Safety and AI CT-AI Practice Questions (Page 4)

Questions 16–20

1. 16expert · hard

   An AI system is used to recommend personalized treatment plans for patients with chronic diseases. The system is highly accurate but is also vulnerable to adversarial attacks that could change its recommendations. The hospital wants to ensure the system is safe. Which approach best addresses the trade-off between accuracy and security?

   • AAccept the risk because the system is highly accurate and adversarial attacks are unlikely in a hospital setting.
   • BImplement adversarial training to make the system more robust, even if it slightly reduces accuracy on normal inputs.
   • CAdd a human review step for all recommendations, which will catch any adversarial changes.
   • DDisconnect the system from the network to prevent attacks.
   Reveal Solution Discussion Select an answer first
2. 17application · medium

   A company is developing an AI system for autonomous drones used in package delivery. The system must be safe, but it also needs to be efficient. The team is considering how to balance safety with other quality characteristics. Which approach best balances these concerns?

   • ADesign the system to be safe and efficient by using a single model that is trained to optimize both objectives.
   • BDesign the system with safety as a constraint, and optimize efficiency within that constraint.
   • CDesign the system to prioritize safety above all else, even if it reduces efficiency.
   • DDesign the system to maximize efficiency, and add safety features only if they do not reduce efficiency.
   Reveal Solution Discussion Select an answer first
4. 18foundation · easy

   Which technique for assuring safety in AI involves mathematically proving that the system meets its safety properties?

   • ARedundancy
   • BAdversarial testing
   • CRuntime monitoring
   • DFormal verification
   Reveal Solution Discussion Select an answer first
5. 19application · medium

   A company is developing an AI system that recommends treatment plans for cancer patients. The system's recommendations are sometimes overridden by doctors. The company wants to define safety in the context of this AI system. Which definition is most appropriate?

   • AThe system must be secure against external attacks.
   • BThe system must not cause harm to patients, either through incorrect recommendations or through failures in its operation.
   • CThe system must always recommend the treatment that has the highest success rate.
   • DThe system must be reliable and available at all times.
   Reveal Solution Discussion Select an answer first
6. 20foundation · easy

   Which of the following is an example of a safety risk specific to AI-based systems?

   • AA denial-of-service attack that overwhelms the system's servers.
   • BA power outage that shuts down the entire data center.
   • CA database administrator accidentally deleting a table of user records.
   • DAn adversarial attack that adds tiny perturbations to an image, causing a self-driving car to misclassify a stop sign.
   Reveal Solution Discussion Select an answer first