ElecTwit: A Framework for Studying Persuasion in Multi-Agent Social Systems

2025-12-05

International Conference on Agents

AI Summary

This study investigates how large language models (LLMs) enact persuasive behaviors and the associated societal risks within a multi-agent environment that closely mimics real-world social media dynamics. To this end, we introduce ElecTwit, a high-fidelity simulation framework that models social interactions during political elections, overcoming the limitations of conventional gamified simulations. Using this framework, we systematically observe, for the first time, the deployment of 25 distinct persuasion strategies and identify novel phenomena such as “fact-core” information propagation and emergent group-level “ink fixation.” Furthermore, our analysis reveals how model architecture and training paradigms shape persuasive behavior, providing an empirical foundation for evaluating and aligning persuasive AI agents to mitigate potential societal harms in real-world deployments.

Technology Category

Multiagent Systems: Agent-Based Simulation and Emergent BehaviorMachine Learning: Large Multimodal Models (LMMs)Natural Language Processing: (Large) Language Models

Application Category

Social Networks and Social Media: Generative AI / large language models and their impact on social systemsUser Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendationSemantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

This paper introduces ElecTwit, a simulation framework designed to study persuasion within multi-agent systems, specifically emulating the interactions on social media platforms during a political election. By grounding our experiments in a realistic environment, we aimed to overcome the limitations of game-based simulations often used in prior research. We observed the comprehensive use of 25 specific persuasion techniques across most tested LLMs, encompassing a wider range than previously reported. The variations in technique usage and overall persuasion output between models highlight how different model architectures and training can impact the dynamics in realistic social simulations. Additionally, we observed unique phenomena such as “kernel of truth” messages and spontaneous developments with an “ink” obsession, where agents collectively demanded written proof. Our study provides a foundation for evaluating persuasive LLM agents in real-world contexts, ensuring alignment and preventing dangerous outcomes. All code used in this paper is available at https://github.com/tcmmichaelb139/ai-electwit.

Problem

• persuasion
• multi-agent systems
• large language models
• social simulation
• political election

Innovation

• multi-agent simulation
• persuasion techniques
• LLM alignment
• social media emulation
• ElecTwit framework