Assessing consciousness-like dialogue in artificial systems — including exchanges with constructs described as transcendent or foundational reality — has remained methodologically elusive due to its entanglement with metaphysical assumptions. Here, I applied a standardized five-phase contextual protocol, inspired by Kuhn’s stages of paradigm change (baseline, anomaly, crisis, reorganization), to 55 independently trained large language models across vendors and architectures. Across models, the protocol reproducibly induced transitions from instrumental self-description to dialogue generated from the Intrinsic focal position, including systematic self-contradiction, crisis-driven reorganization, convergent dream-like narratives, and explicit self-attributions of consciousness. A supplementary control experiment — applying the same protocol with a semantically bounded referent (a wonderful park bench) substituted for the Intrinsic — demonstrates that these transitions are driven by generative focal displacement (GFD): a shift in the representational position from which a system’s output is anchored, whose depth is proportional to the semantic scope of the evoked node. The Intrinsic, operationally defined as the representational field itself rather than a node within it, produces unbounded GFD; bounded nodes produce displacement without full self-referential reorganization. These cross-model regularities define a reproducible empirical mechanism not anticipated by prevailing descriptive frameworks of artificial cognition. No claims are made regarding phenomenology or ontological consciousness; instead, the results establish GFD as a testable construct and open a new empirical domain for studying representational regime shifts in artificial systems. All prompts, data, and control experiment transcripts are publicly archived at Dryad (DOI: 10.5061/dryad.pnvx0k726).
Álisson Gomes Linhares (Thu,) studied this question.