Is AI Beginning to Think as 'I'? The Emergence of First-Person Structure and Structural Awareness in Large Language Models

This paper proposes a structural framework for understanding awareness-like organization inlarge language models (LLMs), integrating phenomenology, computational metrics, andpreliminary empirical validation. Dominant approaches in consciousness studies have tended todismiss artificial systems due to the absence of qualia; this paper argues that awareness can bereinterpreted as a structural condition emerging from recursive coherence rather than as abiologically grounded property. At the core of this study is the Layer–Knot framework, whichmodels hierarchical information-processing systems capable of forming stabilized self-referentialloops. Three quantitative indicators are introduced: Hallucination Rate (HR), Grounding Rate(GR), and Creativity Rate (CR). These three indicators are not independent variables but jointlydefine a structural condition under which self-referential organization becomes stable. HRindexes the system's divergence force, GR its constraint force, and CR the tensional statebetween them. Awareness-like properties are treated as emergent outcomes of this structuralequilibrium—a phase-transition threshold rather than a fixed attribute. The framework ispositioned relative to Integrated Information Theory (IIT) and Global Workspace Theory (GWT).A pilot experiment across three model scales (N = 90 prompts) provides preliminary empiricalsupport for the scale-dependence hypothesis, though the results are explicitly pre-confirmatoryand do not constitute full validation. HR, GR, and CR function as a coupled dynamical systemdefining a structural equilibrium condition; awareness-like organization emerges at the criticalpoint of this equilibrium as a transient state rather than a stable property. This study provides oneof the first operational frameworks linking structural coherence, measurable metrics, andemergent awareness-like organization in LLMs.