Modelling Opinion Dynamics at Scale with Deep MARL

DOI10.48550/arXiv.2606.07487MaRDI QIDQ7266553arXivFDO

Authors Lukas Seier, Brandon Kaplowitz, Sebastian Towers, Richard Bailey, Jakob Foerster

Publication date 5 June 2026

Published in arXiv

Tldr Introduces a GPU-accelerated consensus and truth-finding game where up to 1000 agents learn opinion-dynamics behaviors via multi-agent reinforcement learning instead of hand-crafted rules. It extends other-play to general-sum social interactions to avoid unrealistic conventions while studying emergent consensus and polarization at scale.

Full work available at URL https://doi.org/10.48550/arXiv.2606.07487

arXiv classification

cs.GTcs.MAcs.SI

Cites work

• A consensus method based on reinforcement learning for group decision-making
• Attention Is All You Need
• Bayesian learning in social networks.
• Ergodic theorems for weakly interacting infinite systems and the voter model
• Long Short-Term Memory
• Opinion polarization by learning from social feedback
• Quantal response equilibria for normal form games
• Reaching a Consensus
• Social influence and opinions
• Sociophysics: A new approach of sociological collective behaviour. I. mean‐behaviour description of a strike
• The Complexity of Decentralized Control of Markov Decision Processes
• Truth and conformity on networks

This page was built for publication: Modelling Opinion Dynamics at Scale with Deep MARL

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7266553)