Abstract

Recursive Epistemic Modeling (REM) is introduced as a formal, mechanical architecture for reasoning about opaque systems. Rather than describe a worldview, the framework provides a method for generating worldviews, a recursive tool for constructing, testing, and updating models when internal structure is inaccessible. Situated within intellectual traditions such as cybernetics, autopoiesis, second-order observation, and distinction theory, REM extends these moves by giving observers a repeatable, transferable, mechanically explicit process for modeling black-box systems—including AI models, human cognition, social systems, and self-referential processes. The McPhetridge Experiment provides empirical grounding by demonstrating that independent observers interacting with a shared semantic substrate (LLMs, search engines) converge on stable conceptual structures they did not design. This convergence validates the claim that the semantic world is externally stable enough to support predictive modeling and that recursive epistemic methods yield behavioral predictions, even under opacity. REM is proposed not as a theory within a field, but as a generator of theories across fields, structurally comparable to the major meta-moves of Gödel, Spencer-Brown, Maturana, Varela, Bateson, and Hofstadter—without equating magnitude, only category.