The “Fair” Hiring AI
A company uses an AI system to pre-screen job applications.
After one year, the statistics show: 90 % of the invited candidates are men.
The AI was trained on historical data (the last 10 years).
The developers say: “The AI doesn’t discriminate. It just learns from data.”
The management says: “The outcome is unfair.”
Question:
Who is responsible — and what would be a sensible, concrete measure?
The solution will be unlocked tomorrow.
Solution
The AI is not morally responsible. The responsibility lies with the company (and internally: leadership and the team building/operating the system).
“It just learns from data” explains the result, but it doesn’t excuse it: historical data can encode historical disadvantage.
Concrete measure (practical):
• Audit: identify which signals drive decisions (career paths, gaps, keywords).
• Define a fairness goal (e.g., equal opportunity at equal qualification).
• Improve process: bias mitigation + human-in-the-loop for borderline cases.
• Monitor outcomes regularly to see whether invitation rates improve without hurting quality.
Key line:
Data explains — humans are accountable.