BOUNDARY_SYNC: Measuring Communication-Induced Representational Coupling in Multi-Agent LLM Systems

2026-07-01

AI Summary

This study investigates how communication influences representational convergence or divergence in multi-agent large language model systems. The authors propose the Coupling Amplification Factor (CAF)—the first quantifiable metric for assessing communication-induced representational coupling—and introduce the BOUNDARY_SYNC protocol to systematically analyze the roles of textual and visual communication. Using controlled experiments grounded in GPT-4o, they employ Jensen–Shannon divergence, ablation studies without communication, and prompt perturbation controls. Their findings reveal that both communication modalities significantly drive representational homogenization (CAF < 1), that group size can invert the direction of coupling, and that such coupling is contextually driven by prompt content and exhibits stateless dynamics.

Abstract

As large language models (LLMs) are deployed as communicating agents, does inter-agent communication cause outputs to converge? We introduce BOUNDARY_SYNC, a protocol measuring representational coupling via the Coupling Amplification Factor (CAF = JSD_cond / JSD_baseline), where CAF < 1 indicates homogenization and CAF > 1 indicates diversification. In controlled GPT-4o experiments (N=30, ~9,900 API calls), we measure coupling in text and image communication. Key findings: (1) text communication causes significant homogenization (CAF=0.803 [0.740, 0.873], d=1.30, p<0.001), confirmed by no-communication ablation and prompt-perturbation controls; (2) image communication also homogenizes under within-modality baselines (CAF=0.834 [0.811, 0.858]), with comparable proportional effect; (3) group size moderates coupling direction -- K=5 produces homogenization while K=3 yields CAF > 1.0 (point estimates 1.14 and 1.06, CI pending), suggesting a directional shift toward diversification; (4) cross-model replication shows extreme variation (CAF 0.034-0.803), with DeepSeek dominated by format artifacts; (5) coupling is stateless -- driven by prompt context rather than cumulative updating, with continuous consensus producing monotonic convergence. These results establish LLM agent coupling as real, measurable, and controllable at the prompt level, with direct implications for multi-agent system design.

Problem

• multi-agent LLM systems
• representational coupling
• communication-induced convergence
• output homogenization
• inter-agent communication

Innovation

• representational coupling
• Coupling Amplification Factor
• multi-agent LLM systems
• communication-induced homogenization
• prompt-level control