AI Capabilities, Data Dependencies, and Failure Modes CAIPM Practice Questions (Page 2)

Questions 6–10

1. 6application · medium

   A city government wants to use AI to optimize public bus schedules based on ridership data. The data is collected from sensors on buses, but some sensors are faulty and produce inconsistent readings. The city has a limited budget for data collection. What should the city do first?

   • AClean the existing data by identifying and correcting sensor errors before training the model.
   • BUse a model that is robust to noisy data, such as a decision tree.
   • CCollect more data from additional sensors to overwhelm the errors.
   • DIgnore the faulty sensors and use only data from reliable sensors.
   Reveal Solution Discussion Select an answer first
2. 7application · medium

   A retail bank wants to deploy an AI system to automatically read and categorize incoming customer emails (e.g., loan inquiries, fraud reports, account issues) and route them to the correct department. The bank has a large archive of historical emails with labels assigned by customer service agents. Which AI capability is most directly required, and what is the primary data dependency for this system?

   • ANatural language processing; the historical emails must be representative of the variety and language of future incoming emails.
   • BReasoning; the system must have a complete rule set that covers every possible email scenario.
   • CComputer vision; the system needs enough high-resolution images of emails to learn visual patterns.
   • DLearning; the system needs a continuous stream of unlabeled emails to self-supervise its own routing decisions.
   Reveal Solution Discussion Select an answer first
4. 8foundation · easy

   An AI model trained on historical sales data performs well in backtesting but poorly on new, unseen data. Which data dependency is most likely the cause?

   • AData quality
   • BData storage
   • CData quantity
   • DData relevance
   Reveal Solution Discussion Select an answer first
5. 9application · medium

   A healthcare organization is deploying an AI system to predict patient readmission risk. The model was trained on historical data that underrepresents patients from certain ethnic backgrounds. The organization wants to ensure the model does not perpetuate bias. Which approach is most effective?

   • ARemove sensitive attributes such as ethnicity from the model's input features.
   • BDeploy the model immediately and monitor for bias after a few months.
   • CUse a more complex model to capture subtle patterns in the data.
   • DCollect more representative data and retrain the model, then evaluate performance across patient subgroups.
   Reveal Solution Discussion Select an answer first
6. 10expert · hard

   A government agency uses an AI system to screen visa applications. The system was trained on historical data that reflects past immigration patterns. The agency is concerned about bias against applicants from certain countries. They have a limited budget and must choose a mitigation strategy. Which strategy is most likely to reduce bias effectively?

   • ARemove country of origin from the model's input features.
   • BUse a more complex model to capture subtle patterns in the data.
   • CDeploy the model and monitor for bias after a year.
   • DConduct a bias audit using fairness metrics and then retrain the model with fairness constraints.
   Reveal Solution Discussion Select an answer first