Answers are grounded in this paper and cite their sources.
Simulation study demonstrates emergent multi-agent cooperation in competitive environments without reward functions, indicating the viability of homeostatic architectures.
Second part of a series in three: the engine (DOI 10.5281/zenodo.21026795), the hive (this work), the pack (next). An agent built for machina_1, the competitive environment of Cogs vs Clips (Softmax), governed solely by G-EMV, a published homeostatic geometry: no reward function, no training, no scripted behaviors. Over complete 2,500-step lives, a team of eight identical agents mines, supplies a common pantry, manufactures equipment, conquers and holds junctions, and cooperates through an anonymous notice board, with every number verified against its results file. The work documents the architecture piece by piece, the method (criteria written before looking), the adult agent's biography, the measured value of communication, why its deliberation could not be cached or pruned, and the full map of its limits. Three GIF figures are included as supplementary material; the Spanish version and the source markdown are available in the project repository.Contact: manelenrico@gmail.com — ORCID: 0009-0008-1732-6310 — Repository and Spanish version: https://github.com/Manelenrico/g-emv/tree/main/paper2
No takes yet. Share an insight, caveat, or question.
Manel Enrico (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: