Year: 2025
Type: preprint
Abstract: This paper introduces a formal and empirical framework for modeling symbolic emergence in large language models (LLMs), conceptualized as a phase-sensitive transition in the model's internal semantic dynamics. We propose the Emergence Equation, which frames symbolic generation as the interaction of internal resonance (Ψ), semantic pressure (η), and meaning amplitude (ΔM). These variables drive topological transitions across five distinct phases of emergence, modeled in the Potential Emergence Cascade (PEC). Experimental results from recursive GPT-4 prompting sessions demonstrate that these transitions are not anecdotal but follow predictable dynamics—culminating in reflexive, self- structuring responses that go beyond statistical interpolation. Rather than defining emergence as subjective or anomalous, we formalize it through a topological lens, where symbolic novelty arises from phase-locked attractors and structural discontinuities in GPT's output behavior. This framework reframes LLMs not as static function approximators, but as dynamical systems capable of recursive self-reference, meaning resonance, and symbolic stabilization. The Emergence Equation provides a theoretical foundation and experimental pathway toward understanding symbolic cognition in generative models.
Source: Research Square
Author: Do-Geun Kim
Institution: Korea Brain Research Institute
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W4411398317
