LLM Agents for Deliberative Collaboration: A Study on Joint Decision Making Under Partial Observability

Jul 16, 2026

Upload by Oliver

Deliberation enables effective collaboration, yet joint decision-making becomes challenging when agents share only partial, asymmetric observations. This work studies deliberative large language model (LLM) agents in cooperative joint decision tasks, formalizing the interaction as a partially observable consensus problem with shared reward. A scalable benchmark spans multiple domains and task settings, together with a reference evaluation scaffold and protocol. Systematic experiments show state-of-the-art models still struggle, though diagnostic analysis finds deliberation can support reflection and error correction, sometimes improving over centralized baselines.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages22

Reading Time55 min

Abstract

Introduction

Deliberation to Decision

LLM Agents for Deliberative Collaboration: A Study on Joint Decision Making Under Partial Observability Chenxu Wang*,1 , Yongkun Yang*,1 , Boyuan Du*,2 , Shiwei Lin1 , Huaping Liu1 1Department of Computer Science and Technology, Tsinghua University 2Fuzhou University [jimwangcx@gmail.com](jimwangcx@gmail.com), [hpliu@tsinghua.edu.cn](hpliu@tsinghua.edu.cn) arXiv :2607 .06 157v 1 [ cs .CL] 7 Jul 2026 Abstract Deliberation plays a crucial role in collaboration; when humans work together, they naturally engage in communication to align information and reach an agreement. In this paper, we investigate deliberative large language model (LLM) agents under partially observable joint decision-making tasks. We formalize deliberative collaboration as a cooperative joint decision problem with partial and asymmetric observations, and introduce a scalable benchmark that instantiates this problem across multiple task settings and domains in which agents must exchange information through deliberation to reach a joint decision with a shared reward. We then instantiate a reference scaffold and evaluation protocol for deliberative agents and conduct a systematic evaluation of a range of representative LLMs. The results reveal that complex deliberative collaboration tasks continue to challenge state-of-the-art language models. Even with the aid of external mathematical tools, language models may fail in either the deliberation process for aligning information or the complex reasoning process for making the decision. On the other hand, diagnostic analysis reveals that the deliberation process may also provide opportunities for reflection and error correction, sometimes improving performance over centralized baselines. Altogether, our work establishes a foundation for evaluating and improving LLM agents indeliberative collaboration and provides insights into the strengths, limitations, and properties of current LLM-based multi-agent systems. Code is available at this repository. 1 Introduction Recent advances in large language models (LLMs) and language agents have achieved impressive progress across a variety of domains, including social intelligence (Zhou et al., 2025 ; Mou et al., *Equal contribution. 2025 ; Wang et al., 2024a), multi-agent collaboration (Yan et al., 2025 ; Chen et al., 2025 ; Guo et al., 2024 ; Liu et al., 2024a), negotiation or deliberation (Zhu et al., 2025b ; Choi et al., 2025 ; Hu et al., 2025 ; Bianchi et al., 2024 ; Karanam et al., 2024), and decision making (Dolant and Kumar, 2025 ; Eo et al., 2025) . However, an open question remains: can LLM agents collaborate well through deliberation when they each possess only partial, asymmetric information? We present a representative example of deliberative collaboration in Figure 1. In deliberative collaboration scenarios, the agents must communicate, exchange their knowledge, and reason together to reach a shared decision. This setting is inherently more complex and comprehensive than pure deliberation or collaboration, as it demands task-related communication and joint decision-making. In this paper, we study LLM agents in deliberative collaboration scenarios. We formalize deliberative collaboration as a partially observable joint decision-making problem, where multiple agents each have their own partial observation of the environment, and the agents should reach a consensus on a final decision as the output and share the same reward. Unlike competitive or adversarial settings, the challenge in such scenarios lies in effective information exchange, belief alignment, and coordinated decision-making under partial observability. We build a scalable benchmark that instantiates this problem across multiple related task settings, across collaborative menu design and task allocation domains. The settings differ in observation structure, agent role, and decision mechanism, while sharing the same underlying deliberative collaboration structure. The underlying d

FAQs

How does the paper define deliberative collaboration for LLM agents?

It formalizes deliberative collaboration as a partially observable joint decision-making problem where multiple agents hold partial, asymmetric observations and must reach a shared consensus decision with a shared reward.

What does the proposed benchmark include?

The benchmark instantiates the same underlying deliberative collaboration structure across multiple task settings and domains, varying observation structure, agent roles, and decision mechanisms, while using rule-based numerical rewards.

What do the experimental results suggest about current LLMs?

Complex deliberative collaboration tasks remain difficult for state-of-the-art language models; even with external mathematical tools, models may fail in information alignment during deliberation or in the final decision reasoning.

Jul 16, 2026

Upload by Oliver