When Feasibility of Fairness Audits Relies on Willingness to Share Data: Examining User Acceptance of Multi-Party Computation Protocols for Fairness Monitoring

Privacy mechanisms matter more than fairness benefits in direct evaluation: When asked directly, European users prioritized technical privacy protections in multi-party computation protocols over the fairness monitoring benefits, suggesting that reassurance about data security is essential for gaining participation in bias detection efforts.

User priorities shift in realistic decision-making scenarios: Interestingly, when participants made simulated choices that mimicked real-world decisions, they weighted fairness objectives more heavily than privacy mechanisms, indicating a gap between what people say they care about and what actually influences their choices.

Individual fairness and privacy orientations shape acceptance: The study found that users' pre-existing beliefs about fairness and privacy significantly shaped whether they would accept sharing sensitive demographic data (like ethnicity) for algorithmic audits, suggesting one-size-fits-all communication strategies may be ineffective.

Regulatory compliance depends on public cooperation: Since the EU AI Act requires privacy-preserving methods like multi-party computation for fairness monitoring, the paper highlights a critical dependency: these legally mandated audits can only work if people are willing to share their data, making user acceptance a practical bottleneck for regulation.

Better communication is needed for informed consent: The findings suggest that how organizations communicate about privacy-preserving fairness protocols significantly influences whether people consent to participate, indicating that clearer, more aligned messaging could improve both compliance and public trust in algorithmic accountability.