Question 28
Domain 2: Fairness, Bias, and Societal ImpactA product team is deploying an AI-driven recommendation feed that can shape what users believe and how they act. Which safeguard is the best fit for reducing harmful influence from this content system?
Correct answer: B
Explanation
When a content system can influence beliefs or behavior, safeguards should be built in to limit harmful effects rather than relying only on performance or engagement measures. The key is to treat influence itself as a risk area that requires protective controls. — Source material: "Recommend safeguards for content systems that influence beliefs or behavior."
Why each option is right or wrong
A. Increase personalization so users receive more content aligned with prior preferences
More alignment with prior preferences does not itself function as a protective control against harmful influence.
B. Implement safeguards specifically designed for systems that affect beliefs or behavior
The source material directly recommends safeguards for content systems that influence beliefs or behavior. Because the recommendation feed in the scenario can shape user beliefs and actions, the appropriate response is to implement safeguards targeted to that influence risk.
C. Focus primarily on improving prediction accuracy for recommended content
Prediction accuracy addresses system performance, not whether influence on beliefs or behavior is adequately constrained.
D. Expand content exposure broadly so no single recommendation has much effect
Broader exposure is not identified as the required protective measure for systems that influence beliefs or behavior.