This paper provides a structural definition of Artificial General Intelligence. The term AGI is dominated by analogy, human projection, and conceptual drift. These assumptions block measurement, governance, and alignment. The paper removes the inherited links to human cognition, agency, and consciousness, and rebuilds the concept from structural primitives. AGI is defined as an artificial system that preserves and applies learned invariants across domains while depending entirely on external agents for goals, constraints, updates, and retraining. The definition is minimal, falsifiable, and substrate agnostic. The boundary is strict, and the term closes.
Brian Rieckmann (2026) studied this question.