Constrained Object Hierarchies as a Unified Framework for Artificial General Intelligence

Constrained Object Hierarchies (COH) presents a neuroscience-grounded theoretical framework for modeling artificial general intelligence (AGI) systems across diverse domains. This paper introduces COH as a 9-tuple formalism that integrates compositional structure, adaptive neural components, and multi-type constraints into a unified representation as a unified framework for AGI. We present GISMOL (General Intelligent System Modelling Language) as a practical implementation of COH, providing a toolkit for developing constraint-aware intelligent systems. Through six case studies across healthcare, manufacturing, social science, finance, biology, and astronomy, we demonstrate COH’s ability to model complex adaptive systems with explicit constraint management. The framework addresses the “jagged intelligence” problem by ensuring consistent constraint enforcement across abstraction levels, while providing mechanisms for universal world modeling and agentic system implementation. Our results show that COH/GISMOL enables the development of explainable, safety-critical AGI systems with verifiable constraint satisfaction across multiple domains simultaneously.