Responsible Practices and Governance Standards AIE Practice Questions (Page 8)

Questions 36–40

1. 36expert · hard

   A hospital deploys an AI system that recommends sedation doses for patients in intensive care. The system was approved by the hospital's ethics board. Six months later, a patient receives an incorrect dose and is harmed. An investigation finds that the model's recommendations had been silently changing due to data drift. The hospital wants to prevent recurrence. Which action is most effective?

   • ADiscontinue the AI system and rely solely on clinician judgment.
   • BRetrain the model monthly using the most recent patient data.
   • CImplement continuous performance monitoring with automated alerts when model accuracy or safety metrics fall below thresholds.
   • DRequire that all model recommendations be reviewed by a senior clinician before administration.
   Reveal Solution Discussion Select an answer first
2. 37expert · hard

   A global software company is developing an AI recruitment tool that screens job applicants. The tool will be used in the EU and the US. The company wants to comply with the EU AI Act, GDPR, and US anti-discrimination laws. The tool uses a complex deep learning model that is not explainable. The legal team is concerned about the 'right to explanation' under GDPR and the AI Act's transparency requirements. What is the most appropriate approach?

   • AProvide only aggregate statistics about the model's performance to applicants.
   • BUse an explainable AI technique (e.g., LIME) to provide post-hoc explanations for each candidate's screening decision, and document the system's logic and purpose.
   • CReplace the deep learning model with a simple logistic regression model to ensure full interpretability.
   • DArgue that the AI Act does not apply because the company is based in the US.
   Reveal Solution Discussion Select an answer first
4. 38application · medium

   A financial institution has deployed an AI system that approves or rejects loan applications. Regulators require the institution to demonstrate accountability for the system's decisions. The institution wants to ensure that any decision can be traced back to the responsible team and that the system remains compliant over time. Which set of practices best meets these requirements?

   • AMaintain detailed audit logs of model versions, training data, and decision rationale, and schedule periodic reviews with human oversight.
   • BSet up automated alerts for any model performance drop and rely on the IT team to fix issues as they arise.
   • CUse a third-party model as a black box and accept the vendor's assurances of compliance.
   • DDelete all historical logs after each decision to protect customer privacy.
   Reveal Solution Discussion Select an answer first
5. 39expert · hard

   A non-profit organization is developing an AI system to match refugees with housing and employment opportunities. The organization has limited resources and must decide how to engage stakeholders. The system will be used by caseworkers and refugees. Which stakeholder engagement strategy is most effective given the resource constraints?

   • AConduct a small number of focus groups with caseworkers and refugees, and use the findings to inform design and governance decisions.
   • BDefer stakeholder engagement until after the system is deployed to save time.
   • CLaunch a large-scale public survey to gather input from a broad audience.
   • DRely on the organization's board of directors to represent stakeholder interests.
   Reveal Solution Discussion Select an answer first
6. 40foundation · easy

   Which of the following is a key legal requirement imposed by the GDPR that directly affects AI systems processing personal data?

   • AObtaining valid consent or another lawful basis for processing personal data.
   • BPublishing the source code of all AI algorithms.
   • CRegistering all AI models with a national authority before deployment.
   • DEnsuring that all AI decisions are made without any human involvement.
   Reveal Solution Discussion Select an answer first