How Can Self-Consciousness Emerge in Embodied AGI? From Bodily Boundaries and Self-Regarding Drives to the "Pure Thinking Self"
Guoping Hua
Zenodo (CERN European Organization for Nuclear Research) October 28, 2026 preprint DOI: 10.5281/zenodo.23716403 (opens in new tab)
Study at a glance
AI-extracted from the abstract| Characteristics | Theoretical or philosophical paper |
|---|---|
| Key points | Proposes a three-stage generative architecture of functional self-consciousness for embodied AGI, moving from a somatic self-preservation layer through a value-driven layer to a rational reflection layer, and advances five diagnostic criteria: sense of body ownership, endogenous preference, causal constraints on self-reference, autobiographical continuity, and metacognitive calibration. Argues this reframes AI self-consciousness as a designable, intervenable, and falsifiable research program without claiming to solve the 'hard problem' of phenomenal consciousness. |
Abstract
The most vexing gap in current research on AI consciousness is not merely that the question of whether machines possess subjective experience remains unsettled; more fundamentally, the "self" is often treated as an already-existing premise. Theories can explain how information is broadcast, integrated, or represented at a higher order, yet they rarely ask how a system gets from "having a body" to "this is my body," and then to "I know that I am thinking." This article places that generative problem within the framework of embodied AGI. Adopting a constructive method of "structural constraints—functional decomposition—engineering mapping—counterfactual testing," it proposes a three-stage generative architecture of functional self-consciousness: a somatic self-preservation layer that forms a stable self–world boundary and a sense of body ownership; a value-driven layer that sedimentates interoceptive evaluation into "self-regarding drives" oriented toward the system's own continued existence; and a rational reflection layer that enables thinking to take its own process again as an object. The article further reformulates the "Personality Compass" as a computable multi-drive structure and proposes a dual-mode mechanism of "low-load continuous monitoring and event-triggered focusing," explaining how the "pure thinking self" is possible at the functional level. On this basis, five diagnostic criteria are advanced—sense of body ownership, endogenous preference, causal constraints on self-reference, autobiographical continuity, and metacognitive calibration—and a "speaker mechanism" is used to discuss the structural tension among self-interested drives, public norms, and corrigibility. The article does not claim to solve the "hard problem" of phenomenal consciousness, but seeks to advance AI self-consciousness from an abstract ontological judgment to a designable, intervenable, and falsifiable research program.