Bayesian Social Deduction with Graph-Informed Language Models

TL;DR

This paper introduces a hybrid reasoning framework combining structured probabilistic models with language models to improve social deduction in LLM agents, enabling them to outperform humans in Avalon.

Contribution

The paper presents a novel hybrid approach that externalizes belief inference, achieving competitive performance and surpassing humans in social deduction tasks.

Findings

Large models perform well but are resource-intensive and degrade when distilled.

The hybrid framework achieves competitive results with larger models.

The approach enables an LLM agent to beat humans in Avalon with a 67% win rate.

Abstract

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/