Responsible Practices and Governance Standards AIE Practice Questions (Page 11)

Questions 51–55

1. 51application · medium

   A European healthcare startup is developing an AI system that triages patient symptoms and recommends next steps. The system will process personal health data of EU residents. The startup plans to deploy the system in a public cloud region located in the EU. The compliance team is concerned about GDPR obligations. Which action is most directly required to align with GDPR when processing this data?

   • AConduct a Data Protection Impact Assessment (DPIA) before processing begins.
   • BStore all data on-premises in the EU to satisfy data residency requirements.
   • CObtain explicit consent from every patient whose data is used, even if another lawful basis applies.
   • DEnsure the AI model is trained exclusively on synthetic data to avoid using real patient data.
   Reveal Solution Discussion Select an answer first
2. 52application · medium

   An insurance company is developing an AI model to set premiums. The model uses many variables, including some that could be proxies for protected characteristics like race or gender. The company wants to avoid discriminatory outcomes. What is the most effective way to assess and mitigate this ethical risk?

   • ATest the model only on the overall population to ensure average accuracy.
   • BRemove all variables that could be proxies for protected characteristics, even if they are strong predictors.
   • CConduct a fairness audit using multiple metrics (e.g., disparate impact, equalized odds) and adjust the model if bias is detected.
   • DUse only the variables that are legally allowed and ignore any potential bias.
   Reveal Solution Discussion Select an answer first
4. 53expert · hard

   A bank uses a deep learning model to detect fraudulent transactions. The model is highly accurate, but its decisions are difficult to explain. Regulators require the bank to explain why specific transactions were flagged. The bank's data science team is considering several approaches. Which approach best balances accuracy and explainability?

   • AProvide regulators with the model's training data and feature weights.
   • BUse a post-hoc explainability technique such as LIME to generate local explanations for each flagged transaction.
   • CHire external auditors to review the model and provide a general assurance report.
   • DReplace the deep learning model with a decision tree to ensure full interpretability.
   Reveal Solution Discussion Select an answer first
5. 54foundation · easy

   When assessing ethical risks in an AI project, which of the following is a key consideration?

   • AThe brand of the hardware used for training.
   • BThe programming language used to implement the AI model.
   • CThe number of developers on the project team.
   • DPotential societal impact of the AI system's deployment.
   Reveal Solution Discussion Select an answer first
6. 55application · medium

   A retail company deploys a recommendation system that suggests products to customers. After six months, the marketing team notices that the system's recommendations are becoming less diverse and increasingly favor a few best-selling items. The company wants to maintain ethical standards. What should the company do?

   • AIncrease the frequency of model retraining without measuring any specific metrics.
   • BImmediately switch to a random recommendation algorithm to ensure diversity.
   • CLeave the system unchanged because the decline is a natural result of user preferences.
   • DSet up periodic evaluations that measure recommendation diversity and user satisfaction, and retrain the model when metrics decline.
   Reveal Solution Discussion Select an answer first