Achieving Unanimous Consensus in Decision Making Using Multi-Agents

TL;DR

This paper addresses the challenge of achieving unanimous decision-making on blockchains, proposing a novel deliberation framework where multiple LLM-based agents engage in structured rounds to produce consensus outputs. It formalizes the problem, defines definitive and prioritized decision categories, and proves core system properties—consistency, agreement, liveness, and determinism—under a layer-based architecture integrated with a gossip protocol. The approach is demonstrated via a proof-of-concept implementation on the Nimiq blockchain, evaluating convergence dynamics, block properties, and decision accuracy while discussing remedies for LLM limitations, costs, and security concerns. The work highlights the potential of LLM-driven deliberation to support inclusive, argument-rich governance on decentralized ledgers and outlines future directions for enhancing robustness and scalability.

Abstract

Blockchain consensus mechanisms have relied on algorithms such as Proof-of-Work (PoW) and Proof-of-Stake (PoS) to ensure network functionality and integrity. However, these approaches struggle with adaptability for decision-making where the opinions of each matter rather than reaching an agreement based on honest majority or weighted consensus. This paper introduces a novel deliberation-based consensus mechanism where Large Language Models (LLMs) act as rational agents engaging in structured discussions to reach a unanimous consensus. By leveraging graded consensus and a multi-round deliberation process, our approach ensures both unanimous consensus for definitive problems and graded confidence for prioritized decisions and policies. We provide a formalization of our system and use it to show that the properties of blockchains: consistency, agreement, liveness, and determinism are maintained. Moreover, experimental results demonstrate our system's feasibility, showcasing how our deliberation method's convergence, block properties, and accuracy enable decision-making on blockchain networks. We also address key challenges with this novel approach such as degeneration of thoughts, hallucinations, malicious models and nodes, resource consumption, and scalability.

Figures (9)

• Figure 1: Figure shows the layerwise representation of the system. The bottom layer is the LLM layer that produces arguments used in deliberation. The deliberation layer defines the structure and properties of deliberation where as the blockchain layer makes sure that a secure deliberation can be done.
• Figure 2: The figure illustrates the deliberation process across three rounds. The initial round gathers each model's initial opinion on the problem. The reflection round facilitates deliberation, allowing models to refine their viewpoints. Finally, the conclusion round consolidates the refined utterances from the reflection round, presenting them as viable solutions, with each model expressing its final stance.
• Figure 3: Initial Round for Wason Card game Delidata: The game objective is to select cards that will validate the rule: All cards with vowels on one side will have an even number on the other. The optimal solution is shown in the bottom right side.
• Figure 4: Reflection Round for Wason Card game: At the end of the first reflection round (T=1) all three models unanimously converge on the correct answer.
• Figure 5: BlockStruture: A block consists of block header and block body. The block header stores relevant information that maintains the chain of block where as the block body stores the transactions and the utterances during deliberation.

Theorems & Definitions (11)

• Definition 9.1
• Definition 9.2
• Definition 9.3
• Definition 9.4
• Definition 9.5
• Definition 9.6
• Definition 9.7
• Theorem 10.1
• Corollary 10.1.1
• Theorem 10.2