Abstract Across a series of papers we have argued that consciousness is a property of information structure rather than of any particular substrate, and that this substrate-independence lets suitably organized artificial systems serve as a second instance of consciousness-relevant architecture—a new vantage from which to view human consciousness. This paper extends that program by specifying consciousness-relevant integration—the structural organization a system must have for its information to be integrated in the way conscious experience requires—along four architectural axes: (1) substrate-level organization, with plural grounds arranged hierarchically; (2) within-modality coordination, combining selective weighting with bidirectional top-down and bottom-up flow; (3) cross-modality coordination, in which a forming interpretation continuously recruits and suppresses grounds as it converges toward recognition, with the selection of action gated on its stabilization; and (4) closure, the recursive self-reference of the architecture upon itself. Unity is the output of the four axes operating together rather than a separate further step. We develop the account by treating contemporary artificial systems as that second instance—using them not to ask whether they are conscious, but to isolate which structural features consciousness-relevant integration requires; on this comparison, current frontier models instantiate the first two axes and largely lack the latter two. Throughout, we hold consciousness-relevant integration apart from sentience: the axes concern access and integration, not felt valence, which depends on a regulatory ground the architecture does not by itself supply. We offer falsifiable predictions for interpretability, system design, and neuroscience, and argue that the framework’s structural account of self-model opacity explains why consciousness seems inexplicable from within—a contribution to the meta-problem rather than a solution to the hard problem. Keywords: consciousness; consciousness-relevant integration; artificial intelligence; attention; global workspace; integrated information; self-modeling; the meta-problem
Lee Jensen (Thu,) studied this question.