Privacy, Ethical, Trust and Safety Controls AAISM Practice Questions (Page 6)

Questions 26–30

1. 26foundation · easy

   Which mitigation technique is used to reduce bias in an AI model by adjusting the training data to better represent underrepresented groups?

   • AModel pruning
   • BGradient clipping
   • CData re-sampling
   • DHyperparameter tuning
   Reveal Solution Discussion Select an answer first
2. 27expert · hard

   A research institute is training an AI model on a dataset that includes sensitive personal data. The institute wants to share the trained model with external collaborators without exposing the underlying personal data. The collaborators need to use the model for inference but should not be able to extract personal information. Which data governance practice should the institute implement?

   • AProvide collaborators with a sample of the training data instead of the model.
   • BShare the model with collaborators under a non-disclosure agreement.
   • CRemove all personal data from the training set before training.
   • DUse differential privacy during training to limit the model's ability to memorize individual records.
   Reveal Solution Discussion Select an answer first
4. 28application · medium

   A hospital uses an AI system to recommend treatment plans for cancer patients. The system occasionally makes recommendations that conflict with a physician's judgment. The hospital wants to ensure accountability for patient outcomes. What should the hospital implement?

   • ALimit the AI's recommendations to only those that match standard protocols.
   • BRequire that a physician reviews and approves every AI recommendation before it is implemented.
   • CAllow the AI to implement recommendations automatically to save time.
   • DDocument all AI recommendations and outcomes, but do not require human approval.
   Reveal Solution Discussion Select an answer first
5. 29foundation · easy

   Which ethical principle emphasizes that AI systems should be developed and used in a manner that is fair, equitable, and does not discriminate against individuals or groups?

   • AAccountability
   • BTransparency
   • CPrivacy
   • DJustice
   Reveal Solution Discussion Select an answer first
6. 30expert · hard

   A public sector organization is deploying an AI system to assist in parole decisions. The system provides a risk score, but the organization is concerned about public trust. The system's developers have provided a technical explanation of the model, but the public and the parole board find it difficult to understand. The organization wants to build trust among all stakeholders. What should the organization do?

   • ADevelop a multi-layered explanation strategy: a plain-language summary for the public, a detailed technical report for experts, and a clear appeals process for affected individuals.
   • BLimit the use of the AI system to advisory recommendations only, with no explanation required.
   • CRely on the technical explanation provided by the developers as sufficient for all stakeholders.
   • DPublish the full source code of the AI system to demonstrate transparency.
   Reveal Solution Discussion Select an answer first