AI Capabilities, Data Dependencies, and Failure Modes CAIPM Practice Questions (Page 9)

Questions 41–43

1. 41application · medium

   A financial services firm deploys a credit scoring model that gives lower scores to applicants from a particular postal code. The firm's data scientists find that the model is using postal code as a proxy for income, which correlates with the applicants' race. What is the most appropriate mitigation?

   • AAdd more demographic data to the model to improve accuracy.
   • BRemove postal code from the model's input features.
   • CRetrain the model with fairness constraints that reduce disparate impact across groups.
   • DIgnore the issue because the model is accurate overall.
   Reveal Solution Discussion Select an answer first
2. 42application · medium

   A security company wants to deploy an AI system to detect anomalies in network traffic that may indicate cyberattacks. The system must learn normal patterns and flag deviations. The company has a large dataset of network traffic, but it contains many false labels due to outdated attack signatures. What is the most important consideration for this AI system?

   • AData availability; the system needs real-time traffic feeds to detect attacks.
   • BData quality; the labels must be accurate and up-to-date to train a reliable anomaly detector.
   • CData quantity; the system needs even more data to compensate for the label errors.
   • DData relevance; the system should focus only on attack traffic, not normal traffic.
   Reveal Solution Discussion Select an answer first
3. 43expert · hard

   An e-commerce company deploys a recommendation system that was trained on purchase data from the previous year. After six months, the system's click-through rate drops significantly. Analysis shows that the distribution of product categories browsed by users has shifted, but the model's parameters have not changed. Which failure mode is occurring, and what is the most appropriate first step to address it?

   • AOverfitting; the model memorized the old data, so reduce model complexity.
   • BModel degradation; the model's weights are decaying, so reinitialize and retrain from scratch.
   • CBias; the model is biased against new products, so add fairness constraints.
   • DData drift; retrain the model with recent data to adapt to the new distribution.
   Reveal Solution Discussion Select an answer first