Objective 3.4 Identify and Mitigate Fairness and Non-Discrimination (bias) Risks CERTIFIED-ETHICAL-EMERGING-TECHNOLOGIST Practice Questions (Page 1)

Questions 1–5

1. 1application · medium

   A government agency is deploying an AI system to allocate public housing. They want to ensure that the system does not discriminate against any racial group. They have decided to use a fairness metric to measure the system's performance. Which metric would be most appropriate to ensure that the proportion of positive outcomes (housing allocation) is similar across racial groups?

   • AEqualized odds
   • BDemographic parity
   • CCalibration
   • DIndividual fairness
   Reveal Solution Discussion Select an answer first
2. 2foundation · easy

   A data scientist is building a model to predict loan approval. The historical data used for training contains a disproportionate number of approved loans for one demographic group. Which type of bias is most directly introduced by this data characteristic?

   • AConfirmation bias
   • BAlgorithmic bias
   • CData bias
   • DHuman bias
   Reveal Solution Discussion Select an answer first
4. 3application · medium

   A tech company is building a resume-screening AI to shortlist candidates. The training data is from the past five years and reflects the company's historical hiring patterns, which favored candidates from certain universities. The company wants to ensure the AI does not perpetuate this bias. Which mitigation strategy should they implement?

   • AUse an algorithmic adjustment that penalizes the model for using university names as a feature.
   • BRemove the university name feature and rely on other features like skills and experience.
   • CUse a data preprocessing technique to balance the dataset by oversampling candidates from underrepresented universities.
   • DApply a post-processing technique that randomly selects candidates from the shortlist to ensure diversity.
   Reveal Solution Discussion Select an answer first
5. 4application · medium

   A healthcare organization uses an AI system to prioritize patients for organ transplants. The system was trained on historical data that underrepresents certain ethnic groups. The organization wants to ensure the system does not unfairly disadvantage these groups. Which approach should they take to identify and mitigate this bias?

   • AUse a fairness metric such as demographic parity to compare the selection rates across ethnic groups and then apply a reweighting technique to the training data.
   • BUse a single fairness metric, such as equalized odds, and if the metric is satisfied, declare the model fair.
   • CRemove all demographic attributes from the dataset and retrain the model to ensure fairness.
   • DApply a post-processing technique that adjusts the model's outputs to equalize false-positive rates across groups, without first measuring the baseline disparity.
   Reveal Solution Discussion Select an answer first
6. 5application · medium

   A data science team is using Fairlearn to assess a model's fairness. They want to compare the model's performance across different groups using multiple metrics. Which Fairlearn tool should they use?

   • AFairlearn's Exponentiated Gradient
   • BFairlearn's DemographicParity
   • CFairlearn's MetricFrame
   • DFairlearn's GridSearch
   Reveal Solution Discussion Select an answer first