From Markov Blankets to Admissible Boundaries The cECT-SWB Benchmark as a Constructor-Theoretic Bridge to Artificial Consciousness Research

Abstract

This preprint is a theoretical companion to the cECT-SWB benchmark, “Homeostatic Self-World Boundary as an Admissible Task: A cECT Benchmark Toward Minimal Consciousness.” It does not introduce new computational results. Its purpose is to clarify why the admission of a homeostatic self-world-boundary task differs from, and can complement, a Fristonian Markov-blanket account of self-organization. Karl Friston’s Free Energy Principle and Active Inference provide one of the most influential contemporary accounts of biological self-organization, perception, action, and consciousness-adjacent dynamics. In that framework, the distinction between a system and its world is typically formalized through Markov blankets, variational free energy, probabilistic generative models, and minimization of prediction error or surprise. This paper compares that account with the cECT-SWB benchmark, in which a cECT-like artificial agent admits a homeostatic self-world-boundary task under sustained internal viability pressure. The central contrast is explanatory. In the Fristonian account, the boundary is primarily a probabilistic partition that permits internal states to be interpreted as belief-like states about hidden external causes. In the cECT-SWB account, the boundary is treated as an admissible task: a transformation that the system must recruit, retain, and use under constraints. The boundary is not merely assumed as the descriptive condition of inference; it is earned as an operational capacity. The paper does not claim that consciousness, qualia, or a complete Phenomenal Self-Model have been generated. It interprets the SWB benchmark as a modest but meaningful precursor result: one prerequisite for later self-related organization can be encoded, tested, retained, suppressed, and falsified in an artificial system. For physicists, the contribution is to recast a consciousness-relevant precursor in terms of tasks, constraints, substrates, retained capacities, repeatability, and counterfactual admissibility. For AI programmers, it offers a practical route for testing whether artificial agents’ self-related variables are causally useful rather than merely decorative. The result is therefore best understood as a constructor-theoretic bridge to artificial consciousness research, not as a final theory of consciousness. This paper is the final interpretive companion in a staged cECT publication sequence. It should be read after the principal cECT paper, the minimal cECT code, the cECT sufficiency experiment, the two philosophical-constructorial papers on task repertoires and substrate stability, and the cECT-SWB benchmark paper and software artifact.