The Enactive Approach

In the AI Story

The enactive approach emerged from a specific intellectual lineage: Maturana and Varela's biology of autopoiesis, Merleau-Ponty's phenomenology of embodiment, and the Buddhist contemplative traditions Varela and Thompson drew upon throughout their collaboration. Each contributed a different component. Autopoiesis supplied the organizational definition of life. Phenomenology supplied the analysis of embodied experience. Contemplative practice supplied the disciplined first-person methods that neurophenomenology would later operationalize. The synthesis produced a framework in which the body is not the platform for cognition but the process through which cognition occurs.

The approach stands against what Thompson calls the received view: the assumption that the world exists independently of the mind, that the mind receives information about the world through the senses, and that cognition consists in the computational processing of that information. The received view generates the hard problem of consciousness as an unavoidable consequence of its premises. The enactive approach refuses the premises. There is no pre-given world that is subsequently represented. The organism's world is brought forth through its activity, and the activity is simultaneously physical and experiential — a single process described from two perspectives.

For the AI discourse, the approach's central consequence is the distinction between sense-making and information processing. The bacterium navigating a sugar gradient is making sense of its environment — evaluating it in terms of its own needs, its own survival. A large language model producing plausible text is processing tokens without making sense of anything. The outputs may be indistinguishable at the surface. The underlying processes are categorically different, and the difference determines what the collaboration between human and AI actually is.

The enactive approach does not issue verdicts on engineering timelines. It identifies what would have to be true of a system for that system to count as cognitive in the biological sense, and it demonstrates that current AI architectures do not meet the conditions. The demonstration is not a pessimistic forecast but a conceptual clarification — a map of where understanding actually lives when a human and a machine produce work together.

Origin

The enactive approach was first articulated in The Embodied Mind (1991), co-authored by Thompson, Varela, and Rosch, and developed through Thompson's Mind in Life (2007) and Waking, Dreaming, Being (2015). The January 2025 Nature letter in which Thompson and three colleagues argued that AI will never achieve human-level intelligence represents the framework's most pointed application to the contemporary AI debate.

Key Ideas

Cognition is enacted, not computed. Understanding is the lived activity of an embodied organism, not the manipulation of internal symbols.

World and mind co-emerge. The organism and its environment mutually specify each other through structural coupling; neither pre-exists the relationship.

Sense-making is the minimal form of cognition. From the bacterium to the philosopher, cognition is the organism's creation of significance through its own activity.

Processing and enacting are categorically different. An AI system's outputs may be useful and appropriate, but the system does not enact meaning — meaning is supplied by the human who engages with the output.

Debates & Critiques

Critics argue that the enactive approach conflates two claims: the descriptive claim that biological cognition is embodied, and the normative claim that only embodied systems can be cognitive. Defenders respond that the conflation is illusory — the descriptive claim, properly understood, entails the normative claim, because cognition is not an abstract function that can be separated from the living process through which it occurs.