Select a model whose decisions can be explained: Which design choice best applies responsible-AI policies and
When affected individuals must receive reasons and oversight is required, responsible design favors an explainable, reviewable model even at a bounded accuracy cost.
The question
A bank is designing an AI system to approve or deny loan applications. A complex model offers the highest accuracy but its decisions cannot be explained to rejected applicants, who by policy must receive a reason. The design must reconcile predictive quality, explainability to affected individuals, and meaningful human oversight. Which design choice best applies responsible-AI policies and ethical considerations?
Preparing for AIGP? Take the free 5-min readiness quiz →
- Select a model whose decisions can be explained and reviewed by a human, accepting a bounded accuracy trade-off to meet the explainability and oversight requirements. ✓Correct because when affected individuals must receive reasons and oversight is required, responsible design favors an explainable, reviewable model even at a bounded cost to raw accuracy.
- Deploy the highest-accuracy model as is and generate plausible post-hoc rationales for denials so applicants receive a reason that reads convincingly to them.Plausible because it preserves accuracy and appears to satisfy the reason requirement, but fabricated post-hoc rationales are not faithful explanations and breach transparency and ethics duties.
- Deploy the complex model but restrict its use to only the highest-value applications, on the assumption that fewer decisions reduce the overall explainability obligation.Plausible because narrowing scope can reduce exposure, but every affected applicant is still owed a reason, so limiting volume does not resolve the explainability and oversight gap.
- Deploy the complex model and rely on aggregate fairness statistics to demonstrate the system is non-discriminatory across the applicant population as a whole.Plausible because fairness monitoring is valuable, but population-level statistics do not give an individual applicant the explanation and human review the policy requires.
The trap
Assuming raw accuracy can be preserved with fabricated post-hoc rationales or aggregate fairness stats, when each affected individual is owed a faithful, reviewable explanation. How to remember it
When affected individuals must receive reasons and oversight is required, responsible design favors an explainable, reviewable model even at a bounded accuracy cost.
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 to Govern AI Development questions
- The assessment is incomplete because it omits: As reviewer, what is the most defensible conclusion? →
- Remove the name feature from the design so the identified: Applying a risk mitigation hierarchy that prefers →
- Record each significant design decision with its rationale: Which documentation practice during design and →
- All 426 Understanding How to Govern AI Development questions →
Part of the Certsqill AIGP question bank · Understanding How to Govern AI Development ·
Every answer, right and wrong, comes with its own explanation.
