Autopoiesis
A system's capacity to continuously produce and maintain itself through internal processes.
Autopoiesis, from the Greek for 'self-creation,' describes the property of a system that continuously regenerates and maintains itself by producing its own components through internal operations. Originally coined by Chilean biologists Humberto Maturana and Francisco Varela in 1972 to characterize living cells, the concept captures how a system defines and preserves its own boundary while remaining operationally closed — meaning its internal organization is self-referential rather than directed by external forces. The system interacts with its environment but is not fundamentally defined or controlled by it.
In cognitive science and AI, autopoiesis has served as a theoretical lens for understanding how minds and intelligent systems might be organized. Researchers influenced by Maturana and Varela argued that cognition is not simply information processing but a form of sense-making rooted in a system's ongoing self-maintenance. This perspective challenged input-output models of intelligence and contributed to embodied and enactive approaches to AI, which emphasize that intelligent behavior arises from a system's continuous interaction with its environment rather than from symbolic manipulation of abstract representations.
For machine learning and autonomous systems research, autopoiesis offers a conceptual framework for designing agents that can self-regulate, adapt, and preserve functional integrity under changing conditions. Systems inspired by autopoietic principles aim to maintain internal coherence — adjusting parameters, restructuring representations, or modifying behavior — without requiring constant external supervision. This connects to active areas such as continual learning, self-organizing neural networks, and homeostatic regulation in artificial agents.
While autopoiesis remains more of a theoretical inspiration than a directly implemented algorithm, its influence is visible in research on self-supervised learning, adaptive robotics, and artificial life. The concept challenges engineers and researchers to think beyond static architectures toward systems that are genuinely self-sustaining over time — a goal that grows more relevant as AI is deployed in open-ended, unpredictable real-world environments.
