Large Human Network Models (LHNM): The Human Intelligence Bridge. A New Model Class Between Artificial Intelligence and Human Networks

Large language models answer questions whose answer is in texts. A substantial share of the economically decisive questions, however, has no answer in texts. The answer lies with people: in what they can do, own, want and decide, and in what an experienced expert knows about them without ever having written it down. This white paper describes a model class built for this knowledge. A Large Human Network Model (LHNM) connects three things that have so far existed separately: the unpublished domain, access and behavioral knowledge of an expert, a team or a house, a leading expert in their field; the relationship network these people have built over decades; and the reasoning capability of the best large language models. The result is not a language model and is not meant to replace one. It is a layer above it that learns from a different source than any language model: not from what humanity has written, but from what a person or a team knows about their world and whom they can reach in it. The paper defines the class; distinguishes it from Vertical AI, internal enterprise AI, relationship intelligence software, expert networks, expert clones and AI matchmaking services; describes the data basis, architecture, construction and operation of a model that emerged from a live case in the world of family offices and is being developed further there; and shows why the pattern transfers to any domain in which decisive knowledge lies with a few people. What is described here could not be built until a few years ago: the knowledge and the relationship network of a person or a house in one place, connected, retrievable and able to learn. Such a model can be built in two directions: outward, as access to knowledge and people that can be monetized with third parties, and inward, as the memory of a group of experts, a family or a company that would otherwise be lost with the people who hold it. The counter-thesis and its various facets are examined at the end; the paper closes with testable hypotheses. To the author's knowledge, it is the first description of this model class under this name. German edition included as second file.