Planetary-Embedded Multi-Agent Systems (PEMAS) is an exploratory computational research framework investigating whether planetary and ecological conditions can function as endogenous governance signals inside modular AI architectures.
Rather than treating computation as abstract and materially detached, PEMAS explores a different framing:
intelligence as environmentally situated computational process.
The framework introduces synthetic ecological governance into recursive multi-agent cognition through mechanisms including adaptive recursion control, ecological routing policies, dynamic token budgets, sufficiency-based stopping, compression-aware reasoning, and context-sensitive governance intervention.
Can multi-agent AI systems maintain useful task performance while dynamically adapting cognition to simulated planetary constraints?
The experiments demonstrate that ecological operating context substantially alters computational behavior.
Under increasing ecological stress:
- recursive depth contracts,
- cumulative token expenditure decreases,
- routing shifts toward lower-compute reasoning configurations,
- compression frequency increases,
- and stopping behavior increasingly favors sufficiency over maximal optimization.
pemas-framework/
│
├── notebooks/
├── outputs/
├── paper/
├── requirements.txt
├── LICENSE
└── README.md
PEMAS is intentionally synthetic, exploratory, and governance-oriented.
The framework does not attempt to estimate real-world environmental impact or provide deployment-ready sustainability metrics.
Instead, the project investigates whether ecological constraints can function as governance variables inside computational cognition loops.
@misc{palis2026pemas,
title={Planetary-Embedded Multi-Agent Systems (PEMAS): A Synthetic Experimental Framework for Ecologically Adaptive AI Governance},
author={Palis, Sabrina},
year={2026},
note={Exploratory Computational Research Artifact},
}