3.1 Explain Responsible AI Principles (fairness, Transparency, Accountability, Bias Mitigation, Safety) AI-TECHNICAL-PRACTITIONER Practice Questions (Page 4)

Questions 16–20

1. 16expert · hard

   A financial regulator requires that all AI models used for loan approvals be auditable. A bank uses a complex ensemble model that is difficult to audit. The bank must comply but wants to minimize disruption to its operations. What is the best approach?

   • ARequest an exemption from the regulator based on the model's high accuracy.
   • BReplace the ensemble model with a single, interpretable model.
   • CDevelop an internal audit process that uses model explanation tools to document decisions.
   • DOutsource the audit to a third-party company that specializes in AI auditing.
   Reveal Solution Discussion Select an answer first
2. 17foundation · easy

   How can fairness in AI be assessed?

   • ABy measuring the accuracy of the model on a single test set
   • BBy evaluating the model's performance and outcomes across different demographic groups
   • CBy checking that the model's predictions are consistent over time
   • DBy ensuring the model is trained on the largest possible dataset
   Reveal Solution Discussion Select an answer first
4. 18application · medium

   A recruitment firm uses an AI system to screen resumes. The system was trained on resumes from the past five years, which predominantly came from one gender. The firm wants to reduce gender bias in the system. Which strategy is most effective?

   • ARemove gender-related keywords from the resumes before training.
   • BBlind the system to the applicant's name and gender during screening.
   • CUse a more sophisticated natural language processing model to better understand resumes.
   • DRe-balance the training data by including more resumes from the underrepresented gender.
   Reveal Solution Discussion Select an answer first
5. 19foundation · easy

   What is a common source of bias in AI systems?

   • AUsing too many features in the model
   • BBiased training data that does not represent the target population
   • CUsing a simple algorithm instead of a complex one
   • DDeploying the model on a cloud platform
   Reveal Solution Discussion Select an answer first
6. 20application · medium

   A university uses an AI system to screen student applications. The admissions team is concerned that the system may inadvertently disadvantage applicants from certain regions. What is the most appropriate way to assess fairness?

   • AReview the model's overall accuracy on a test set.
   • BEnsure the model does not use region as an input feature.
   • CCompare acceptance rates across different demographic groups and investigate significant disparities.
   • DAsk the AI vendor to guarantee that the system is fair.
   Reveal Solution Discussion Select an answer first