Ethical, Societal, and Security Concerns in AI AIE Practice Questions (Page 4)

Questions 16–20

1. 16application · medium

   A hospital uses an AI system to recommend treatment plans for patients with chronic conditions. The system was trained on historical data from a single region and performs well for that population, but the hospital now serves a more diverse patient base. A physician notices that the system recommends different treatments for patients with similar symptoms but different ethnic backgrounds. The hospital's ethics board wants to ensure the system is fair and accountable. What should the hospital do?

   • ARetrain the model on a more diverse dataset and evaluate performance across demographic subgroups
   • BRemove ethnicity as a feature from the model
   • CPublish the model's training data to the public for independent review
   • DRequire physicians to manually override the system for minority patients
   Reveal Solution Discussion Select an answer first
2. 17application · medium

   A social media platform uses an AI content moderation system. Attackers have learned to add tiny, imperceptible noise to images to make the system misclassify harmful content as safe. The platform wants to defend against this type of attack. Which approach is most effective?

   • AIncrease the model's confidence threshold for flagging content.
   • BRely on human moderators to review all content flagged by the AI.
   • CUse adversarial training by augmenting the training set with perturbed images.
   • DUse a simpler model that is less sensitive to small changes.
   Reveal Solution Discussion Select an answer first
4. 18expert · hard

   A bank uses an AI system for loan approval. The system is a trade secret, and the bank is concerned about model theft by competitors. The bank also faces regulatory pressure to explain loan decisions to rejected applicants. The bank must balance protecting its intellectual property with the legal requirement for transparency. Which approach best satisfies both concerns?

   • ARefuse to provide any explanation, citing trade secret protection
   • BProvide a generic explanation that the decision was based on a 'proprietary algorithm'
   • CProvide applicants with a specific explanation of the factors that influenced the decision, without revealing the model's weights or full architecture
   • DPublish the model's full architecture and weights to demonstrate transparency
   Reveal Solution Discussion Select an answer first
5. 19expert · hard

   A multinational corporation deploys an AI recruitment tool across different countries. The tool is trained on data from one region and performs well there, but when used in other regions, it shows significant bias against local candidates. The company must ensure fairness across all regions while maintaining a consistent hiring standard. What is the best approach?

   • AUse the same model globally to ensure consistent standards.
   • BRemove demographic attributes from the model to make it 'blind' to region.
   • CCollect more data from all regions and train one global model.
   • DTrain separate models for each region using local data, and validate each model for fairness locally.
   Reveal Solution Discussion Select an answer first
6. 20expert · hard

   A security company uses an AI system to detect network intrusions. The system is trained on a public dataset of network traffic. An attacker discovers that the system has a blind spot for certain types of attacks and exploits it to breach a client's network. The company wants to improve the system's robustness without sacrificing its ability to detect known attacks. The company has a limited budget and must choose a mitigation. Which approach is most effective?

   • AUse an ensemble of multiple AI models to improve overall accuracy
   • BAugment the training data with adversarial examples and retrain the model
   • CIncrease the model's sensitivity to catch more potential intrusions
   • DReplace the AI system with a signature-based intrusion detection system
   Reveal Solution Discussion Select an answer first