Towards Involving End-users in Interactive Human-in-the-loop AI Fairness

End-users can meaningfully participate in fairness decisions: The research demonstrates that ordinary people without technical backgrounds can identify and help address fairness issues in AI systems when given appropriate interactive tools, challenging the assumption that fairness oversight must be limited to ML experts.

Interpretability is essential for public engagement: The prototype system allowed end-users to see why specific loan decisions were made, which was critical for enabling them to spot potential bias and suggest corrections—highlighting that transparency about AI reasoning is foundational to broader fairness accountability.

Cultural values shape fairness judgments: The study examined how cultural dimensions influence people's fairness perspectives globally, suggesting that one-size-fits-all fairness standards may not be appropriate and that diverse stakeholder input across different cultural contexts should inform AI fairness design.

Interactive "debugging" makes fairness actionable: By allowing end-users to adjust feature weights in the system, the research shows that fairness isn't just an abstract concept to evaluate—it can be made concrete and actionable through interfaces where people can directly experiment with and modify how AI systems make decisions.

Co-design with end-users improves system usability: The workshops where end-users helped design the prototype resulted in a system that was actually useful and understandable to non-experts, suggesting that involving affected communities early in development produces more effective fairness tools than top-down expert approaches alone.