No Free Signal: A Negative Result for Substrate-Evolution Around Fixed LLMs in an Embodied Multi-Agent Population
Sterling Morrison · GDF Research · DOI 10.5281/zenodo.20048295
Abstract
Can evolutionary pressure on the communication substrate — rather than on model weights — make a frozen LLM more adaptive inside a 25-creature multi-agent system? A seven-arm design with 140 controlled runs, including mute baselines, scrambled models, and cadence-matched noise emitters, finds the full treatment does not outperform any control on fitness. Published as the null result it is. A behavioral analysis still detects emission-source effects on receiver responses, with methodological contributions around cadence-matched noise controls.
The question
Can evolutionary pressure on the communication channel around a frozen language model, instead of on the model's weights, make that model more useful inside a population of embodied agents?
What we did
A seven-arm design, 20 seeds per arm, 15,000 ticks per run, 140 runs in all, with a 25-creature population and Amazon Nova Lite as the frozen model. The controls were chosen to rule out cheap explanations: mute creatures, no emitter, a scrambled model, replayed model output, and a random emitter matched to the model's cadence.
What we found
The full treatment didn't outperform any control on fitness. The cleanest internal contrast, replayed model output against a random emitter at the same cadence, was a coin flip. A behavioral analysis did find small but real differences in how receivers responded to different emission sources, which population fitness alone hid.
It's published as the null result it is, with the cadence-matched noise control as a methodological contribution for anyone running similar studies.
Limitations
- One frozen model and one simulated world. A different environment or model could behave differently.
- Raw behavioral logs are 22 GB and available on request; the analyzed output is in the repository.
Cite
Morrison, S. (2026). No Free Signal: A Negative Result for Substrate-Evolution Around Fixed LLMs in an Embodied Multi-Agent Population. Zenodo. https://doi.org/10.5281/zenodo.20048295
@misc{morrison2026no,
author = {Morrison, Sterling},
title = {No Free Signal: A Negative Result for Substrate-Evolution Around Fixed LLMs in an Embodied Multi-Agent Population},
year = {2026},
publisher = {Zenodo},
doi = {10.5281/zenodo.20048295},
url = {https://doi.org/10.5281/zenodo.20048295}
}