AI Design, Development/Procurement, and Documentation AAIR Practice Questions (Page 5)

Questions 21–25

1. 21application · medium

   A bank is designing a credit-scoring AI system. The design team proposes using a model trained on historical loan data that includes a protected attribute (e.g., race) as a feature. The compliance officer is concerned about fairness and regulatory risk. Which action should the design team take FIRST to address this concern?

   • ADefer the decision to the development team, as they are best positioned to handle data-related issues.
   • BRemove the protected attribute from the training data and proceed with model development.
   • CConduct a fairness impact assessment to identify potential bias and define fairness metrics before model training.
   • DProceed with training and then test for bias after deployment to see if any issues arise.
   Reveal Solution Discussion Select an answer first
2. 22foundation · easy

   What is the primary purpose of communicating AI design decisions to stakeholders?

   • ATo ensure transparency and alignment with organizational goals
   • BTo reduce the need for testing
   • CTo hide risks from stakeholders
   • DTo increase the complexity of the system
   Reveal Solution Discussion Select an answer first
4. 23application · medium

   An AI development team is building a predictive maintenance system for industrial equipment. During development, they make several changes to the model architecture and retrain the model with new data. To support future audits and regulatory compliance, which documentation should the team maintain throughout this process?

   • AA versioned development log that records changes, data versions, model parameters, and rationale for each decision.
   • BA final report summarizing the model's performance on the test set.
   • CA copy of the final model's source code only.
   • DAn email thread among team members discussing potential improvements.
   Reveal Solution Discussion Select an answer first
5. 24application · medium

   During the development of an AI system for loan approval, the testing team finds that the model's performance degrades when applied to new customer data that differs from the training distribution. The team needs to ensure the model remains reliable in production. What should they do?

   • AFreeze the model and deploy it, as the testing results are sufficient.
   • BReduce the model's complexity to make it more robust to distribution shifts.
   • CImplement continuous monitoring and a feedback loop to retrain the model as new data becomes available.
   • DDocument the issue and defer action until after deployment.
   Reveal Solution Discussion Select an answer first
6. 25foundation · easy

   Which procurement risk is most concerned with the ability to maintain and update the AI system over time?

   • ALong-term maintainability
   • BUser interface aesthetics
   • CMarketing strategy
   • DInitial cost
   Reveal Solution Discussion Select an answer first