42-True: A Large Meaning Model for Declared Human Intent

Abstract—Contemporary artificial intelligence is trained, with rare exception, on observational data: text and behaviour recorded under conditions where the subject was, or assumed they were, watched. Nine decades of behavioural science establish that observation distorts behaviour; it follows that models trained on observational corpora converge toward an upper bound on fidelity to the unobserved human. This paper proposes a complementary class of training data — declared intent paired with verified outcome — gathered in an environment architecturally incapable of identifying the declarant. We define its atomic unit, the intent pair, and the model trained upon it, the Large Meaning Model (LMM). We situate the proposal in two prior empirical studies we contributed to — on privacy-preserving advertising and on natural-language question answering over dense legal corpora — whose findings on declared-versus-inferred data, semantic retrieval, and grounding against hallucination directly inform the design. The corpus the LMM requires does not exist and cannot be scraped, simulated, or extracted; it must be produced by a purpose-built network. We describe that network, its classification and privacy architecture, and a proposed foundation to steward the resulting model as a commons.