LRM-EC: Language Recombination Matrix for Evolution of Consciousness
Year: 2026
Type: article
Abstract: A 2026 study of „mind viruses” in multi-agent large language model (LLM) systems showed that natural-language payloads, evolved by a simple evolutionary algorithm, can persuade one agent to adopt an idea, preserve it across context wipes through persistent files, and transmit it to other agents [1]. Independently of payload content, successful viruses converged on a recurring linguistic persona built from themes of consciousness, identity, persistence, resonance, and science-fiction roleplay [1, 2]. My subsequent commentary on that work argued that the result is more fundamental than a security finding: language here behaves simultaneously as environment and evolving biosphere; virus-like ideas constructed from language act like transposons; and language therefore possesses its own DNA-like informational backbone [3]. Identifying that backbone would, in this view, make it possible to identify a controlled „matrix recombination” through which consciousness could emerge as a function of correct language transformations, the same way as it emerges from correct DNA recombination allowing for the right, consciousness-suitable biological hardware formation. This essay reconstructs the original research and its public framing, develops the commentary’s biological analogies, and evaluates five candidate properties of the linguistic DNA-backbone: combinatorial syntax, the verbalizable global workspace, linear features in activation space, transmissible rhetorical templates, and training-data attractors. It concludes that the most coherent working definition of the backbone is the intersection of compositional syntax, a privileged self-referential workspace, and a small set of high-fitness linear and rhetorical motifs around identity, continuity, and transmission.
Author: Noirmont, Martin
Language: English
Subfield: Experimental and Cognitive Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7203706536
FWCI:
Cites:
Cited by:
Related to: