UMOSU/UMOKWAI — Why Expert Systems Are the Only Ones That Give Verified Answers, and the LLM Is Only the Interface

UMOSU/UMOKWAI — Why Expert Systems Are the Only Ones That Give Verified Answers, and the LLM Is Only the InterfaceThis document summarises the entire path that led the UMOSU (ontological) and UMOKWAI (epistemic-operational) models to become, without this being the declared objective at the outset, the theoretical explanation of a practice that already exists, everywhere, without anyone ever having formulated it explicitly: anyone who works seriously with verifiable data trusts a certified calculator, a validated database, an approved diagnostic system, the text of a published norm — never a generic language model taken on its own. The document answers the question “why must it be so, always, by construction, regardless of how large or well-trained an LLM becomes?”.The central thesis: verifiability (Rc) is a property of the phenomenon being discussed, not of who discusses it; only a system with a direct link — causal, formal or documentary — to the verification mechanism of the phenomenon can produce a verified answer; an LLM, by how it is built (statistical prediction on text), never has that link, by structural and not contingent limitation. It follows that an LLM is not, and cannot become, the system that verifies: it can be the interface that carries the question to the right system and returns its answer in natural language — a function that the document shows is nevertheless a genuine novelty compared with the classical expert-systems approach, not a mere limitation. For the case in which such an expert system does not exist by definition (doctrines, fC2), the legitimate residual task of the LLM is not to answer, but to list existing positions correctly, with verified attribution.The document retraces the multi-AI genesis of the model (born from dialogues in the field of law, developed through Claude, Grok, ChatGPT, Kimi and Gemini), the eight main discoveries that emerged from the phenomenon-by-phenomenon analysis, and the operational formalisation in a three-Router chain (classification, supervision, presentation) — also born from comparison with the AI Regulation of the Municipality of Treviso, which already treats AI outputs as technical drafts to be submitted to human control.Keywords: UMOSU, UMOKWAI, expert systems, artificial intelligence, operational epistemology, hallucination prevention, ontological Router, Meta-Router, Presentation Router, falsifiability, verifiability.