Testing for Algorithmic, Sample and Inappropriate Bias CT-AI Practice Questions (Page 5)

Questions 21–25

1. 21application · medium

   A university uses an AI system to predict student success and allocate scholarships. The system is found to give lower predictions to students from certain high schools, even when their grades are similar. The university is concerned about the fairness of the system. What is the most appropriate testing technique to assess whether this is inappropriate bias?

   • ARetrain the model with more data from the affected high schools.
   • BReview the model's training data to see if certain high schools are underrepresented.
   • CCompare the model's predictions across high school groups while controlling for grades.
   • DConduct a survey to ask students from those high schools about their experience.
   Reveal Solution Discussion Select an answer first
2. 22expert · hard

   A team is testing a recommendation system for an e-commerce platform. The model is found to recommend higher-priced products to users from certain postal codes. The training data is balanced across postal codes. The team suspects algorithmic bias. Which testing technique would be most effective in confirming this?

   • AReview the model's feature importance to see if postal code is a top predictor.
   • BRun a fairness audit comparing the average price of recommended products across postal codes.
   • CTest the model with a synthetic dataset where postal codes are randomly assigned.
   • DRetrain the model without postal code as a feature.
   Reveal Solution Discussion Select an answer first
4. 23foundation · easy

   Which stakeholder group is most directly affected by inappropriate bias in a credit scoring model?

   • ARegulators who enforce data protection laws.
   • BThe developers who wrote the code for the model.
   • CThe cloud provider hosting the model.
   • DApplicants who are denied loans due to discriminatory patterns in the model's decisions.
   Reveal Solution Discussion Select an answer first
5. 24application · medium

   A team is testing a loan approval model. They want to ensure that the model does not discriminate based on gender. The model does not use gender as a feature, but it uses 'occupation' and 'zip code', which may correlate with gender. Which testing technique would be most effective in detecting this indirect bias?

   • AReview the model's feature importance to see if 'occupation' and 'zip code' are highly weighted.
   • BRemove 'occupation' and 'zip code' from the model.
   • CTest the model with a synthetic dataset where gender is explicitly included.
   • DRun a fairness audit that measures approval rates across gender groups, even though gender is not a feature.
   Reveal Solution Discussion Select an answer first
6. 25foundation · easy

   What distinguishes inappropriate bias from other types of bias in AI systems?

   • AInappropriate bias only occurs in models that use protected attributes as input features.
   • BInappropriate bias specifically leads to unfair or discriminatory outcomes that violate ethical or legal standards.
   • CInappropriate bias is always caused by malicious intent of the developers.
   • DInappropriate bias is a measure of how much the model's predictions deviate from the training data.
   Reveal Solution Discussion Select an answer first