Artificial Intelligence Circumvents Identity-Driven Biases in Source Selection

Social identity profoundly shapes whom people choose as information sources, constraining exposure to diverse perspectives. While people are motivated to seek accurate information, they systematically avoid outgroup sources even when group membership is irrelevant to the task at hand. Here we investigate whether artificial intelligence (AI) can circumvent these identity-driven biases in source selection. In Study 1, a nationally representative sample of American adults (n = 1,054) preferred AI over human sources when seeking information about political conflicts. In Study 2 (n = 284), an incentivized political fact-checking experiment revealed that participants preferred AI sources over outgroup (d = 0.470) and even ingroup (d = 0.230) partisan sources, despite recognizing they were of equal competence. In Study 3 (n = 277), using an identity-irrelevant shape categorization task, participants only preferred AI over outgroup sources (d = 0.191), with no difference between AI and ingroup sources. Computational modeling revealed that these preferences emerge through selectively accumulated evidence against partisan advisors during deliberation, rather than differences in priors. These findings suggest that AI's perceived neutrality enables it to bypass identity-based discrimination. These results highlight the potential of AI to reduce echo chambers and broaden epistemic exposure by serving as an identity-neutral conduit for information acquisition.