Conduct Bias Checks (e.g., Model, Data, Algorithm) PMI-CPMAI Practice Questions (Page 1)

Questions 1–5

1. 1expert · hard

   A predictive policing model is found to flag a higher proportion of minority neighborhoods for patrol. The police department wants to investigate whether this is a discriminatory pattern or a reflection of actual crime rates. What should they do?

   • ACompare the model's predictions to actual crime rates in those neighborhoods
   • BRemove the neighborhood feature from the model
   • CImmediately stop using the model because it is biased
   • DRetrain the model with a fairness constraint
   Reveal Solution Discussion Select an answer first
2. 2expert · hard

   A multinational company deploys a hiring model across different countries. The model's bias metrics are monitored globally, but the company notices that the disparate impact ratio is acceptable overall, yet it is significantly worse in one country. The company wants to ensure fairness across all regions. What should they do?

   • ARetrain the model using only data from the problematic country
   • BAccept the global metric as sufficient because the overall ratio is within the acceptable range
   • CMonitor the disparate impact ratio separately for each country and set region-specific thresholds
   • DRemove the country attribute from the model to ensure fairness
   Reveal Solution Discussion Select an answer first
4. 3foundation · easy

   What is the purpose of bias mitigation techniques in AI model development?

   • ATo reduce identified biases in the model
   • BTo make the model more complex
   • CTo remove all protected attributes from the dataset
   • DTo increase the model's accuracy at any cost
   Reveal Solution Discussion Select an answer first
5. 4application · medium

   A government agency is developing a model to predict recidivism risk for parole decisions. The training data is collected from a region where a particular ethnic group is overrepresented in arrests due to historical policing practices. The agency wants to ensure the model does not perpetuate this bias. What should they do first?

   • AApply a bias mitigation technique like adversarial debiasing
   • BAnalyze the training data to determine the representation of each ethnic group and compare it to the general population
   • CUse a fairness metric like equalized odds to evaluate the model after training
   • DCollect more data from the overrepresented group to balance the dataset
   Reveal Solution Discussion Select an answer first
6. 5application · medium

   A healthcare provider uses an AI model to prioritize patients for follow-up care. The model was trained on historical data where a particular ethnic group had lower engagement with the healthcare system. The model now assigns lower priority scores to patients from that group. The provider wants to ensure the model is fair across all ethnic groups. Which approach should the team take?

   • AReview the model's predictions for discriminatory patterns by examining false negative rates across ethnic groups
   • BApply a fairness metric like equalized odds to compare false positive and false negative rates across ethnic groups
   • CPerform a calibration analysis to ensure that for a given predicted score, the actual risk is the same across groups
   • DConduct a demographic parity analysis to compare priority scores across ethnic groups
   Reveal Solution Discussion Select an answer first