Testing for Algorithmic, Sample and Inappropriate Bias CT-AI Practice Questions (Page 2)

Questions 6–10

1. 6foundation · easy

   Which outcome is an example of inappropriate bias in an AI system?

   • AA spam filter correctly identifies 95% of spam emails and 5% of legitimate emails as spam.
   • BA hiring algorithm consistently ranks male candidates higher than equally qualified female candidates.
   • CA model's accuracy is lower on new data than on training data due to overfitting.
   • DA recommendation system suggests different products to users based on their past purchases.
   Reveal Solution Discussion Select an answer first
2. 7expert · hard

   A team is testing a credit risk model. They have identified potential sample bias because the training data under-represents young adults. They also suspect algorithmic bias because the model may use proxies for age. The team has limited resources and must choose a testing strategy. Which strategy would be most effective?

   • AImmediately retrain the model with a balanced dataset and then test for accuracy.
   • BRun a fairness audit first, and if bias is detected, assume it is sample bias and collect more data.
   • CUse a single fairness metric, such as equalized odds, to assess the model.
   • DFirst, analyze the training data to confirm sample bias, then run a fairness audit to detect algorithmic bias.
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   A team is testing a resume-screening AI. They find that the model ranks candidates lower if their names are associated with a particular ethnicity. The training data was balanced across ethnic groups, and the model's accuracy is similar across groups. The team suspects the bias comes from the model learning subtle patterns in the text, such as names of universities or clubs that correlate with ethnicity. Which type of bias is this, and what testing technique would best reveal it?

   • AAlgorithmic bias; perform a fairness audit using metrics like equalized odds across ethnic groups.
   • BAlgorithmic bias; use a code review to check for hardcoded ethnicity variables.
   • CInappropriate bias; review the model's decision rules for explicit discriminatory criteria.
   • DSample bias; use a test set with more candidates from the affected ethnicity.
   Reveal Solution Discussion Select an answer first
5. 9expert · hard · select all that apply

   A company uses an AI system to screen job applicants. The system is found to rank applicants from certain universities lower, even when their qualifications are similar. The company is concerned about inappropriate bias. Select all that apply: Which testing techniques would be appropriate to assess this bias?

   • ARun a disparate impact analysis on the final hiring decisions.
   • BConduct a user acceptance test with a sample of applicants.
   • CCompare the model's ranking outcomes across university groups, controlling for qualifications.
   • DReview the model's training data to see if certain universities are underrepresented.
   • ERetrain the model without university as a feature.
   Check answers Discussion Select an answer first
6. 10application · medium

   An insurance company uses an AI model to set premiums. A testing team discovers that the model charges higher premiums to customers in certain postal codes, which correlate with lower-income neighborhoods. The company's legal team warns that this could be considered discriminatory. What is the most appropriate testing technique to assess the impact of this bias?

   • AReview the model's feature importance to see if postal code is a top predictor.
   • BPerform a disparate impact analysis comparing premium rates across postal code groups.
   • CRetrain the model without postal code as a feature.
   • DConduct a user acceptance test with a sample of customers from different postal codes.
   Reveal Solution Discussion Select an answer first