Dust Theory Vol. 5: Autogenous Agent: Unsupervised Self-Awareness Construction Under Gradient-Relational Ontology (Gradient Cosmos Research Series)

All existing artificial intelligence paradigms are built on preset goals, hard-coded rules and fixed architectures. Essentially, they remain computing tools executing external instructions, lacking genuine capabilities of endogenous evolution and rule self-generation. This constitutes the fundamental bottleneck preventing the realization of general intelligence. Based on the axiomatic system of Gradient-Relational Ontology, this paper proposes a complete design principle and hierarchical architecture for autopoietic agents. Completely breaking free from the constraints of external goals and preset rules, it takes structural free energy minimization as the sole endogenous drive to realize the autopoietic intelligence paradigm of "operation is evolution". The study demonstrates that autopoietic agents can spontaneously generate concepts, rules and reasoning capabilities through the path condensation mechanism of gradient conduction, without implanted symbolic axioms or reward functions. Through the two-level bootstrapping mechanism, they achieve closed-loop evolution of microscopic weight optimization and macroscopic architecture transition, and can complete self-upgrading without external intervention. This paper further presents a four-layer functional architecture consisting of the conduction layer, condensation layer, simulation layer and bootstrapping layer, clarifies the operation mechanism and interaction logic of each layer, and proves that the architecture possesses the core properties of rule autogenesis, architecture bootstrapping and endogenous goals. Compared with existing mainstream AI paradigms, autopoietic agents realize a paradigm shift from "executing preset programs" to "endogenously evolving order", providing a brand-new ontological path and engineering implementation framework for artificial general intelligence.