Powered by OpenAIRE graph
Found an issue? Give us feedback
image/svg+xml art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos Open Access logo, converted into svg, designed by PLoS. This version with transparent background. http://commons.wikimedia.org/wiki/File:Open_Access_logo_PLoS_white.svg art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos http://www.plos.org/ ZENODOarrow_drop_down
image/svg+xml art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos Open Access logo, converted into svg, designed by PLoS. This version with transparent background. http://commons.wikimedia.org/wiki/File:Open_Access_logo_PLoS_white.svg art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos http://www.plos.org/
ZENODO
Preprint . 2026
License: CC BY
Data sources: ZENODO
ZENODO
Preprint . 2026
License: CC BY
Data sources: Datacite
addClaim

Why Internal Agent Markets Fail at Small Scale: Five Hypotheses Tested on a Multi-Objective GridWorld

Authors: Yang, Haolin;

Why Internal Agent Markets Fail at Small Scale: Five Hypotheses Tested on a Multi-Objective GridWorld

Abstract

Can a “market” of competing internal modules—each with distinct preferences, staking their own resources on decisions, and subject to monopoly-breaking revolution mechanisms—produce better collective behavior than a single unified agent? We build a minimal experimental framework where multiple neural policy modules compete for action execution rights in a multi-resource GridWorld survival environment, testing five sequential hypotheses over more than fifty experimental configurations. We find that: (1) naive market competition collapses into monopoly; (2) an antitrust revolution mechanism successfully breaks monopoly and maintains diversity; (3) state-aware arbitration improves resource collection by 6× but does not surpass the single-agent baseline; (4) input-level differentiation via optimistic perceptual bias is cleaner than reward-level distortion; and (5) a learnable arbitrator achieves the first instance where the market surpasses the single agent on survival steps (83.3 vs. 73.6, with 3.5× lower variance). However, the market consistently underperforms in raw resource collection efficiency. We identify the coordination cost ceiling—the overhead of multi-module arbitration exceeds the benefit of specialization at small scale—as the fundamental limiting factor, and discuss conditions under which this ceiling may be overcome in larger systems.

Keywords

AI Alignment, Coordination Cost, artificial-intelligence, reinforcement-learning, Agent Specialization, multi-agent-systems, Internal Market Mechanism, Evolutionary Game Theory, Modular AI Systems, Multi-Agent Reinforcement Learning, Mixture of Experts

  • BIP!
    Impact byBIP!
    citations
    The total number of citations received by all articles of the researcher of interest.
    0
    popularity
    This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
    Average
    influence
    This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    Average
    impulse
    This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
    Average
Powered by OpenAIRE graph
Found an issue? Give us feedback
citations
The total number of citations received by all articles of the researcher of interest.
BIP!Citations provided by BIP!
popularity
This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
BIP!Popularity provided by BIP!
influence
This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Influence provided by BIP!
impulse
This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
BIP!Impulse provided by BIP!
0
Average
Average
Average