Moving beyond Principles: Identifying Actionable AI Fairness Practices

Fairness principles alone are insufficient: The paper identifies a critical "principles-to-practice gap," showing that existing AI fairness frameworks rely too heavily on abstract ethical principles without providing concrete, actionable steps that organizations can actually implement in their work.

Fairness requires role-specific guidance across the entire AI lifecycle: Rather than one-size-fits-all approaches, the research develops a matrix of fairness practices organized by organizational role and obligation level, recognizing that different teams (developers, managers, auditors, etc.) need tailored guidance at different stages—from initial design through deployment and monitoring.

Implementation requires a "socio-technical" perspective: The study emphasizes that AI fairness isn't just a technical problem; it involves organizational structures, workflows, and human decision-making. Effective fairness governance must address both the technology and the people and processes surrounding it.

Actionable practices need to be modular and dynamic: The research provides a structured, reusable "governance scaffold" that organizations can adapt to their specific contexts, rather than prescribing rigid rules—allowing for flexibility while maintaining accountability.

Fairness governance must be sustained, not one-time: By spanning the entire AI lifecycle, the framework emphasizes that fairness isn't achieved through a single audit or design phase, but requires ongoing practices, monitoring, and adjustment throughout an AI system's operational life.