The Cybernetic Turn: Generative AI as a Dynamical System of Meaning

Abstract

Generative artificial intelligence is usually explained in the vocabulary of computer science: architectures, parameters, objectives, scale. That vocabulary describes the machine accurately but leaves its most consequential property unexplained: why a system directed by short, local, context-dependent interventions can assemble artifacts of essentially unbounded complexity across every domain of meaning. This paper argues that the missing explanation is cybernetic rather than algorithmic. It rests on a principle long used in modeling and in science education, here promoted from pedagogical observation to ontological claim: the global description of a dynamic process is typically far more complex than its local description, often admitting no closed form at all, while the local description is the one closer to human understanding and the only one that can be constructed. On that basis, the paper treats generative AI as a dynamical system of a new kind, whose state space is a learned space of meaning and whose local law is learned and indexical, and it argues that the system’s effectiveness is the constructability advantage of the local form, extended for the first time from number and symbol to meaning. Four elements carry the account: the embedding space as state space, the token as microscopic law, indexicality as the semantics of the local step, and navigation as the human mode of control. Ashby’s law of requisite variety is relocated, not violated. The comparative thesis is that cybernetics, which stands between mathematics and computer science, explains generative AI better than the algorithmic frame, because it captures the mode of the system — modeling, not implementation — and not only its mechanism. The account is conceptual and programmatic: it relocates the question of what generative AI is, and makes a class of phenomena describable, measurable, and open to refutation.