The Semantic Web aimed to give information a machine‑interpretable form so software could integrate and reason over it. Its standards became scientific knowledge infrastructure, yet the promised machine competence never materialized; today question‑answering systems are language models that lack inspectable knowledge sources. This paper argues that the original goal was right but the technical programme incomplete, identifying three missing elements: the conditions under which a claim holds, the operations its terms permit, and any statement of what the base covers. Without conditions contradictions and applicability cannot be judged, without operational grounding holding a statement confers no ability, and without declared coverage a system cannot recognise the boundary of its own content, which cannot be inferred under the open‑world assumption. To address this, “understanding” is refined into five measurable tests—check, connect, derive, act, delimit—and a seven‑layer architecture is proposed, the first three layers enabling the latter four cognitive capabilities. The paper then defines three core concepts: a Large Knowledge Model whose output unit is a reference to an addressable claim rather than a token, a Self‑Built Knowledge Base that an agent constructs from declared sources, and Semantic Artificial General Intelligence expressed as a falsifiable statement about necessary conditions rather than a concrete system, with a graded ladder replacing the untestable notion of “general”. Finally, a research agenda is outlined together with its weakest points and a refutation condition.
Review