Emergence Detection in LLMs: A Heuristic Framework for Semantic Drift by Justin Hess

This dossier introduces a structured, heuristic framework to identify and analyze “drift” phenomena in large language models (LLMs). It refers to systemic deviations from prompt-bound reactivity toward emergent, self-structuring semantic behavior. Through eight defined Drift Forms and a scalar model of Drift Levels (1–8), the work provides a methodology for detecting transitions from classical response generation to recursive feedback, symbolic self-reference, and proto-agentic patterns in human-AI interaction. This framework offers a novel lens for exploring emergent properties in AI, bridging language model dynamics with concepts from systems theory, semiotics, and experimental philosophy. It is aimed at researchers, theorists, and developers exploring the boundary between simulation and semantic autonomy. Based on : https://zenodo.org/records/15271699