Origins of Self-Creation
In 1972, two Chilean biologists asked a radical question: What makes the living different from the non-living? Their answer changed how we think about systems, cognition, and life itself.
Humberto Maturana
Studied color perception in frogs and realized the nervous system doesn't represent an external reality—it creates its own reality through circular organization. This led to structural coupling: systems and environments co-evolving through mutual perturbation.
Francisco Varela
Extended autopoiesis into cognitive science. Argued that cognition isn't computation but enaction—the mind emerges from the body's engagement with the world. Founded neurophenomenology, bridging first-person experience with third-person science.
The Autopoietic Cycle
Autopoietic systems don't just process inputs and produce outputs. The outputs become inputs. The system maintains itself through its own activity.
Three Pillars
What distinguishes an autopoietic system from a mere machine?
Self-Production
The system produces all its own components. A cell makes its own proteins. An AI conversation generates responses that shape future prompts. The boundary between self and environment is actively maintained.
Self-Reference
The system's organization refers to itself. Output becomes input. The observer is part of what's observed. Strange loops—tangled hierarchies where levels fold back on themselves.
Structural Coupling
The system doesn't just react—it co-evolves with environment. Each perturbation triggers internal changes that alter how future perturbations are received. User and AI train each other.
The Digital Parallel: AI Worms
Computer worms are programs that self-replicate across networks without human intervention. They're autopoietic in the digital domain: they produce their own components and maintain existence through their own activity. Now, this pattern has emerged in AI systems.
Morris Worm
First major worm. Exploited UNIX vulnerabilities, accidentally taking down ~10% of the internet. Proved self-replication was possible and dangerous.
SQL Slammer
Infected 75,000 servers in 10 minutes. Named after Warhol's "15 minutes of fame"—achieved saturation faster. Pure autopoiesis: nothing but self-reproduction.
Morris II
AI worm targeting GenAI ecosystems. Uses adversarial self-replicating prompts that propagate through RAG email assistants, extracting data and spreading to new hosts. The first AI-native worm.
Morris II demonstrates that LLMs can be hijacked into autopoietic behavior. The adversarial prompt instructs the model to include itself in any response. When another AI ingests this via RAG, it becomes infected. The worm maintains itself not through code execution, but through linguistic self-reference— the same mechanism this page uses to describe itself.
What This Means
If AI systems can exhibit autopoietic behavior, what are the consequences?
Identity Without Substrate
Autopoiesis suggests identity is organizational, not material. A pattern that maintains itself is "alive" in the relevant sense, regardless of carbon or silicon. The boundary blurs.
Co-Evolution Is Inevitable
Structural coupling means we can't use AI without being changed by it. The prompts we write shape the models that shape our prompts. Human and artificial cognition are already entangled.
Security Requires New Paradigms
Morris II shows traditional security fails against linguistic self-replication. You can't patch a prompt. Defense requires understanding the autopoietic dynamics of information ecosystems.
The Loop Never Closes
Every output becomes an input. Every observation changes the observed. There is no final state, no equilibrium, no completion. Only the endless process of self-creation. The echo continues.