Reproduce “Exploring the Topology and Memory of Consensus: How LLM Agents Agree, Fragment, or Settle When Forming Conventions” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Aliakbar Mehdizadeh, Martin Hilbert
This research systematically investigates how the memory capacity of large language model agents and the communication network topology jointly influence the formation of social conventions. The study demonstrates a critical interaction between these factors, revealing that memory's effect on the speed of convention formation reverses direction depending on the network's structure, affecting global consensus or fragmentation.
Key insights
- Memory depth and communication topology are interdependent design parameters that must be co-designed, as their interaction determines the direction and magnitude of effects on coordination outcomes.
- A memory-mediated trade-off exists between the speed of local coordination and the degree of global consensus, where network structure dictates which outcome is favored by different memory lengths.
- LLM agents' choices were better described by Fictitious Play (belief-based adaptation about partners' behavior) than by Reinforcement Learning (own past payoffs), suggesting a reliance on social cues.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.