Key points are not available for this paper at this time.
As AI is increasingly used in public services, concerns have grown about bias and misalignment with community values. To investigate how ordinary citizens reason about fairness in public-sector AI, we conducted a scenario-based study with 110 participants who evaluated 13 algorithmically-informed government decision systems. Drawing on their responses, we analyze how auditors evaluate risks, identify impacted groups, and navigate fairness trade-offs. Our findings reveal that participants do not audit arbitrarily; they adjust their evaluations to context, reason from values, and exhibit principled subjectivity by applying fairness orientations consistently while remaining sensitive to scenario-specific details. However, we also observe tensions between bias recognition and prioritization, suggesting a gap between awareness and actionable design focus. Building on literature in participatory AI, algorithmic impact assessments, and user-driven auditing, we argue that good auditors are not merely detectors of error but normative agents who surface latent value tensions. We conclude by offering design implications for audit tooling that supports individual and group deliberation, trade-off reasoning, and equity-centered decision-making in the oversight of public algorithms.
Jackson et al. (Thu,) studied this question.