Managing Artificial Intelligence GSLC Practice Questions (Page 8)

Questions 36–40

1. 36application · medium

   A security team notices that a fraud-detection model, which previously flagged suspicious transactions accurately, has started missing obvious fraud cases. The model was retrained last week on a new dataset that included a large batch of manually reviewed transactions. The team suspects the retraining introduced a vulnerability. Which action best addresses the most likely cause while also preparing for future incidents?

   • ADisable the model and rely on manual review of all transactions until the next scheduled retraining.
   • BRoll back to the previous model version and initiate an incident response to investigate the data used in retraining.
   • CIncrease the model's confidence threshold so it only flags transactions with very high probability of fraud.
   • DRetrain the model again using the same dataset but with more training epochs.
   Reveal Solution Discussion Select an answer first
2. 37expert · hard

   A company's executive team wants to deploy AI to automate customer support. The security team has identified significant risks, including potential data leakage and reputational harm from incorrect responses. The company has a low risk appetite and limited budget. Which approach best aligns the AI initiative with the organization's risk appetite?

   • AOutsource the entire customer support function to an AI vendor, transferring all risk to the vendor.
   • BImplement a pilot program with a small subset of customers, with human oversight and clear escalation procedures.
   • CReject the AI initiative entirely because the risks are too high.
   • DDeploy the AI system to all customers immediately to maximize cost savings and efficiency.
   Reveal Solution Discussion Select an answer first
4. 38foundation · easy

   Which technique is commonly used to mitigate bias in AI models?

   • ARe-sampling the training data to balance representation
   • BIncreasing the number of model layers
   • CUsing a larger learning rate
   • DDisabling model interpretability tools
   Reveal Solution Discussion Select an answer first
5. 39application · medium

   A financial services firm is required to audit its AI-based credit scoring system to ensure it complies with fair lending laws. The audit must verify that the model does not discriminate based on protected characteristics. Which audit approach is most effective?

   • ACompare the model's predictions to those of a human loan officer to see if they align.
   • BInterview the data scientists who built the model to understand their intentions and assumptions.
   • CPerform a statistical analysis of the model's outcomes across demographic groups, using fairness metrics such as disparate impact ratios.
   • DReview the model's source code to ensure that protected attributes are not used as input features.
   Reveal Solution Discussion Select an answer first
6. 40expert · hard

   A healthcare provider is evaluating an AI vendor for diagnostic support. The vendor's model was trained on data from a different country and has not been validated locally. The provider must ensure patient safety and regulatory compliance. What is the most critical requirement to include in the contract?

   • AA requirement to use only cloud-based deployment
   • BA clause that the vendor is liable for all patient outcomes
   • CA requirement that the vendor share the model's source code
   • DA requirement for local validation studies and ongoing performance monitoring
   Reveal Solution Discussion Select an answer first