Question 16
Domain 2: Fairness, Bias, and Societal ImpactA team is building a machine learning system and wants to reduce bias throughout the entire AI lifecycle rather than only at deployment. Which approach best aligns with recommended bias-mitigation controls across the lifecycle?
Correct answer: B
Explanation
Bias mitigation should be addressed with controls at each major stage of the AI lifecycle, including data collection, preprocessing, model design, testing, and monitoring. Focusing on only one stage leaves other sources of bias unmanaged. — Source material: Propose controls during data collection, preprocessing, model design, testing, and monitoring.
Why each option is right or wrong
A. Apply bias controls only during model testing, because earlier stages are primarily technical preparation.
Controls are proposed during data collection, preprocessing, model design, testing, and monitoring.
B. Implement bias controls during data collection, preprocessing, model design, testing, and ongoing monitoring.
The source material explicitly identifies five lifecycle stages for proposed controls: data collection, preprocessing, model design, testing, and monitoring. Because the team wants mitigation across the entire lifecycle, the option covering all five stages is the best match.
C. Delay bias mitigation until monitoring, since real-world performance is the first reliable point for evaluation.
Controls are proposed before deployment as well as during monitoring.
D. Concentrate bias controls on data collection and preprocessing, because model design and testing do not affect bias outcomes.
Model design and testing are also listed stages for proposed controls.