What steps constitute a bias audit in CPMAI?

• A. Rely on overall accuracy on the full dataset without subgroup analysis.
• B. Define fairness criteria but skip subgroup evaluation.
• C. Test data represent population but do not document findings.
• D. Define fairness criteria, test data represents population, evaluate outcomes across subgroups, and document findings.

Answer: (D)

Bias audits in CPMAI require a structured approach that combines clear fairness standards, representative data, subgroup checks for disparate impacts, and thorough documentation. Defining fairness criteria gives you a concrete benchmark to judge whether results are biased, rather than relying on a vague sense of fairness. Using data that truly represent the population ensures findings reflect real-world conditions rather than a skewed sample that could mask issues. Evaluating outcomes across subgroups is essential because overall accuracy can hide differential impacts; disparities may exist even when overall performance looks good. Documenting the findings provides transparency, accountability, and a record of assumptions, methods, and limitations that others can review and build upon. If any of these elements are missing, you risk missing biases, misinforming stakeholders, or being unable to reproduce or challenge conclusions.