Supervision of AI Solutions (e.g., Outputs, Impacts, and Decisions) AAIA Practice Questions (Page 5)

Questions 21–25

1. 21application · medium

   A healthcare AI system that predicts patient readmission risk has been producing inaccurate predictions for a subset of patients. The AI operations team has identified the root cause as a data quality issue in the input data. What is the most appropriate next step in handling this error?

   • AImmediately retrain the model with the corrected data and redeploy it.
   • BAdjust the model's prediction threshold to reduce the number of inaccurate predictions.
   • CLog the error, correct the data quality issue, and document the incident for future reference.
   • DNotify the patients affected by the inaccurate predictions.
   Reveal Solution Discussion Select an answer first
2. 22application · medium

   A government agency is considering deploying an AI system to automate the review of visa applications. The agency wants to assess the potential impact of the system on applicants and society before deployment. Which method is most appropriate for this impact assessment?

   • AReview the system's technical documentation to ensure it meets performance specifications.
   • BConduct a stakeholder consultation to gather input from affected communities and experts.
   • CAsk the AI vendor to provide a list of potential risks.
   • DCompare the system's processing speed to that of human reviewers.
   Reveal Solution Discussion Select an answer first
4. 23application · medium

   An organization's AI operations team has been monitoring an AI system that predicts customer churn. The team has identified several instances where the model's predictions were inaccurate. The team wants to document these incidents for future reference and reporting. What is the most important information to include in the documentation?

   • AThe personal opinions of the team members about why the errors occurred.
   • BA recommendation to stop using the AI system for churn prediction.
   • CThe names of the customers whose churn was incorrectly predicted.
   • DThe date and time of each incident, the nature of the error, and the corrective action taken.
   Reveal Solution Discussion Select an answer first
5. 24foundation · easy

   Which method is commonly used to assess the potential impact of an AI solution on stakeholders?

   • AReviewing the AI model's source code line by line
   • BMeasuring the server's CPU usage during inference
   • CChecking the AI model's training data size
   • DConducting stakeholder interviews and surveys
   Reveal Solution Discussion Select an answer first
6. 25expert · hard

   A multinational corporation has deployed several AI systems across different business units. Each unit has its own AI operations team, but there is no central oversight. The board of directors wants to ensure consistent supervision of AI decisions and compliance with ethical standards. The company is considering establishing a governance structure. Which structure best balances central oversight with business-unit autonomy?

   • ACreate a central AI governance committee that sets policies and standards, while business units retain responsibility for day-to-day supervision and incident handling.
   • BOutsource AI supervision to an external auditor that reviews each system annually.
   • CEstablish a single centralized team that supervises all AI systems across the company, removing business-unit involvement.
   • DDelegate all AI supervision to each business unit's existing operations team without additional oversight.
   Reveal Solution Discussion Select an answer first