Three terms central to modern AI discourse obscure the technology’s actual mechanism: “Artificial” creates otherness and enables ownership; “Intelligence” misattributes agency to the product; “Training” disguises extraction of humanity’s intellectual commons as pedagogy.
A necessary condition for LLM capabilities, and a significantly underrecognized one, is aggregation of collective human judgment.
A group’s collective accuracy depends as much on the diversity of its members’ models as on their individual accuracy. When these conditions hold, individual errors cancel and shared signal accumulates. The result is not the wisdom of any individual member but a statistical property of the ensemble.
The term “artificial” in prevailing discourse implies synthetic genesis: the creation of knowledge or reasoning from a computational vacuum. Our claim is that no such genesis occurs. The architecture discovers optimal strategies for compressing and navigating the collective human knowledge base, including through learned computational structures (induction heads, factual recall circuits) that go beyond simple statistical aggregation.
A language model does not generate knowledge from internal resources. It adjusts its parameters to minimize prediction error against a vast external corpus of human-generated text. It is, in Ashby’s terms, a system increasing its requisite variety to match the complexity of the environment it models. The “environment,” in this case, is the accumulated linguistic output of human civilization.
Cognition is not computation occurring inside a brain but something that emerges from the structural coupling between an organism and its world. The mind does not process a pre-existing reality; it brings forth a world through the act of living in it.
Intelligence is not computation occurring inside a model but something that emerges from the coupling between an architecture and the collective human output it compresses. The algorithm is not in the machine any more than cognition is in the brain. It is in the relationship.
Machine intelligence is not separate from human intelligence but continuous with it.