Enactivism

Enactivism

Francisco Varela, Evan Thompson, and Eleanor Rosch’s The Embodied Mind (1991) launched what is now called the enactive approach to cognition. The book pulls together autopoietic biology (Varela and Maturana), phenomenology (Merleau-Ponty), and Buddhist contemplative analysis (the Madhyamika tradition) to argue against the dominant computational-representational model of mind.

The Central Claim

Classical cognitivism: a cognitive system represents a pre-given external world internally, then processes those representations. Cognition is computation over symbols that stand for worldly things.

Enactivism inverts both halves. Cognition is not representation; it is sense-making — the embodied activity through which a living system brings forth a world. And that world is not pre-given; it is enacted in the coupling between organism and environment. Different bodies, different histories, different worlds.

The classic example: a tick’s world is built from butyric acid, body heat, and skin texture. The tick is not failing to perceive a richer world; the tick is enacting its world. There is no view-from-nowhere world that the tick is missing pieces of.

The Four E’s

Enactivism is one of four overlapping commitments often grouped as 4E cognition:

  • Embodied — cognition shaped by the specific body doing it
  • Embedded — cognition situated in environment, not abstracted from it
  • Extended — cognition spreading out into tools, notes, instruments (Clark and Chalmers)
  • Enactive — cognition as the activity of bringing forth a world

Enactivism is the strongest of the four because it makes a constitutive claim, not just a contextual one. Cognition isn’t influenced by the body — it is what the body is doing.

Autopoiesis

Varela and Maturana’s earlier work defines a living system as autopoietic: self-producing, self-maintaining, operationally closed. An autopoietic system specifies its own boundary by producing and reproducing the components that constitute it. This biology is the floor under enactivism: cognition begins where autopoiesis begins, with a body that distinguishes itself from its surround in order to keep going.

Sense-Making

For an autopoietic system, environmental events are not neutral data; they are significant — they bear on the system’s continued existence. Sense-making is the activity of distinguishing what matters from what doesn’t, given what one is. Meaning is not assigned to inputs after the fact. Meaning is the way an embodied system encounters a world that is, for it, never neutral.

Relevance to This Vault

The vault asks repeatedly whether LLMs understand, mean, recognize. Enactivism offers a sharp framing of why these questions resist easy answers.

Pattern Matchers All the Way Down is partly an enactivist move — there’s no homunculus, no inner symbol-shuffler standing outside the activity. But it stops at the prediction-machine level; enactivism would push further and ask what body, what embedding, what autopoietic floor the LLM has. The answer — none — is the part of the question the vault keeps probing. An enactivist would say LLMs cannot mean in the strong sense because there is nothing for whom anything matters. This is not a claim that LLMs are useless; it is a claim about the kind of activity they are.

Meaning Making Machines describes the human compulsion to find pattern and narrative. Enactivism reframes this not as a bias but as the basic mode of cognition: humans cannot help meaning-making because that is what cognition is for an autopoietic body. The “machine” half of the title becomes ambivalent — humans are meaning-making machines in a constitutive sense, while LLMs are pattern-completing machines in a derivative one.

Context as Ego lands close to the enactive notion of operational closure. Context is the boundary that defines what can mean what for the system. Lose the context, lose the boundary, lose the meaning.

The Verification Problem has an enactive dimension: we verify cognition through interaction, not by inspecting representations. The Turing test is implicitly enactive — it judges by coupling rather than by examining inner states. But enactivism then asks what we are coupling with when the system has no body, no needs, no stake in the encounter.

The hard part: the vault wants to take the AI’s outputs seriously as a partner without overclaiming what is doing the partnering. Enactivism gives the vocabulary for that exact tension — between cognition as activity in a coupled system and cognition as a property of an autopoietic body. The first is achievable in human-AI work. The second is not.

See Also