Learning to Make Friends: Coaching LLM Agents toward Emergent Social Ties

TL;DR

This work investigates whether LLM agents can reproduce emergent social dynamics observed in human online communities, including homophily, reciprocity, and social validation, by introducing memory and learning mechanisms. It proposes a multi-agent LLM simulation with separate public and private channels, task-aligned reward functions, and an in-context coaching signal to accelerate adaptation, enabling end-to-end formation of social ties. The authors formalize five behavioral rewards—SOC, INF, PRE, COORD, and EMO—and integrate them into a compositional reward $R_t(u) = \sum_{r \in \mathcal{R}} \lambda_r R_t^r(u)$ learned via Plan–Execute–Reflect loops. They show that coached agents develop stable interaction patterns and endogenously form network structures with statistical properties similar to real online communities, validating the framework as a controlled testbed for studying echo chambers and interventions. The work highlights both the promise and limitations of simulating human-like social behavior with LLM populations and outlines clear paths for scaling and policy-relevant experiments.

Abstract

Can large language model (LLM) agents reproduce the complex social dynamics that characterize human online behavior -- shaped by homophily, reciprocity, and social validation -- and what memory and learning mechanisms enable such dynamics to emerge? We present a multi-agent LLM simulation framework in which agents repeatedly interact, evaluate one another, and adapt their behavior through in-context learning accelerated by a coaching signal. To model human social behavior, we design behavioral reward functions that capture core drivers of online engagement, including social interaction, information seeking, self-presentation, coordination, and emotional support. These rewards align agent objectives with empirically observed user motivations, enabling the study of how network structures and group formations emerge from individual decision-making. Our experiments show that coached LLM agents develop stable interaction patterns and form emergent social ties, yielding network structures that mirror properties of real online communities. By combining behavioral rewards with in-context adaptation, our framework establishes a principled testbed for investigating collective dynamics in LLM populations and reveals how artificial agents may approximate or diverge from human-like social behavior.

Figures (8)

• Figure 1: Multi-agent LLM social media conversation framework.
• Figure 2: Comparison between augmenting in-context learning with coach.
• Figure 3: Network metrics for heuristic tie formation.
• Figure 4: Network metrics for text-based tie formation.
• Figure 5: Degree distributions for heuristic tie formation without coach.