Explaining how your AI system is fair

Fairness in AI isn't one-size-fits-all: The paper emphasizes that "fairness" has multiple, sometimes conflicting definitions, meaning organizations must choose the fairness approach that best fits their specific use case rather than applying a universal standard.

Context and ethics should drive fairness choices: The appropriate fairness definition depends on "ethical standards and legal requirements" specific to each application, suggesting that fairness decisions require input from ethicists, legal experts, and stakeholders—not just technologists.

Decision trees can make fairness selection transparent: The authors propose using decision trees to help AI practitioners map ethical principles to concrete fairness definitions, making the selection process "straightforward and transparent" rather than opaque.

Documenting fairness reasoning builds public trust: By "sharing the reasons and principles expressed during the decision making process with the broader audience," organizations can help end users understand and trust how fairness was implemented in AI systems affecting them.

Specifying fairness upfront is essential for accountability: Rather than treating fairness as an afterthought, the paper argues that explicitly defining what "fairness" means for a given system from the start is "the best way forward to maintain confidence in AI systems."