Test the features for proxy discrimination: Considering nondiscrimination and adverse-action expectations
Test proxy features for unjustified disparate impact, remedy them, and give applicants the specific principal reasons for each denial.
The question
A lender's AI uses ZIP code and shopping patterns as features; regulators note these correlate with a protected characteristic, and applicants ask why they were denied. Considering nondiscrimination and adverse-action expectations together, which response is most defensible?
Preparing for AIGP? Take the free 5-min readiness quiz →
- Keep the features because they boost accuracy and refuse to explain individual denials, since a proprietary model's internal logic is a trade secret shielded from any applicant disclosure.Plausible on accuracy, but proxy features correlated with a protected class can create disparate impact, and lending contexts generally require meaningful reasons for adverse decisions.
- Add the protected characteristic explicitly as a feature to measure and offset its effect, then treat the presence of that variable as conclusive proof the model is now fully fair.Almost right that measuring disparities matters, but simply adding a protected attribute is not conclusive proof of fairness and can itself raise legal concerns.
- Replace individual explanations with a single aggregate fairness statistic published annually, since a portfolio-level metric fully satisfies both the disparate-impact and adverse-action expectations.Plausible-sounding but an aggregate metric does not substitute for the individualized adverse-action reasons that applicants are generally owed.
- Test the features for proxy discrimination, drop or justify those causing unjustified disparate impact, and provide applicants with the specific principal reasons for each denial. ✓Correct: nondiscrimination law reaches proxy features producing unjustified disparate impact, and lending norms require giving applicants the specific reasons for an adverse decision.
The trap
Believing accuracy gains or a portfolio-level fairness metric excuses proxy discrimination and skipping individual reasons. How to remember it
Test proxy features for unjustified disparate impact, remedy them, and give applicants the specific principal reasons for each denial.
How many of these would you get right?
One of 1581 AIGP questions on Certsqill. Take a free five-minute check and see your score per domain — not one number, but which section to open tonight.
Test your AIGP readiness — freeMore Understanding How Laws, Standards and Frameworks Apply to AI questions
- The research-only licence does not authorize commercial: Considering intellectual property law, which →
- Substantiate or drop the lowest-price claim: Under consumer protection law, which governance step is most →
- A defect rooted in training data can be a design: Considering product liability law, which statement is most →
- All 394 Understanding How Laws, Standards and Frameworks Apply to AI questions →
Part of the Certsqill AIGP question bank · Understanding How Laws, Standards and Frameworks Apply to AI ·
Every answer, right and wrong, comes with its own explanation.
