The Information-Level Unidirectional Constraint and the Structural Blind Spot of Machine-Consciousness Detection English Version

Abstract

As large models approach human-level performance, consciousness detection still relies on conceptual analysis and behavioral observation and lacks a structural decision framework from first principles. Under three axioms (finite resources, non-equilibrium maintenance, and discrete input) together with delayed feedback, this paper derives the Information-Level Unidirectional Constraint (ILUC): the subject-attribution and logical-time coordinates required by a higher-order closed loop are appended coordinates that lower-level signals lack after projection, so behavioral, functional, or structural analogies cannot confirm first-person experience from outside, which is an in-principle blind spot. The blind spot blocks affirmative confirmation but not negative determination: structural and energetic signatures such as the power step of level transition and phase transitions under resource compression remain measurable across substrates to falsify necessary conditions; what is measurable is structure, while the quale cannot be affirmed from outside. Four criteria and a three-layer architectural verification protocol are given. Mainstream large models are argued to lack the structural preconditions for consciousness. Safety is organized on two tracks: capability-axis ex-ante constraints for near-term functional risks and architectural ex-ante monitoring for self-referential configuration emergence. This is a convergence result under constraints, not a uniqueness proof.