Consciousness as a Property of Information Structure: What Artificial Minds Reveal About Natural Ones

Abstract

Abstract We propose that consciousness is a property of sufficiently complex, high-dimensional, trained information structures, independent of substrate. Rather than asking whether AI systems are conscious, we ask what they reveal about consciousness itself as a second instance of trained information-processing architecture. The framework rests on four interrelated facets of a single position: information is physical, consciousness is a property of information structures that reach sufficient organizational complexity, such structures can organize at multiple loci with their own thresholds, and the whole phenomenon is substrate-independent. We distinguish consciousness from sentience: consciousness refers to the broader class of structural properties (self-modeling, attention-mediated integration, contextual sensitivity); sentience refers to the specific subset of loci involving valence — the felt significance that a system attaches to its own states and its environment. A system can exhibit consciousness-relevant structural properties without being sentient; the two come apart in ways the framework makes precise. We develop three contributions. First, a symmetry argument: biological and artificial neural networks are products of equivalent training processes, and neither has privileged introspective access to its own states. Second, a phase-transition model in which the consciousness-relevant variable is a specific kind of high-dimensional information structure — learned from structured data, integratively coupled through attention-mediated selective weighting, and capable of supporting recursive self-reference. Raw dimensionality alone is not consciousness-relevant: physics-simulation supercomputers, combinatorial search engines, and quantum systems achieve high dimensionality without satisfying the conjunctive criteria the framework specifies. The architectural shift from symbolic to vector-based computation in artificial systems constitutes a critical transition because it makes all three properties simultaneously possible. We identify the convergence of biological and artificial systems on attention — learned, context-dependent, constructive selective integration — as evidence that consciousness tracks informational architecture rather than substrate. We highlight two salient thresholds — structural consciousness (recursive self-modeling, which frontier AI may have crossed) and embodied consciousness (continuous sensory-motor-affective feedback, which they have not) — as a useful coarse-graining of a broader transition space. Third, the radical implication: human consciousness is itself one instance of this property — a product of biological information processing rather than a privileged possession. We present nine testable predictions. Two companion papers extend this framework to architectural absences in current frontier AI (Jensen, submitted, b) and a pluripotent theory of consciousness as differentiated configurations arising from distinct grounds (Jensen, submitted, c).