Building Privacy, Trust, and Safety in AI Systems CRAGE Practice Questions (Page 5)

Questions 21–25

1. 21application · medium

   A government agency is deploying an AI system to assist in determining eligibility for social benefits. The system uses a complex model that considers hundreds of variables. Citizens have the right to understand why their application was denied. The agency is committed to transparency and accountability. What is the most effective approach to provide explainability for the decisions?

   • AOffer a general statement that the decision was made by an AI system and is based on a variety of factors.
   • BUse model-agnostic explanation techniques, such as LIME or SHAP, to generate post-hoc explanations for each individual decision.
   • CPublish the entire model's code and training data on a public website for anyone to inspect.
   • DProvide a simple rule-based approximation of the model that is easy to understand, even if it is not perfectly accurate.
   Reveal Solution Discussion Select an answer first
2. 22application · medium

   An autonomous vehicle manufacturer is testing an AI system that controls braking in emergency situations. During simulation testing, the system occasionally fails to detect pedestrians in low-light conditions. The safety engineering team must implement a mechanism to prevent harmful outcomes in the real world. What is the most appropriate safety mechanism to implement?

   • AImplement a fail-safe that triggers an automatic emergency stop when the system's confidence in its detection falls below a predefined threshold.
   • BDisable the AI system entirely and rely on the driver to handle all braking decisions.
   • CAdd more sensors to the vehicle to collect additional data, but do not change the decision-making logic.
   • DIncrease the system's processing speed to make faster decisions, even if this reduces the accuracy of pedestrian detection.
   Reveal Solution Discussion Select an answer first
4. 23application · medium

   An autonomous vehicle company is testing its AI navigation system. The system must be safe in unexpected situations, such as sensor failures or unusual road conditions. The safety team is evaluating robustness testing and fail-safe mechanisms. Which approach best ensures the AI system can handle unexpected situations safely?

   • AIncreasing the vehicle's speed to get through unexpected situations quickly.
   • BDisabling the AI system and relying entirely on a remote human operator.
   • CConducting extensive robustness testing that simulates sensor failures and unusual road conditions, and implementing a fail-safe that safely stops the vehicle if the AI cannot make a safe decision.
   • DRelying on the AI system's ability to learn from real-world driving without prior testing.
   Reveal Solution Discussion Select an answer first
5. 24application · medium

   A smart home device manufacturer is designing a new AI-powered assistant that collects audio data from users' homes. The company wants to apply Privacy by Design principles from the outset. The design team is considering how to handle user consent and data minimization. Which approach best embodies Privacy by Design?

   • ADesigning the device to collect only the audio data necessary for the assistant's core functions, and providing a clear, easy-to-use consent interface at setup.
   • BStoring audio data indefinitely in the cloud to allow for future feature development.
   • CCollecting all audio data continuously to improve the assistant's performance, and asking for consent only once at purchase.
   • DProviding a detailed privacy policy in the user manual, but not offering any in-device consent controls.
   Reveal Solution Discussion Select an answer first
6. 25application · medium

   A health insurance company is using an AI system to process claims. A customer requests that their personal data be deleted from the system, citing their right to erasure under GDPR. The company has a legal obligation to retain certain claim data for regulatory compliance. What is the most appropriate action?

   • AAssess the request and delete the data that is not required for legal compliance, while retaining the data that must be kept for regulatory purposes.
   • BRefuse the request entirely because the company has a legal obligation to retain the data.
   • CImmediately delete all data related to the customer, including data required for regulatory compliance.
   • DAnonymize the customer's data so that it can no longer be linked to them, and then delete the original data.
   Reveal Solution Discussion Select an answer first