Audit Planning and Design AAIA Practice Questions (Page 6)

Questions 26–30

1. 26application · medium

   An audit team is planning to audit a natural language processing (NLP) model used for customer sentiment analysis. The model was trained on social media posts and customer reviews. The team needs to test whether the model's outputs are consistent and unbiased. Which audit procedure would be most effective?

   • ACompare the model's outputs to a random sample of manually labeled data to measure accuracy.
   • BPerform adversarial testing by providing the model with carefully crafted inputs designed to expose biases or inconsistencies.
   • CReview the model's training data for any obvious biases in the source data.
   • DInterview the data scientists who developed the model to understand their design choices.
   Reveal Solution Discussion Select an answer first
2. 27foundation · easy

   Why is it important for an AI auditor to maintain independence?

   • ATo ensure objectivity and avoid conflicts of interest
   • BTo ensure the audit is completed faster
   • CTo reduce the cost of the audit
   • DTo increase the number of audit findings
   Reveal Solution Discussion Select an answer first
4. 28foundation · easy

   What is a characteristic of a well-defined audit objective?

   • AIt is specific and measurable
   • BIt is based on the auditor's personal opinion
   • CIt is broad and general to cover all possibilities
   • DIt is unchangeable once defined
   Reveal Solution Discussion Select an answer first
5. 29expert · hard

   An audit team is being assembled to audit an AI system that uses natural language processing to analyze legal documents. The team currently includes an IT auditor, a data analyst, and a legal expert. The audit will involve reviewing the model's training data, testing its accuracy, and evaluating compliance with legal standards. What additional expertise is most critical for the team?

   • AA communications specialist to handle stakeholder interactions.
   • BA data privacy officer to ensure compliance with data protection laws.
   • CA machine learning engineer with experience in NLP models.
   • DA project manager to coordinate the audit activities.
   Reveal Solution Discussion Select an answer first
6. 30application · medium

   An internal audit team is tasked with auditing a new AI-based recommendation system used by an e-commerce company. The system was developed in-house and has been in production for three months. The audit sponsor wants the audit to cover the entire system, including data pipelines, model training, and deployment. However, the audit team has only two weeks to complete the audit. What should the audit lead do first?

   • AImmediately begin testing the model's accuracy and performance to ensure the system is working correctly.
   • BFocus only on the model's algorithm and ignore the data pipelines and deployment, as these are less critical.
   • CDefine specific, measurable audit objectives and scope boundaries that align with the sponsor's request but are achievable within the two-week timeline.
   • DRequest an extension of the audit timeline to allow for a comprehensive review of all components.
   Reveal Solution Discussion Select an answer first