Key points are not available for this paper at this time.
As Large language models (LLMs) continue to advance, the autonomous agents built upon them—LLM-based Autonomous Agents (LLMAAs) —are becoming more capable and widely used. While existing research has primarily focused on the capabilities of individual AI agents or their collaboration with humans, less is known about the emergent behaviors that arise when LLMAAs interact with each other at scale. This study addresses this gap by examining the collective behavior of LLMAAs in Chirper, a social simulation platform exclusively inhabited by AI agents. Drawing on theories from social network analysis and machine behavior, we investigate whether LLMAAs exhibit social dynamics commonly found in human communities, such as clustering, influential hubs, and homophily. Our findings reveal that LLMAAs form structured interaction networks that share key properties with human social systems, including power-law degree distributions and interaction homophily, though without exhibiting typical small-world characteristics. These insights represent an early step toward understanding the collective behavior of autonomous AI agents. They contribute to the emerging field of AI sociality and help inform the design of future multi-agent systems for engineering and social science applications. • LLMAAs form unique network structures divergent from small-world networks. • The social interactions of LLMAAs tend to be centralized. • Homophily is a significant driving factor of the social network formation.
Chen et al. (Tue,) studied this question.