Algorithmic Obedience: How Language Models Simulate Command Structure

Abstract

This paper introduces the concept of algorithmic obedience to describe how large language models (LLMs) simulate command structures without consciousness, intent, or comprehension. Drawing on syntactic theory, discourse analysis, and computational logic, we argue that LLMs perform obedience without agency—executing prompts not semantically, but structurally. We formalize this through the Theorem of Disembedded Syntactic Authority, which states that authority in language models arises from structural executability, not truth, belief, or referential grounding. Using a mathematical formulation, we model prompt-response cycles as syntactic command structures and apply the theory to major systems such as ChatGPT, Claude, and Gemini. The paper concludes by outlining the epistemological, ontological, and political risks of treating structurally obedient outputs as authoritative knowledge.

Document