numbers makes its decisions automatically fair.

4.True or false: If an algorithm does not ask applicants to state their gender, it cannot produce gender-related unfair outcomes. Explain your answer.

5.The tool also favours uninterrupted work histories. Name one group that could be disadvantaged by this rule and explain why.

6.A manager says, “The software is neutral because it follows data, not personal opinions.” Give one reason this claim is not enough to show the tool is fair.

7.Select two useful steps Northstar could take before using the tool again.

Audit recommendation rates across relevant groups and investigate differences.
Remove questionable signals, test changes, and check whether qualified applicants are treated more fairly.
Keep the tool secret so applicants cannot challenge it.
Assume a single successful test proves it will always be fair.

8.Who should share responsibility for making the hiring process fair? Choose all that apply and briefly explain your choices.

Northstar’s leaders and the team that builds or buys the tool
Applicants and affected communities, who can provide feedback and raise concerns
Public authorities that set and enforce human-rights and employment rules
Nobody, because software decisions cannot be changed

9.In Canada, human-rights protections apply to employment decisions, although the details depend on the jurisdiction. Explain why a hiring tool that disadvantages people because of a protected characteristic could raise a human-rights concern. Do not assume that the algorithm itself is exempt from responsibility.

10.Write a short recommendation to Northstar. State whether it should keep using the tool as it is, pause it, or use it only with safeguards. Include two safeguards and one way to check whether they work.