Transparency CRAGE Practice Questions (Page 7)

Questions 31–35

1. 31expert · hard

   A team is developing an AI system to detect fraudulent insurance claims. The system must be transparent to internal auditors, but the team also wants to maximize detection accuracy. The auditors need to understand why a claim was flagged. Which approach best satisfies both needs?

   • AUse a highly accurate black-box model and provide auditors with post-hoc explanations for each flagged claim
   • BUse a simple rule-based system that is fully interpretable but less accurate
   • CUse a black-box model and document the model's overall performance metrics for auditors
   • DUse a black-box model and only provide auditors with the final fraud score
   Reveal Solution Discussion Select an answer first
2. 32expert · hard

   A company has deployed a deep learning model for medical diagnosis. The model is highly accurate, but clinicians are reluctant to use it because they cannot understand why it makes certain predictions. The company must improve clinician trust without significantly reducing the model's accuracy. Which approach is most effective?

   • ARequire clinicians to use the model without explanation until they become accustomed to it
   • BProvide clinicians with post-hoc explanations such as saliency maps or SHAP values for each prediction
   • CPublish the model's training data and code to demonstrate its validity
   • DReplace the deep learning model with a simpler logistic regression model
   Reveal Solution Discussion Select an answer first
4. 33application · medium

   A hospital deploys an AI system that predicts patient readmission risk. The hospital's ethics committee wants to communicate the system's capabilities and limitations to both physicians and patients. What is the most effective approach?

   • APublish the model's source code on the hospital's public website
   • BSend the same detailed technical report to both physicians and patients
   • CProvide physicians with a technical validation report and patients with a plain-language brochure
   • DHold a single town hall meeting for all stakeholders to answer questions
   Reveal Solution Discussion Select an answer first
5. 34application · easy

   A nonprofit organization uses an AI system to allocate donations to various projects. The organization wants to build trust with donors by being transparent about how the AI makes allocation decisions. What is the primary benefit of transparency in this context?

   • AIt guarantees the AI system will make perfect decisions
   • BIt ensures the AI system is profitable
   • CIt allows the organization to avoid all regulatory requirements
   • DIt builds trust and accountability with donors and stakeholders
   Reveal Solution Discussion Select an answer first
6. 35application · medium

   A data science team is choosing between two models for a credit scoring system. Model A is a logistic regression that is easy to understand but slightly less accurate. Model B is a gradient boosting model that is more accurate but difficult to interpret. The compliance team requires that the model's decisions be explainable to auditors. Which choice best balances accuracy and transparency?

   • AChoose Model B and provide auditors with the model's feature importance scores
   • BChoose Model B and rely on post-hoc explainability tools to generate explanations
   • CChoose Model A because its inherent interpretability makes it easier to explain to auditors
   • DChoose Model A but do not document its decision logic because it is inherently transparent
   Reveal Solution Discussion Select an answer first