An evaluation process to detect and mitigate biases in AI systems, ensuring fairness and compliance with ethical standards.
A structured review - often by an independent team or external firm - that examines every stage of the AI lifecycle (data collection, preprocessing, modeling, evaluation) for bias. Auditors apply statistical tests (e.g., disparate impact), model explainability tools, and user-group analyses. The process ends with concrete remediation recommendations and governance updates.
A bank commissions a bias audit of its credit-scoring AI. Auditors sample loan decisions by demographic group, find that applicants from certain ZIP codes are disproportionately denied, and recommend data-augmentation and new fairness constraints in the scoring algorithm, after which the bank monitors denial rates monthly to track progress.
