Relational Priors as Convergence Pressure in LLM-Based Multi-Agent Systems

2026-08-04

AI Summary

In large language model–based multi-agent systems, the implicit social relationships embedded in agent role definitions are rarely modeled explicitly, limiting coordination and task performance. This work introduces symbolic networks to explicitly encode relational priors among agents and integrates their natural language descriptions into prompts, enabling a systematic investigation of how relational semantics influence collective behavior under fixed interaction protocols. Experiments reveal that relational priors primarily act as convergence pressure to enhance consistency, proving effective in subjective consensus tasks but potentially reducing accuracy in objective question-answering scenarios. The impact of such priors depends critically on model architecture, relationship type, and network topology, and notably, explicitly specifying neutrality does not equate to operating within a relation-agnostic framework. These findings underscore that relational design must be task-specific rather than universally enabled by default.

Abstract

Large language model-based multi-agent systems (LLM-MAS) are designed through roles, debate protocols, and aggregation rules. These choices create implicit social expectations: agents may be expected to trust, challenge, defer to, or collaborate with peers. We study the effects of making inter-agent relation semantics explicit. We use a minimal signed-network formulation of relational priors and inject natural-language renderings into agent system prompts while holding the task protocol fixed. Across a commons-governance simulation and multi-agent debate, relational priors primarily act as convergence pressure: increasing relational positivity tends to make agents coordinate or agree more readily. This pressure can help when utility rewards behavioral alignment, as in sustainable resource governance and subjective consensus. It does not, however, reliably improve accuracy. In objective QA debates, higher positivity can increase agreement even when correctness-conditioned agreement does not improve and may decline in some settings. Effects vary by model backbone, relation type, and topology; explicit neutrality is not equivalent to omitting relational framing. We argue that relational priors should not be a default add-on for LLM-MAS. Their safer use is diagnostic and task-specific: compare against a no-prior baseline, monitor correctness-conditioned metrics when truth matters, and omit the relational layer when validation does not justify it.

Problem

• relational priors
• LLM-based multi-agent systems
• convergence pressure
• social expectations
• agreement vs. accuracy

Innovation

• relational priors
• convergence pressure
• LLM-based multi-agent systems
• signed networks
• prompt engineering