The Intelligence That Was Never Artificial: LLMs as Collective Human Cognition and the Cybernetics That Predicted Them

In 1907, Francis Galton demonstrated that the median estimate of 787 people guessing an ox's weight was accurate to within 0.8 percent, outperforming any individual expert. We argue that a structurally analogous mechanism is a primary, systematically underrecognized source of Large Language Model capabilities: next-token prediction aggregates billions of human judgments about language, reasoning, and meaning through a sophisticated compression function, producing capabilities no individual contributor possessed.

We trace the historical erasure of this insight. Wiener's cybernetics (1948) described intelligence as emergent from feedback systems. The Dartmouth conference (1956) reframed this relational phenomenon as 'Artificial Intelligence,' severing the intelligence from its collective human source. Three terms central to modern AI discourse obscure the actual mechanism: 'Artificial' creates otherness and enables ownership; 'Intelligence' misattributes agency to the product; 'Training' disguises extraction of humanity's intellectual commons as pedagogy.

Using Surowiecki's four conditions for crowd wisdom, we derive testable predictions: LLM capability should degrade when these conditions are violated through monoculture corpora, RLHF narrowing, or synthetic data loops. Evidence from model collapse research, the alignment tax literature, and recent replications of the Galton effect with LLM ensembles provides empirical support. We conclude that what is called 'AI' is collective human intelligence mediated by silicon, and that the naming determines who benefits.