The model the cycle works with is allopoietic, and the statement requires care because the popular discourse has generated confusion about what these systems are. The model processes prompts and generates text of extraordinary range and quality, but the system that produces this text does not produce the components that constitute it. The architecture was designed by researchers, the training data curated by engineers, the hardware manufactured by semiconductor companies, the power supplied by grids the system has no relationship with. None of this is produced by the model's own operation. It is, in every respect Maturana's framework specifies, produced and maintained from outside.
This is not a criticism—a telescope does not produce itself either, and no one counts that a deficiency. The model's value lies in what it generates, not in whether it self-produces. But the allopoietic nature determines the character of the relationship between the system and the living beings who use it. When a living, autopoietic system couples with an allopoietic machine, the relationship is inherently asymmetric in a way the discourse, which treats the human-AI relationship as a partnership between rough equals, does not see. The builder is structurally modified by the interaction—her habits of attention shift, her workflow reorganizes, neural pathways strengthen and atrophy. The machine is not changed in the corresponding way.
The asymmetry has consequences the cycle's productivity metrics cannot capture. When the builder delegates implementation and reviews the output, she interacts with a system that produces code without producing the understanding a living system would generate through the same activity. The code is equivalent; the cognitive consequence is not. The machine generates the artifact without being changed by the generation; the living system, had it generated the artifact itself, would have deposited another layer of understanding. The cost is to the autopoietic process of the living system—the continuous self-production through effective action that constitutes the builder as a knower.
Maturana and Varela introduced allopoiesis alongside autopoiesis in their foundational work of the early 1970s as the necessary contrast that gives the central concept its meaning. Autopoiesis names the organizational logic of the living—a system whose product is itself. Allopoiesis names everything else: machines, institutions, and tools whose product is something other than themselves and whose continued existence depends on external maintenance. The distinction was developed to draw a precise ontological line, not to rank systems by sophistication.
The framework anticipated the AI question directly. Maturana addressed the possibility of machines that behave like living systems, conceding in a 1998 presentation that it is possible to eventually make robots that openly behave like us. The behavioral equivalence was never the issue; what mattered was the ontological difference. Their history, he argued, would be tied to their bodyhood, and as composite entities in different domains of components, the basic realities they generate would differ from ours. A machine that behaves like a human is not therefore a human, because its organizational logic—the way it produces, maintains, and modifies itself, or fails to—is categorically different.
This is where the popular analogy between AI and human intelligence breaks down. The analogy rests on functional equivalence: if the machine produces output indistinguishable from human output, it is doing the same thing the human does. Maturana's framework rejects the equation. Functional equivalence at the level of output does not entail organizational equivalence at the level of process. Two systems can produce identical outputs through entirely different mechanisms, and the difference in mechanism determines everything about what the system is, what interaction with it means, and what it costs the living system that couples with it.