AI Assurance, Testing, and Auditing CRAGE Practice Questions (Page 7)

Questions 31–35

1. 31application · medium

   An organization is about to audit its AI recruitment tool. The audit team has defined the scope and gathered relevant documentation. According to standard AI auditing processes, what should they do next?

   • AConduct the audit fieldwork, including testing and evidence collection.
   • BPublish the final audit report immediately.
   • CDefine the audit objectives and criteria.
   • DImplement remediation actions for any findings.
   Reveal Solution Discussion Select an answer first
2. 32application · medium

   An online lending platform is auditing its AI loan approval model. The audit team finds that the model's overall approval rate is similar across demographic groups, but the interest rates offered to one group are consistently higher. Which fairness testing approach would best reveal this issue?

   • ATest the model's robustness by adding noise to the input features.
   • BMeasure the equalized odds by comparing false-positive and false-negative rates across groups.
   • CAnalyze the distribution of predicted interest rates across demographic groups.
   • DReview the model's calibration curves for each group.
   Reveal Solution Discussion Select an answer first
4. 33application · medium

   A retail company is deploying an AI chatbot for customer service. The legal team requires assurance that the chatbot will not provide harmful or non-compliant responses. What is the primary goal of AI assurance in this context?

   • AMaximize the chatbot's response speed.
   • BEnsure the AI system is reliable, safe, and compliant with relevant regulations.
   • CIncrease the number of features the chatbot supports.
   • DReduce the cost of customer service operations.
   Reveal Solution Discussion Select an answer first
5. 34expert · hard

   A medical device company is seeking regulatory approval for an AI diagnostic tool. The regulator requires both high accuracy and explainability. The company has developed two models: Model A is a deep learning model with 95% accuracy but is a black box; Model B is a logistic regression with 88% accuracy but is fully interpretable. The company must choose one model to submit. Which consideration is most important in this decision?

   • AChoose Model B because interpretability is essential for regulatory approval in medical devices.
   • BSubmit both models and let the regulator decide.
   • CChoose Model A because higher accuracy is the primary requirement for regulatory approval.
   • DChoose Model A and use post-hoc explanation methods to satisfy the explainability requirement.
   Reveal Solution Discussion Select an answer first
6. 35application · medium

   A logistics company uses an AI model to predict delivery delays. After deployment, the model's accuracy degrades over time because of seasonal changes. What is the most appropriate continuous assurance action?

   • ARetrain the model every day regardless of performance.
   • BMonitor model accuracy and retrain the model when accuracy falls below a defined threshold.
   • CIgnore the degradation because it is seasonal and will recover.
   • DStop using the model during seasonal changes.
   Reveal Solution Discussion Select an answer first