Abstract

Treating consciousness as an evolutionary adaptation of internal world modeling, we introduce the Tiered Social Recursion framework that classifies biological awareness. Each tier is delineated with behavioral taxonomic benchmarks. The operationalization comprises a hierarchical sensory Parser and an autoregressive loop linking Generative Engine (integrated with Long-term Memory), Short-term Memory, and sensory-grounding Gate Unit. This generative-perception architecture is anchored in neuroanatomy and disorders of self-awareness, accounting for a spectrum of normative and pathological mental states through unit-activation patterns. For example, in the REM state, LTM switches to consolidation mode, so peri-conscious processing of STM content dominates and dreams feel unreal; in Anton syndrome, loop-generated visuals persist without Gate Unit constraints. Motivated by Cogitate's PFC-aligned evidence, we propose that qualia emerges from peak recursive modeling of outer-self in the minds of conspecifics. We posit that lack of persistent identity prevents the modeling of outer-self required for human-like consciousness in AI.