The Politics of Using AI in Policy Implementation: Evidence from a Field Experiment

Abstract The use of AI by government agencies in guiding important decisions (for example, on policing, welfare, education) has triggered backlash and demands for greater public input in AI regulation. Yet it remains unclear what such input would reflect: general attitudes towards new technologies, personal experience with AI, or learning about its implications. We study this question experimentally by tracking the attitudes of over 1,500 workers whose task assignments were randomly determined by either a human or an AI ‘boss’, with task content and valence also randomized. Across a three-wave panel, we find that personal experience with AI-as-boss affected workers’ job performance but not their attitudes on using AI in public decision making. In contrast, exposure to information about the technology produced significant attitudinal change, even when it conflicted with participants’ prior disposition or direct experience. The results highlight the promise of incorporating public input into AI governance.