Case-Based Lesson: Ethical Decision-Making in AI Deployment

Case Overview

A mid-sized logistics company in Southeast Asia has deployed an AI-powered hiring tool to shortlist warehouse staff. Six months in, HR notices that female applicants and candidates from certain ethnic minorities are being systematically filtered out. The tool was trained on five years of historical hiring data — data that reflected the company’s previously homogeneous workforce. Leadership must now decide whether to suspend the tool, retrain it, or scrap it entirely, while managing pressure from investors who funded the AI initiative.

Lesson Flow

• Opening (10 min): Learners read the scenario individually. No discussion yet — they jot down their first instinct on what the company should do and why.
• Context Building (15 min): Facilitator briefly explains how training data bias works. No lecture — use two or three pointed examples to ground the case technically.
• Small Group Analysis (25 min): Groups of four examine the case from an assigned stakeholder lens — HR manager, investor, affected applicant, or AI vendor.
• Cross-Group Discussion (20 min): One representative per group shares their stakeholder position. Class maps areas of agreement and conflict on a shared board.
• Individual Synthesis (10 min): Each learner writes a revised recommendation, explaining what changed — or didn’t — in their thinking after discussion.

Guiding Questions

• What is the root cause of the bias — the data, the model, or the decision to deploy?
• At what point did the company have an obligation to intervene, and why didn’t they?
• How does each stakeholder define “harm” differently in this situation?
• If the tool is retrained, what guarantees that the problem won’t recur?
• Is suspending the tool the ethical choice, the legal choice, or both — and does that distinction matter?

Discussion Plan

The cross-group discussion is structured in two rounds. In the first round, each stakeholder group presents their position without interruption. In the second round, the facilitator introduces a constraint — the company’s legal team has advised that publicly admitting fault could trigger regulatory penalties — and groups must revisit their recommendations under this new pressure. The facilitator tracks shifting positions visibly and closes by asking the class: whose interests were centred, and whose were peripheral?

Assessment Method

Assessment is through the written synthesis produced in the final ten minutes. Learners are evaluated on three criteria:

• Reasoning quality: Does the recommendation follow logically from the evidence in the case?
• Perspective integration: Does the learner account for more than one stakeholder view?
• Position evolution: Does the learner reflect meaningfully on how discussion shaped — or challenged — their thinking?

[…]