The More, The Stronger? Investigating How Multi-Agent AI Shapes Human Opinions

As AI agents become increasingly interactive, their potential to shape human opinions raises concerns about bias reinforcement, misinformation, and manipulation. While prior research has examined how individual AI agents influence users, it remains unclear whether multi-agent AI systems exert stronger influence, similar to human group effects.Drawing on social influence theory, we investigate whether a group of AI agents can amplify opinion shifts compared to a single agent. In an empirical study where participants discussed two paintings with either one or five AI agents, we found that multi-agent interactions led to significantly stronger opinion shifts. Participants aligned more closely with the AI group's expressed stance, suggesting that increasing the number of agents enhances social influence.These findings highlight both the opportunities and risks of multi-agent AI in shaping user opinions, with implications for persuasive and ethical AI design. Additionally, we identified several factors that may moderate the influence of multi-agent AI, including users' prior beliefs on the topic, perceived in-authenticity of AI-generated comments, and alignment between human and AI preferences. Considering these factors in future AI system design could help balance influence effectiveness with user autonomy and trust.