Two Geneses of Cognition Beyond Classical AI: A Conceptual Framework for Language-First Cognitive Emergence in Large Language Models

Abstract

This paper proposes a conceptual distinction between two possible geneses of cognition. The first is the familiar biological path described by autopoietic and enactivist traditions: life, perception, language, and then reflective thought. The second is a hypothetical machine-native path in which higher-order organization emerges above language without first reproducing the full biological stack. On this view, large language models do not instantiate biological cognition, consciousness, or life. Yet sufficiently large-scale recursive operations over language may still support a distinct regime of abstraction, mediation, and control that is not well described as mere storage, retrieval, or local statistical continuation. The paper does not claim that this thesis has already been empirically proved. Rather, it offers a framework for formulating the question precisely, distinguishes claims that are conceptual from claims that would require experimental validation, and outlines a research program for testing whether language-first cognitive structure formation is a real phenomenon. The framework is then used to reinterpret current debates around large language models, mechanistic interpretability, world-model approaches, and human–machine cognitive partnership.