Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Influencing LLM Multi-Agent Dialogue via Policy-Parameterized Prompts

Hongbo Bo, Jingyu Hu, Weiru Liu

TL;DR

This work investigates whether prompt-as-action can be parameterized so as to construct a lightweight policy which consists of a sequence of state-action pairs to influence conversational behaviours without training and shows that prompt parameterization can influence the dialogue dynamics.

Abstract

Large Language Models (LLMs) have emerged as a new paradigm for multi-agent systems. However, existing research on the behaviour of LLM-based multi-agents relies on ad hoc prompts and lacks a principled policy perspective. Different from reinforcement learning, we investigate whether prompt-as-action can be parameterized so as to construct a lightweight policy which consists of a sequence of state-action pairs to influence conversational behaviours without training. Our framework regards prompts as actions executed by LLMs, and dynamically constructs prompts through five components based on the current state of the agent. To test the effectiveness of parameterized control, we evaluated the dialogue flow based on five indicators: responsiveness, rebuttal, evidence usage, non-repetition, and stance shift. We conduct experiments using different LLM-driven agents in two discussion scenarios related to the general public and show that prompt parameterization can influence the dialogue dynamics. This result shows that policy-parameterised prompts offer a simple and effective mechanism to influence the dialogue process, which will help the research of multi-agent systems in the direction of social simulation.

Influencing LLM Multi-Agent Dialogue via Policy-Parameterized Prompts

TL;DR

This work investigates whether prompt-as-action can be parameterized so as to construct a lightweight policy which consists of a sequence of state-action pairs to influence conversational behaviours without training and shows that prompt parameterization can influence the dialogue dynamics.

Abstract

Large Language Models (LLMs) have emerged as a new paradigm for multi-agent systems. However, existing research on the behaviour of LLM-based multi-agents relies on ad hoc prompts and lacks a principled policy perspective. Different from reinforcement learning, we investigate whether prompt-as-action can be parameterized so as to construct a lightweight policy which consists of a sequence of state-action pairs to influence conversational behaviours without training. Our framework regards prompts as actions executed by LLMs, and dynamically constructs prompts through five components based on the current state of the agent. To test the effectiveness of parameterized control, we evaluated the dialogue flow based on five indicators: responsiveness, rebuttal, evidence usage, non-repetition, and stance shift. We conduct experiments using different LLM-driven agents in two discussion scenarios related to the general public and show that prompt parameterization can influence the dialogue dynamics. This result shows that policy-parameterised prompts offer a simple and effective mechanism to influence the dialogue process, which will help the research of multi-agent systems in the direction of social simulation.
Paper Structure (37 sections, 2 equations, 5 figures, 19 tables)

This paper contains 37 sections, 2 equations, 5 figures, 19 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. LLMs Agents and Social Simulation.
  4. LLM Agents Formalization.
  5. Prompt Control as Lightweight Policy Parameterization
  6. Multi-Agent Formalization
  7. Agent
  8. State and Retrieval
  9. Prompt-as-Action
  10. Action Execution and Output
  11. Policy Parameterization
  12. Rule Templates R
  13. Weights Design.
  14. Adaptive Weights
  15. Time-based trend updating
  16. ...and 22 more sections

Figures (5)

  • Figure 1: The overall framework illustrates the process from an agent’s state representation to action generation, LLM-based action execution, and evaluation.
  • Figure 2: Round-wise changes of three dialogue metrics under different rule templates. Each curve shows the mean value (with 95% confidence interval).
  • Figure 3: An Example of Adaptive Weight Changes. Farmer agent's $W_D$ and $W_M$ changes in 10 rounds under the Q1 topic.
  • Figure 4: Round-wise changes of three dialogue metrics with adaptive weights.
  • Figure 5: Ablation Study of Components T, M and D.