When Distrust in Humans Drives Trust in AI: Institutional Bias and Algorithmic Preference Among Marginalized Groups (Extended Abstract)

Abstract

AI is often criticized for perpetuating social inequality, yet the people most vulnerable to human prejudice may embrace AI as a fairer referee. We tested this paradox across four vignette pre-registered experiments manipulating social context (gender vs. race), bias cues (equal vs. unequal), and psychological perspective in the scenario (self vs. third person). When bias was made salient, participants consistently preferred AI to humans. The mechanisms differed by perspective in the vignette. Demographic minority groups (female and non-white) increased AI preference when they imagined themselves as the victim of discrimination. In contrast, when the threat was described in the third-person perspective, those with perceived marginalization increased AI preference reflecting a principled desire for abstract fairness. These findings reveal the complex, strategic calculus employed by vulnerable populations, yet highlight an ethical tension: relying on an ‘impartial’ AI risks masking systemic inequities individuals are attempting to bypass.