AI0-001 AI Governance and Ethics Practice Question

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Use a different evaluation metric such as equalized odds

  Why it's wrong here

  Swapping to equalized odds changes which error-rate balance is measured, not the 0.2 versus 0.4 selection-rate gap itself, so the disparity persists. It is tempting because equalized odds is a legitimate fairness metric for comparing true-positive and false-positive rates across groups, and would be the right choice when the concern is error parity rather than selection parity.

• ✗

  Remove the sensitive attribute from the training data

  Why it's wrong here

  Dropping the sensitive attribute does not remove disparate impact, since correlated proxies such as postcode or job history still encode group membership. It is tempting because it is a recognised pre-processing fairness technique, and would be correct where no proxy variables exist and the attribute is genuinely non-predictive.

• ✗

  Collect more data for Group A to increase its representation

  Why it's wrong here

  Adding more Group A samples does not change the model's selection-rate disparity; demographic parity compares outcome rates, not dataset size. Collecting more data is appropriate when a group is underrepresented and estimates are noisy, but here the measured rates themselves differ, so the decision threshold or training labels must be adjusted.

• ✓

  Retrain the model with a fairness constraint that enforces demographic parity

  Why this is correct

  The 0.2 versus 0.4 selection rates show a demographic parity gap. Retraining with an explicit fairness constraint optimises the model to equalise selection rates across groups, directly targeting the measured disparity rather than adjusting thresholds post hoc.