Conduct Bias Checks (e.g., Model, Data, Algorithm) PMI-CPMAI Practice Questions (Page 4)

Questions 16–20

1. 16expert · hard

   A model is used to approve or deny loan applications. The model's overall accuracy is 95%, but the false negative rate for applicants over 60 is 20% versus 5% for younger applicants. The company wants to ensure fairness for older applicants. Which fairness testing method should they use?

   • ADemographic parity, to compare approval rates between age groups
   • BPredictive parity, to ensure that the positive predictive value is equal across age groups
   • CCalibration, to ensure that predicted probabilities match actual outcomes for each age group
   • DEqualized odds, to compare false negative and false positive rates between age groups
   Reveal Solution Discussion Select an answer first
2. 17foundation · easy

   What is the main purpose of bias detection metrics in an AI system?

   • ATo eliminate all bias from the model permanently
   • BTo quantify and track bias over time
   • CTo increase the complexity of the model
   • DTo replace human oversight in AI governance
   Reveal Solution Discussion Select an answer first
4. 18application · medium

   An e-commerce company deploys a recommendation model that shows job ads. They notice that the model recommends high-paying job ads to male users more often than to female users. The company wants to monitor this bias over time. What should they implement?

   • AA periodic manual review of model outputs to identify any obvious discriminatory patterns
   • BA monitoring system that calculates the disparate impact ratio for job ad categories on a weekly basis
   • CA system that automatically removes any user from the recommendation pool if they are in a protected group
   • DA dashboard that tracks the demographic composition of users who see high-paying job ads
   Reveal Solution Discussion Select an answer first
5. 19application · medium

   A tech company develops a resume screening model. They want to ensure that the model does not discriminate against candidates based on gender. The model's overall accuracy is high, but they suspect it may be biased. Which fairness testing method should they use to evaluate the model?

   • ACompare the model's predictions against a random baseline
   • BTest the model on a small sample of resumes to see if it makes obvious errors
   • CReview the model's feature importance to see if gender is a top feature
   • DCalculate the demographic parity by comparing the selection rates for male and female candidates
   Reveal Solution Discussion Select an answer first
6. 20application · medium

   A university uses an AI system to screen student applications. The system assigns a 'success score' to each applicant. The admissions team notices that applicants from rural high schools consistently receive lower scores than applicants from urban high schools with similar academic records. What should the team do to investigate this pattern?

   • ARetrain the model using only academic performance as a feature
   • BReview the model's predictions for applicants from rural versus urban schools, controlling for academic performance
   • CCalculate the overall accuracy of the model on the entire applicant pool
   • DRemove the 'high school location' feature from the model
   Reveal Solution Discussion Select an answer first