Fair by design: A sociotechnical approach to justifying the fairness of AI-enabled systems across the lifecycle

Statistical fairness metrics alone are insufficient: The paper argues that most current approaches to AI fairness rely too heavily on statistical measures, which fail to capture the full complexity of fairness in real-world applications. A broader, context-dependent analysis is needed beyond these technical metrics.

Fairness must be considered across the entire AI lifecycle, not just at one stage: Rather than addressing fairness only during model development, the framework emphasizes that fairness considerations should be identified and addressed from initial design through deployment and ongoing use—a more comprehensive approach than typical current practices.

A practical framework (TEA) can help organizations document and justify fairness decisions: The paper presents the Trustworthy and Ethical Assurance framework as a concrete tool that enables transparent, participatory reasoning about fairness. This allows organizations to systematically identify, document, and justify their fairness considerations in a shared way.

Real-world healthcare applications reveal multiple fairness challenges: Using a clinical diagnostic system for hypertension risk prediction as a case study, the paper demonstrates that fairness concerns are concrete and varied—including potential discrimination against specific patient subgroups—highlighting why context-specific analysis matters.

Participatory approaches improve fairness assessment: The framework emphasizes involving multiple stakeholders in fairness reasoning rather than leaving it solely to technical teams, suggesting that collaborative decision-making produces more robust and defensible fairness outcomes.