AI Assurance, Testing, and Auditing CRAGE Practice Questions (Page 4)

Questions 16–20

1. 16foundation · easy

   What is the primary purpose of continuous monitoring in AI systems post-deployment?

   • ATo replace the need for periodic audits
   • BTo ensure the model never makes a mistake
   • CTo detect performance degradation, drift, and emerging issues over time
   • DTo automatically retrain the model without human intervention
   Reveal Solution Discussion Select an answer first
2. 17application · medium

   An insurance company uses an AI model to set premiums. A review shows that the model charges higher premiums to residents of certain postal codes, which correlate with race. The company wants to ensure fairness. What should they do?

   • AConduct a bias audit to determine if the postal-code variable is a proxy for race and assess the impact on protected groups.
   • BImmediately remove the postal-code variable from the model.
   • CIgnore the correlation because postal codes are not a protected class.
   • DIncrease premiums for all customers to balance the difference.
   Reveal Solution Discussion Select an answer first
4. 18foundation · easy

   What is the purpose of bias testing in AI systems?

   • ATo confirm that the model's predictions are always correct
   • BTo verify that the model's code is free of syntax errors
   • CTo identify and measure systematic differences in model outcomes across protected groups
   • DTo ensure the model runs efficiently on all hardware platforms
   Reveal Solution Discussion Select an answer first
5. 19expert · hard

   A company is adopting an AI assurance framework to guide its auditing processes. The company operates in a highly regulated industry and needs to demonstrate compliance to multiple regulators. The company's AI systems vary widely in risk, from low-risk chatbots to high-risk medical diagnostics. Which framework adoption strategy is most effective?

   • AApply the same framework requirements uniformly to all AI systems to ensure consistency.
   • BAdopt multiple frameworks and apply them all to every system to maximize coverage.
   • CUse the framework as a baseline and scale the depth of assurance activities based on the risk level of each system.
   • DUse the framework only for high-risk systems and rely on internal best practices for low-risk ones.
   Reveal Solution Discussion Select an answer first
6. 20expert · hard

   A bank has deployed an AI system for credit risk assessment. The system's performance is monitored quarterly. Recently, the bank's compliance officer noticed that the model's predictions have become less stable, with frequent changes in risk scores for the same customer. The data science team suspects the model is being retrained too often. Which action would best address the instability while maintaining model accuracy?

   • AReduce the retraining frequency and add a stability check before deploying each new model version.
   • BSwitch to a more complex model that can capture more patterns.
   • CIncrease the retraining frequency to ensure the model adapts to the latest data.
   • DStop retraining the model and use the original version permanently.
   Reveal Solution Discussion Select an answer first