AGI Needs Hunger: Metabolic Grounding as a Constraint on Artificial Agency
Abstract
Current AI systems exhibit impressive pattern recognition, reasoning, and planning, yet remain fundamentally unmotivated: they optimize externally specified objectives without intrinsic pressure to persist, adapt, or care. We argue that this limitation reflects a missing constraint shared by all biological intelligence: metabolic scarcity. In living systems, cognition is inseparable from energy regulation. Biological agents must continuously act to maintain viability. Building on biological computationalism, which views cognition as embodied computation realized through physical self-maintenance, we advance the position that artificial general intelligence will require an analogous internal energetic imperative ("artificial hunger'') to generate real agency, self-directed learning, and consciousness. We propose a framework for metabolically grounded computation, outline architectural guidelines, and derive falsifiable predictions for future AI systems.