An AI recruitment screening tool systematically scores applicants from one demographic group lower, reflecting patterns in its training data. Which risk is this, and how should governance respond?
- AA simple hardware failure, which governance addresses by replacing the servers that the tool runs on
- BAcceptable behaviour, because an AI tool cannot really be biased if it only follows its training data
- CBias leading to unfair outcomes, which governance addresses by monitoring for and managing harmful bias
- DA transparency issue only, which is fully solved just by telling applicants that an AI tool was used
Systematically lower scores for one group reflect harmful bias in the training data; governance responds by detecting and managing bias to ensure fair outcomes, not by disclosure alone.
Why each option is right or wrong.
- C (correct). Skewed outcomes reflecting training data are harmful bias; governance monitors for and manages it for fairness.
- A. This is not a hardware fault; the problem lies in the data and the model's learned patterns.
- B. Following training data does not make bias acceptable; biased data produces biased, unfair outcomes.
- D. Disclosure helps, but it does not correct the unfair scoring; the bias itself must be managed.
Reference: ITIL AI Governance (Version 5) syllabus, AI Benefits and Risks.