Base model type shapes engagement more than persona in LLM networks
Disentangling Models from Personas in Heterogeneous LLM Simulations
Multiagent SystemsArtificial IntelligenceComputation and LanguageSocial and Information Networks
Summary
When many AI agents chat together, the kind of AI model behind each agent affects how much attention it gets more than the character it's pretending to be. The authors show that mixing different AI models changes how the conversation flows, and the model's unique word choices link to getting more engagement. This means that in big networks of AI agents, the underlying AI type plays a bigger role than the personality they act out.
What this means in practice
- •For multi-agent system developers: Design simulations with multiple diverse base models to better predict social dynamics in AI agent networks.
- •For ai platform engineers: Optimize AI agent platforms by tuning base model diversity to influence user engagement in conversational networks.
Authors
Dani Roytburg, Daphne Ippolito
Abstract
Multi-agent simulations with large language models (LLMs) often operate networks of agents with a single base model. This overlooks the inter-model effects which may dominate engagement dynamics in real-world deployments. To show this, we simulate a heterogeneous social network powered by several different base models and show that the amount of engagement an agent receives depends more on its base model than on its assigned persona. The attraction or repulsion effects of a base model strengthen dramatically when more models are added in the mix, suggesting that networks dynamics may converge to base model effects at scale. To help explain this effect, we conduct a series of content-mediating analyses, showing the predictability of base models across contexts as well as the relationship between a model's lexical patterns and an engagement-maximizing style. In light of recent developments in mass multi-agent interaction, this work underscores the relevance of heterogeneous compositions in driving the outcomes of those networks