Ethical, Societal, and Security Concerns in AI AIE Practice Questions (Page 7)

Questions 31–35

1. 31application · medium

   A bank deploys an AI model to approve small business loans. During routine monitoring, the data science team discovers that an attacker has been submitting thousands of fraudulent loan applications designed to shift the model's decision boundary so that more high-risk applicants get approved. The bank's compliance officer wants to prevent this type of attack and maintain fairness in lending decisions. Which two-pronged approach should the bank implement?

   • ARestrict access to the model's training data and retrain only with internal historical data.
   • BImplement adversarial training and add input validation to reject anomalous applications.
   • CIncrease the model's complexity so it can better distinguish legitimate from fraudulent applications.
   • DAdd a fairness constraint to the model's loss function and re-run the current training data.
   Reveal Solution Discussion Select an answer first
2. 32application · medium

   An autonomous vehicle company uses AI for object detection. A security researcher demonstrates that placing a small sticker on a stop sign can cause the AI to misclassify it as a speed limit sign. The company wants to make the system more robust. Which approach is most appropriate?

   • ARely on multiple sensors and fuse their outputs.
   • BAdd a rule that ignores small objects on signs.
   • CUse a higher-resolution camera to capture more details.
   • DTrain the model with adversarial examples that include sticker-like perturbations.
   Reveal Solution Discussion Select an answer first
4. 33foundation · easy

   Which mitigation strategy is specifically aimed at making AI systems more resilient to adversarial attacks and other security threats?

   • AGovernance
   • BRobust design
   • CImpact assessment
   • DRegulation
   Reveal Solution Discussion Select an answer first
5. 34expert · hard

   A financial regulator is reviewing an AI model used by banks to assess credit risk. The model is a deep neural network with high accuracy, but it is opaque. The regulator requires that banks be able to explain individual credit decisions to customers. The banks argue that the model's accuracy would drop if they switched to a simpler, interpretable model. What is the most practical solution?

   • AUse a surrogate model (e.g., LIME or SHAP) to generate local explanations for each decision.
   • BAllow banks to keep the model but prohibit them from explaining decisions to customers.
   • CRequire banks to publish the model's weights and architecture.
   • DForce all banks to use only linear regression models.
   Reveal Solution Discussion Select an answer first
6. 35application · medium

   A retail chain introduces an AI scheduling system that predicts customer traffic and automatically generates employee work schedules. Employees complain that the system often schedules them for split shifts with long unpaid breaks, reducing their effective hourly wage and making it hard to plan childcare. The chain wants to maintain the efficiency gains from the system while addressing employee concerns. What should the chain do?

   • APublish the algorithm's code so employees can see how schedules are generated
   • BAdd constraints to the scheduling algorithm to minimize split shifts and provide a minimum number of consecutive hours
   • CLet employees manually swap shifts after the schedule is published
   • DReplace the AI system with a human scheduler who uses the same traffic predictions
   Reveal Solution Discussion Select an answer first