Getting Fairness Right: Towards a Toolbox for Practitioners

There is no one-size-fits-all fairness solution: Different fairness metrics are mathematically incompatible with each other, meaning practitioners must choose which fairness definition best fits their specific use case rather than applying a universal standard.

Fairness requires more than just math: Implementing fair AI systems demands input from multiple stakeholders—including ethicists, legal experts, and community members—to define what fairness actually means in a given context, not just technical adjustments by data scientists.

Context matters for fairness decisions: The "right" fairness approach depends on the nature of the application, available data, legal requirements, and the ethical, philosophical, and cultural values of the communities affected by the AI system.

A practical toolbox can help non-experts navigate fairness: By structuring the complex landscape of fairness metrics and options, practitioners without deep technical backgrounds can better understand and select appropriate fairness objectives for their AI systems.

Fairness requires intentional commitment from stakeholders: Rather than assuming AI systems will be fair by default, organizations must actively commit to defining and implementing fairness standards that reflect their local context and values.