Principles of Ethical and Fair AI Use AIE Practice Questions (Page 10)

Questions 46–49

1. 46application · medium

   A bank uses an AI system to detect fraudulent transactions. The system flags some legitimate transactions as fraudulent, causing inconvenience to customers. The bank wants to ensure accountability for the system's errors. Which measure is most appropriate?

   • AEstablish a process for customers to dispute flagged transactions and have them reviewed by a human.
   • BAutomatically block all flagged transactions without customer notification.
   • CRely on the AI system's high accuracy and ignore occasional errors.
   • DDisclose to customers that the system may make errors but offer no recourse.
   Reveal Solution Discussion Select an answer first
2. 47application · medium

   A financial institution deploys an AI system that approves or denies loan applications. The system is developed by an external vendor. The institution's compliance officer wants to ensure accountability for the system's decisions. Which measure is most effective?

   • ADisclose to customers that the decision was made by AI.
   • BEstablish a contract with the vendor that defines roles, responsibilities, and audit rights.
   • CRely on the vendor's internal testing and certification.
   • DRequire all loan decisions to be manually reviewed by a human.
   Reveal Solution Discussion Select an answer first
4. 48application · medium

   A bank is using an AI system to detect credit card fraud. The system blocks transactions that appear fraudulent. A customer's legitimate transaction was blocked, and the customer demands an explanation. The bank's customer service team needs to provide a clear reason. What is the best way to ensure explainability?

   • AUse a model that provides a reason code, such as 'transaction amount exceeds typical pattern'.
   • BRefuse to explain because revealing the model's logic could help fraudsters evade detection.
   • CProvide the customer with the exact fraud score and the model's internal weights.
   • DTell the customer that the system flagged the transaction as suspicious and that is the only information available.
   Reveal Solution Discussion Select an answer first
5. 49expert · hard

   An insurance company uses an AI model to set premiums. The model uses a wide range of data, including credit score, driving record, and geographic location. A regulator finds that the model results in higher premiums for residents of certain zip codes, which correlate with racial demographics. The company argues that the model is accurate and does not use race as a feature. Which action best addresses the fairness concern while maintaining the model's business purpose?

   • ARetain the model but add a post-processing step that adjusts premiums to ensure no protected group is disproportionately burdened.
   • BRemove all geographic features and retrain the model without them.
   • CReplace the model with a simple rule-based system that uses only driving record.
   • DArgue that the model is fair because it does not use race as an input.
   Reveal Solution Discussion Select an answer first