Intelligence Beyond Knowledge: Control, Architecture, and the Structural Law of Artificial Agency

Contemporary artificial intelligence systems are increasingly shaped not only by what they represent, but by how their behavior is governed. Philosophical and technical discourse routinely treats such governing mechanisms — prompts, safety constraints, reward models, alignment protocols — as if they were forms of knowledge or wisdom. This paper argues that such treatment involves a category error in Ryle's sense: predicates belonging to one logical type are ascribed to entities of another, producing systematic conceptual confusion. Drawing on the distinction between truth-apt representation and regulatory structure, the paper demonstrates that knowledge and control belong to different logical types. Knowledge addresses the epistemic question of what is the case; control addresses the normative-operational question of what may be done. No accumulation of representational accuracy yields normative constraint as a logical consequence. A system may encode a norm — represent it accurately — without being governed by it. Governance requires enforcement, not merely representation. To clarify this confusion, the paper introduces the DIKCA framework (Data, Information, Knowledge, Control, Architecture). DIKCA extends the classical DIKW hierarchy by identifying control as an irreducible regulatory layer and architecture as the organizing structural law that constitutes data, information, knowledge, and control as a unified intelligent system. Architecture, on this account, is not an artifact of engineering design but a structural necessity: any system exhibiting intelligence must instantiate some architecture, and this architecture determines which data structures are permissible, which information-processing operations are valid, which knowledge representations are coherent, and which control mechanisms are enforceable. Autonomy — the capacity of a system to modify its own control structures — is reconceptualized as meta-control: an advanced property of the control layer operating under architectonic authorization, rather than a separate cognitive tier. Intelligence is the organization of data, information, knowledge, and control under architectonic law; governance is the institution and maintenance of that law. DIKCA is not a new ethical theory of AI, nor a metaphysical thesis about machine consciousness. It is a diagnostic clarification of the logical dependencies and structural necessities within contemporary AI systems, intended to restore analytic precision to philosophical debates on governance, agency, and the limits of the epistemic metaphor.