AI Assurance, Testing, and Auditing CRAGE Practice Questions (Page 9)

Questions 41–45

1. 41application · medium

   A government agency is planning an audit of an AI system used for benefit eligibility determinations. The audit team has limited time and must focus on the highest-risk aspects. The system uses a complex model and processes sensitive personal data. Which step should the audit team perform first to ensure the audit is scoped effectively?

   • AConduct a risk assessment to identify the areas of highest potential harm and regulatory exposure.
   • BCollect all model training data and labels for independent re-evaluation.
   • CSelect an assurance framework, such as NIST AI RMF, to guide the audit process.
   • DInterview the system's developers to understand the model's architecture.
   Reveal Solution Discussion Select an answer first
2. 42application · medium

   A company is deploying an AI system that makes hiring decisions. The risk assessment team must identify potential risks. Which risk is most directly related to AI assurance?

   • AThe server hosting the model may experience downtime.
   • BThe marketing team may not promote the product effectively.
   • CThe model may produce biased outcomes that lead to discriminatory hiring practices.
   • DThe development team may exceed the project budget.
   Reveal Solution Discussion Select an answer first
4. 43application · medium

   A credit union uses a gradient-boosting model for loan approvals. Members are denied loans and demand explanations. The credit union must provide understandable reasons for each denial. Which testing approach should be used to validate that the model can produce such explanations?

   • ARobustness testing by adding noise to input data.
   • BPerformance testing to ensure the model runs within acceptable latency.
   • CExplainability testing using SHAP values to verify that each denial can be attributed to specific input features.
   • DBias testing across demographic groups.
   Reveal Solution Discussion Select an answer first
5. 44foundation · easy

   What is the typical final step in an AI auditing process?

   • ADefining the audit scope and objectives
   • BImplementing corrective actions
   • CCollecting audit evidence and data
   • DReporting findings and recommendations to stakeholders
   Reveal Solution Discussion Select an answer first
6. 45application · medium

   A team is testing a facial-recognition system for a security application. They need to verify that the system works correctly for different skin tones and lighting conditions. Which combination of testing methods is most appropriate?

   • AFunctional testing with diverse datasets and robustness testing under varying lighting conditions.
   • BExplainability testing using saliency maps.
   • CBias testing only, comparing error rates across groups.
   • DPerformance testing only, measuring throughput.
   Reveal Solution Discussion Select an answer first