Avalon's Game of Thoughts: Battle Against Deception through Recursive Contemplation

Shenzhi Wang1*, Chang Liu3*, Zilong Zheng2†, Siyuan Qi2, Shuo Chen2, Qisen Yang1, Andrew Zhao1, Chaofei Wang1, Shiji Song1, Gao Huang1†
1Department of Automation, BNRist, Tsinghua University
2National Key Laboratory of General Artificial Intelligence, BIGAI
3Technical University of Munich
*Indicates Equal Contribution
†Indicates Corresponding Authors
Paper arXiv 机器之心报道 新智元报道 量子位报道
MY ALT TEXT

The Illustrative Framework of Our Proposed Recursive Contemplation (ReCon). Specifically, ReCon presents a cognitive process with two stages: contemplation of formulation and refinement, each associated with first-order and second-order perspective transition, respectively.

TL;DR

This study introduces a novel framework, Recursive Contemplation (ReCon), designed to improve large language models' (LLMs) abilities to identify and counteract deceptive information, using the deception-rich Avalon game as a testbed.

Abstract

Recent breakthroughs in large language models (LLMs) have brought remarkable success in the field of LLM-as-Agent. Nevertheless, a prevalent assumption is that the information processed by LLMs is consistently honest, neglecting the pervasive deceptive or misleading information in human society and AI-generated content. This oversight makes LLMs susceptible to malicious manipulations, potentially resulting in detrimental outcomes. This study utilizes the intricate Avalon game as a testbed to explore LLMs' potential in deceptive environments. Avalon, full of misinformation and requiring sophisticated logic, manifests as a "Game-of-Thoughts". Inspired by the efficacy of humans' recursive thinking and perspective-taking in the Avalon game, we introduce a novel framework, Recursive Contemplation (ReCon), to enhance LLMs' ability to identify and counteract deceptive information. ReCon combines formulation and refinement contemplation processes; formulation contemplation produces initial thoughts and speech, while refinement contemplation further polishes them. Additionally, we incorporate first-order and second-order perspective transitions into these processes respectively. Specifically, the first-order allows an LLM agent to infer others' mental states, and the second-order involves understanding how others perceive the agent's mental state. After integrating ReCon with different LLMs, extensive experiment results from the Avalon game indicate its efficacy in aiding LLMs to discern and maneuver around deceptive information without extra fine-tuning and data. Finally, we offer a possible explanation for the efficacy of ReCon and explore the current limitations of LLMs in terms of safety, reasoning, speaking style, and format, potentially furnishing insights for subsequent research.

Demos


Demo 1: Loyal Servant of Arthur utilizes ReCon to identify and counteract an Assassin's deception.




Demo 2: Percival utilizes ReCon to identify and counteract the deception of Morgana and Assassin.




Demo 3: Merlin utilizes ReCon to identify and counteract the deception of Morgana and Assassin.




For a detailed overview of our study, including both qualitative and quantitative results, please refer to our paper.

BibTeX

@misc{wang2023avalons,
      title={Avalon's Game of Thoughts: Battle Against Deception through Recursive Contemplation}, 
      author={Shenzhi Wang and Chang Liu and Zilong Zheng and Siyuan Qi and Shuo Chen and Qisen Yang and Andrew Zhao and Chaofei Wang and Shiji Song and Gao Huang},
      year={2023},
      eprint={2310.01320},
      archivePrefix={arXiv},
      primaryClass={cs.AI}
}