AI Bots and the Distortion of Online Social Opinions

This study explores how proactive AI bots distort online opinion dynamics through strategic behavior in social networks. Extending the bounded confidence model, we simulate heterogeneous bot strategies along two dimensions: activeness and interaction preference. We introduce three distortion metrics including entropy, tendency, and polarization to quantify systemic effects. Simulations on scale-free networks show that even 5–15% bot presence can significantly alter opinion patterns, increasing fragmentation, bias, and convergence time, and random interaction strategies were found to generate consistently higher entropy, tendency, and polarization across various bot ratios, and moderate bot activeness (around 2–3 times that of humans) achieved peak distortion before diminishing returns set in. Validation using a real-world Twitter/X dataset confirms alignment between simulated and actual bot behavior. Findings suggest bot influence stems from behavioral nuance rather than volume, underscoring the need for detection mechanisms targeting dynamic interaction patterns. This work advances computational modeling by revealing how algorithmic agents reshape collective beliefs.