Title: Economy of Minds: Emerging Multi-Agent Intelligence with Economic Interactions

Abstract:

In the absence of centralized command, how can a collective of agents autonomously organize and adapt to foster superior group intelligence? Drawing inspiration from Friedrich Hayek’s theories on decentralized market coordination, this study investigates this phenomenon through a simulated "agent economy." Within this framework, agents engage in auctions to secure the right to perform actions, facilitating the exchange of payments and the accumulation of wealth derived from environmental rewards. These fundamental economic signals serve to distribute credit assignment across the system, enabling sophisticated planning without the need for global orchestration or predefined communication protocols.

The population undergoes an evolutionary process driven by economic selection. Agents that demonstrate efficacy accumulate wealth and are refined through exploitation, whereas those that fail to succeed face bankruptcy and are subsequently replaced through exploration. Our experiments demonstrate that even when initialized with relatively weak agents, this economic model generates emergent multi-step reasoning capabilities. Consequently, it surpasses stronger monolithic baseline models across five distinct agentic domains: mathematical reasoning, financial research, scientific research, accelerator design, and distributed-system optimization.

Additionally, we offer theoretical perspectives on how economic dynamics influence agent behavior, establishing a connection between local incentives and long-term global performance. These findings propose a novel trajectory for achieving multi-agent intelligence: rather than manually engineering coordination mechanisms, we can construct decentralized incentive structures that allow such coordination to arise organically.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC