Testing for Algorithmic, Sample and Inappropriate Bias CT-AI Practice Questions (Page 1)

Questions 1–5

1. 1application · medium

   A hospital uses an AI system to prioritize patients for organ transplants. The system was trained on historical data that reflects past decisions. An audit reveals that the system consistently assigns lower priority scores to patients from a particular ethnic group, even when clinical factors are identical. The hospital's ethics board is concerned about discriminatory outcomes. What is the most appropriate immediate action for the testing team?

   • ATemporarily suspend the system and perform a root-cause analysis to identify the source of the bias.
   • BRetrain the model with more data from the affected ethnic group.
   • CAdjust the model's decision threshold to equalize priority scores across all ethnic groups.
   • DConduct a bias impact assessment and report the findings to the ethics board.
   Reveal Solution Discussion Select an answer first
2. 2application · medium

   A team is testing a facial recognition system for a security application. The training data was collected from public web images, which are predominantly of light-skinned individuals. The team wants to test for sample bias. Which testing approach would be most effective?

   • AEvaluate the model on a test set that is balanced across skin tones and compare accuracy per subgroup.
   • BTest the model on the same training data to check for overfitting.
   • CReview the model's architecture to ensure it uses appropriate convolutional layers.
   • DUse a larger test set with more images of light-skinned individuals.
   Reveal Solution Discussion Select an answer first
4. 3foundation · easy

   How does algorithmic bias differ from sample bias in an AI system?

   • AAlgorithmic bias is caused by the model's design or learning process, while sample bias is caused by non-representative training data.
   • BAlgorithmic bias is a type of sample bias that occurs during data collection.
   • CAlgorithmic bias affects only the model's outputs, while sample bias affects only the model's inputs.
   • DAlgorithmic bias is always intentional, while sample bias is always unintentional.
   Reveal Solution Discussion Select an answer first
5. 4expert · hard

   A health insurance company uses an AI model to predict healthcare costs. The model is found to under-predict costs for patients with chronic conditions, leading to inadequate coverage. The company is concerned about the impact on these patients. The testing team must decide how to assess the bias. Which approach would be most appropriate?

   • AConduct a focus group with patients to understand their concerns.
   • BRetrain the model with more data from patients with chronic conditions.
   • CReview the model's feature engineering to see if chronic conditions are properly represented.
   • DMeasure the model's prediction error separately for patients with and without chronic conditions.
   Reveal Solution Discussion Select an answer first
6. 5application · medium

   A bank is developing a credit-scoring model. The training data was collected from loan applications submitted through the bank's mobile app over the past two years. The model performs well on the test set, which was also drawn from the same app. However, when deployed, the model shows significantly lower accuracy for applicants who apply through physical branches. Which type of bias is most directly indicated, and what is the most appropriate testing technique to detect it?

   • AAlgorithmic bias; perform a code review of the model's feature engineering logic.
   • BInappropriate bias; conduct a fairness audit comparing approval rates across age groups.
   • CSample bias; retrain the model with more features from the mobile app data.
   • DSample bias; evaluate model performance on a stratified holdout set that includes branch applicants.
   Reveal Solution Discussion Select an answer first