An internal AI audit team is planning an audit of a natural language processing system used for contract review. Which approach BEST ensures audit coverage of model bias risk?
Select an answer to reveal the explanation.
Short Explanation and Infographic
Here's the deal — b is correct because bias audit procedures require empirical testing of model outputs on representative subgroups — stratified sampling across demographic groups and document types reveals whether the model performs consistently across segments. A is insufficient as code review alone does not surface emergent bias.
Full explanation below image
Full Explanation
B is correct because bias audit procedures require empirical testing of model outputs on representative subgroups — stratified sampling across demographic groups and document types reveals whether the model performs consistently across segments. A is insufficient as code review alone does not surface emergent bias. C is a governance indicator but does not test actual model behavior. D addresses security, not bias.