Socializing with Artificial Intelligence: An Experiment of AI Interaction Effects on Sociability Attitudes

The rapid integration of generative AI into everyday life has intensified scholarly interest in its potential to fulfill social roles traditionally reserved for humans, including companionship and emotional support. Prior experimental research has shown that chatbot and social-robot characteristics can shape social presence, perceived humanness, parasocial responses, and technology-related attitudes. Less explored remains how brief direct interaction with an open-ended generative AI chatbot compares with informational exposure in shaping immediate perceptions of AI’s social and relational potential. To address this gap, we conducted a pre-registered online experiment with 423 U.S. adults. Participants were randomly assigned to one of four conditions: control (no AI exposure), information-only (reading about AI capabilities and biases), interaction-only (a 4-minute social conversation featuring an API’s customized GPT-3.5-Turbo AI prompted for social exchanges), or information plus interaction. We measured perceptions of AI socialization viability, attitudes toward AI socialization, emotional support, friendship potential, and social presence. Results showed that both friend-framed interaction conditions generally produced more favorable perceptions than control, whereas informational exposure alone produced no significant differences. Anthropomorphism moderated only AI socialization viability, so that interaction–control differences narrowed as anthropomorphism increased. All in all, brief, friend-framed AI interaction can shape immediate perceptions of AI’s social potential.