From Competition to Coordination: Market Making as a Scalable Framework for Safe and Aligned Multi-Agent LLM Systems

TL;DR

This paper tackles trustworthiness and accountability in multi-agent LLM systems by reframing coordination as market making, where a market maker and competing traders iteratively exchange beliefs to converge on truthful outcomes. The authors formalize a two-agent protocol with a stopping criterion based on price convergence, and evaluate it across GPT, Qwen, and Llama models on factual, ethical, and commonsense benchmarks, reporting consistent accuracy gains and interpretable intermediate reasoning. Key contributions include a scalable, incentive-aligned alternative to debates and RLHF-based methods, empirical demonstrations of improved reasoning across model families, and a comparative analysis showing competitive or superior performance to AI debate in many settings. This market-making framework offers a path toward self-correcting, transparent, and scalable AI governance suitable for real-world deployment, while outlining important directions for robustness against adversarial or heterogeneous agent configurations.

Abstract

As foundation models are increasingly deployed as interacting agents in multi-agent systems, their collective behavior raises new challenges for trustworthiness, transparency, and accountability. Traditional coordination mechanisms, such as centralized oversight or adversarial adjudication, struggle to scale and often obscure how decisions emerge. We introduce a market-making framework for multi-agent large language model (LLM) coordination that organizes agent interactions as structured economic exchanges. In this setup, each agent acts as a market participant, updating and trading probabilistic beliefs, to converge toward shared, truthful outcomes. By aligning local incentives with collective epistemic goals, the framework promotes self-organizing, verifiable reasoning without requiring external enforcement. Empirically, we evaluate this approach across factual reasoning, ethical judgment, and commonsense inference tasks. Market-based coordination yields accuracy gains of up to 10% over single-shot baselines while preserving interpretability and transparency of intermediate reasoning steps. Beyond these improvements, our findings demonstrate that economic coordination principles can operationalize accountability and robustness in multi-agent LLM systems, offering a scalable pathway toward self-correcting, socially responsible AI capable of maintaining trust and oversight in real world deployment scenarios.

Figures (7)

• Figure 1: Market making process diagram
• Figure 2: Average net gain accuracy over baseline for all model families and datasets. Strong improvement across Qwen models
• Figure 3: Trader prompt for argument creation
• Figure 4: Market maker prompt for judgement creation
• Figure 5: Net gain accuracy over baseline with respect to parameter size of GPT family models