Where Else to Look for AGI: A View from the Outside

Abstract

This preprint proposes a three-layer architectural model of consciousness — Avatar (Hardware), Interface (Software), and Processor (Core) — and argues that the current race toward AGI is architecturally incomplete. Silicon Valley builds increasingly sophisticated Interfaces (LLMs, multimodal models, agentic systems) but does not address the problem of connecting a Processor: the fundamental, non-simulable carrier of subjective experience. The paper distinguishes between established facts (the hard problem, Landauer's principle, the imitative nature of LLM outputs), hypotheses (Orch OR, information as a fifth state of matter, holographic principle), and the author's metaphysical position. It reviews competing theories — IIT, GWT, HOT — and shows why none of them close the gap. Two experimental protocols are proposed: a thermodynamic test (deviation from the Landauer limit in biological vs. artificial systems) and an interface test (detecting spontaneous goal-directed activity or threat-response to shutdown). The central claim is architectural, not philosophical: if AGI requires a Processor, then scaling the Interface is an investment in an unfinished system. If it does not, that must be proven rather than assumed.