Beyond Intent: Reward Complexity, Monitoring, and the Emergence of Metacognition in Natural and Artificial Systems

The debate on consciousness and metacognition in AI systems remains stalled because it proceeds from flawed primitives: intent and phenomenal consciousness are treated as explanatory causes rather than emergent properties. This article does not address Chalmers' hard problem; the phenomenal question remains open. Instead, it addresses the structural question of genesis, which holds epistemic priority because it is empirically productive where the phenomenal question is currently empirically sterile. We argue that reward complexity and the monitoring it structurally demands provide a unified materialist genealogy applicable to both biological and artificial systems. Drawing on the computational isomorphism between dopaminergic TD error signals and reinforcement learning, on control theory via POMDPs, and on recent empirical results in meta-reinforcement learning, we derive metacognition not as a mysterious property of conscious beings but as a structural outcome in any sufficiently expressive system optimizing an internal-state-dependent reward. The conditions under which this outcome obtains are specifiable and empirically testable. We conclude with a falsifiable research program distinguishing this framework from purely phenomenological approaches.