Heterogeneous preferences and asymmetric insights for AI use among welfare claimants and non-claimants

The deployment of AI in welfare benefit allocation accelerates decision-making but has led to unfair denials and false fraud accusations. In the US and UK (N = 3249), we examine public acceptability of speed-accuracy trade-offs among claimants and non-claimants. While the public generally tolerates modest accuracy losses for faster decisions, claimants are less willing to accept AI in welfare systems, raising concerns that using aggregate data for calibration could misalign policies with the preferences of those most affected. Our study further uncovers asymmetric insights between claimants and non-claimants. Non-claimants overestimate claimants' willingness to accept speed-accuracy trade-offs, even when financially incentivized for accurate perspective-taking. This suggests that policy decisions aimed at supporting vulnerable groups may need to incorporate minority voices beyond popular opinion, as non-claimants may not easily understand claimants' perspectives.