Bayesian Social Deduction with Graph-Informed Language Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Rahimirad, Shahab, Gergerli, Guven, Romero, Lucia, Qian, Angela, Olson, Matthew Lyle, Stepputtis, Simon, Campbell, Joseph
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911581971939328
author Rahimirad, Shahab
Gergerli, Guven
Romero, Lucia
Qian, Angela
Olson, Matthew Lyle
Stepputtis, Simon
Campbell, Joseph
author_facet Rahimirad, Shahab
Gergerli, Guven
Romero, Lucia
Qian, Angela
Olson, Matthew Lyle
Stepputtis, Simon
Campbell, Joseph
contents Social reasoning - inferring unobservable beliefs and intentions from partial observations of other agents - remains a challenging task for large language models (LLMs). We evaluate the limits of current reasoning language models in the social deduction game Avalon and find that while the largest models demonstrate strong performance, they require extensive test-time inference and degrade sharply when distilled to smaller, real-time-capable variants. To address this, we introduce a hybrid reasoning framework that externalizes belief inference to a structured probabilistic model, while using an LLM for language understanding and interaction. Our approach achieves competitive performance with much larger models in Agent-Agent play and, notably, is the first language agent to defeat human players in a controlled study - achieving a 67% win rate and receiving higher qualitative ratings than both reasoning baselines and human teammates. We release code, models, and a dataset to support future work on social reasoning in LLM agents, which can be found at https://camp-lab-purdue.github.io/bayesian-social-deduction/
format Preprint
id arxiv_https___arxiv_org_abs_2506_17788
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Bayesian Social Deduction with Graph-Informed Language Models
Rahimirad, Shahab
Gergerli, Guven
Romero, Lucia
Qian, Angela
Olson, Matthew Lyle
Stepputtis, Simon
Campbell, Joseph
Artificial Intelligence
Computation and Language
Machine Learning
Multiagent Systems
I.2.1; I.2.7
Social reasoning - inferring unobservable beliefs and intentions from partial observations of other agents - remains a challenging task for large language models (LLMs). We evaluate the limits of current reasoning language models in the social deduction game Avalon and find that while the largest models demonstrate strong performance, they require extensive test-time inference and degrade sharply when distilled to smaller, real-time-capable variants. To address this, we introduce a hybrid reasoning framework that externalizes belief inference to a structured probabilistic model, while using an LLM for language understanding and interaction. Our approach achieves competitive performance with much larger models in Agent-Agent play and, notably, is the first language agent to defeat human players in a controlled study - achieving a 67% win rate and receiving higher qualitative ratings than both reasoning baselines and human teammates. We release code, models, and a dataset to support future work on social reasoning in LLM agents, which can be found at https://camp-lab-purdue.github.io/bayesian-social-deduction/
title Bayesian Social Deduction with Graph-Informed Language Models
topic Artificial Intelligence
Computation and Language
Machine Learning
Multiagent Systems
I.2.1; I.2.7
url https://arxiv.org/abs/2506.17788