AI Assurance, Testing, and Auditing CRAGE Practice Questions (Page 11)

Questions 51–53

1. 51foundation · easy

   In AI risk assessment, what does the term 'likelihood' refer to?

   • AThe severity of consequences if the risk materializes
   • BThe probability that a specific risk event will occur
   • CThe potential impact of a risk event on the organization
   • DThe number of risk events identified in the system
   Reveal Solution Discussion Select an answer first
2. 52application · medium

   A bank deploys a credit-scoring model. During validation, the model shows a 12% approval-rate gap between two demographic groups. The compliance team asks for evidence that the model does not discriminate. Which testing approach best addresses this concern?

   • APerform load testing to ensure the model responds within 200 ms under peak transaction volume.
   • BRun a differential outcome analysis across protected groups and compare approval rates against a pre-defined fairness threshold.
   • CReview the model's feature-importance scores to identify which variables drive predictions.
   • DConduct a robustness test by adding small perturbations to input data to check prediction stability.
   Reveal Solution Discussion Select an answer first
3. 53expert · hard

   A city is deploying an AI system to allocate police resources. The system is trained on historical crime data. The city council is concerned about potential racial bias. The data science team finds that the model is accurate overall but has higher false-positive rates in certain neighborhoods. The team must decide how to address this. Which approach is most appropriate?

   • ARetrain the model with a fairness constraint that equalizes false-positive rates across neighborhoods.
   • BRemove all demographic data from the model to ensure it is 'blind' to race.
   • CReplace the AI system with a human-only decision process.
   • DAccept the false-positive disparity because the model is accurate overall.
   Reveal Solution Discussion Select an answer first