Reported trust varies with graded value alignment in AI-attributed economic–environmental choices

Abstract

As artificial intelligence (AI) systems increasingly support value-laden decisions, users may judge whether an advisor’s observable choices align with their own priorities. Prior theories of value similarity predict that value alignment should support trust, but less is known about whether self-reported trust varies across graded, individually calibrated levels of objective value alignment. We tested this question in a controlled economic–environmental trade-off task with 250 participants. Participants first completed matching and titration tasks to estimate their subjective equivalence point between economic gain and additional good-air-quality days. They then observed five pre-programmed choices attributed to a simulated AI advisor. These choices varied by objective value alignment, with four levels defined by distance from each participant’s equivalence point, and by decision orientation, economy-leaning versus environment-leaning. Reported trust increased monotonically as AI-attributed choices moved closer to participants’ calibrated trade-off points. Perceived value alignment also increased across objective value alignment levels and was closely associated with trust, although this post-task association should not be interpreted as causal mediation. Environment-leaning choices showed a small, less robust trust advantage. These findings characterize objective value alignment as a graded cue for self-reported trust in a controlled AI-attributed decision scenario.