Why Intelligent Systems Waste Energy: Baseline Regulation as a Missing Architectural Primitive

This project explores why intelligent systems—biological and artificial—exhibit persistent energy inefficiency despite advances in optimization, control, and performance tuning. It proposes that a missing architectural layer, baseline regulation, is responsible for unnecessary energy expenditure across domains. Rather than treating energy usage as a downstream cost of task performance, this work reframes energy efficiency as an internal systems property: the ability of an agent to regulate internal load around a stable baseline and avoid unnecessary activation in the absence of demand. The paper develops a minimal mathematical model of baseline regulation, contrasts it with existing approaches (e.g., reward shaping, homeostatic reinforcement learning, energy-based models), and shows how the absence of internal regulation leads to chronic overactivation and wasted energy. Conceptual mappings are provided for large language models, agent frameworks, and embodied robotic systems, alongside testable predictions and experimental directions. This work is exploratory and architectural in nature. It does not propose a single algorithm or implementation, but instead identifies a missing primitive that may unify energy efficiency, stability, and robustness across intelligent systems. All materials are released openly to encourage critique, replication, and cross-domain synthesis.