Jour Fixe - Artificial (Partisan) Intelligence? Political Affiliations and Human-AI Interaction

Artificial (Partisan) Intelligence? Political Affiliations and Human-AI Interaction

Political affiliations can shape individuals' behaviors and interaction with each other. But do political affiliations also play a role in the context of human-AI interaction? We conduct a survey and a controlled online experiment to examine on an individual level whether perceptions about an AI system developer's political affiliations affect users' interaction with the system and its assessment. We produce consistent evidence that perceived differences between the users' and the developers' political affiliations decrease the users' likelihood of accessing and utilizing an AI system. Examining the causal mechanisms underlying these patterns, we find that neither changes in users' beliefs about the AI system's accuracy nor trust calibrations can account for the observed effects. This finding lets us conclude that an in-group bias underlies the influence of political affiliations on human-AI interaction. Our findings offer first evidence that perceptions about the developers behind AI systems, in particular their political affiliations, can constitute a source of algorithm aversion. From a practical standpoint, our research provides valuable insights for managers contemplating AI development outsourcing and policymakers focused on transparency regulations.