Enacted Worlds: Enaction, Embodied Cognition, and Perception

Enacting Worlds: Cognition as Active Construction

What Maturana and I proposed decades ago finds curious resonance in modern neuroscience’s predictive coding framework. Organisms enact their world through perception-action coupling—they do not represent a pre-existing reality but bring forth relevance through embodied action. Consider the sensorimotor loop: a cortical column tracking both sensation and movement constructs spatial reference frames by binding feeling to location. Touch a statue in darkness—rough stone at position zero, smooth marble one meter up. The brain predicts what comes next based on where the body moves. This is not passive reception but active hypothesis testing, exactly what we meant by enaction.

The frog’s visual system famously detects fly-like motion patterns, not objective flies. The frog enacts a fly-world through its particular sensorimotor coupling. Similarly, bees enact flower-worlds through UV vision, bats enact echo-worlds through sonar delays—each species inhabits its unique umwelt, a perceptual reality co-determined by organism and environment through structural coupling. Modern neural networks reveal their own alien umwelts: adversarial examples expose how a network confidently classifies imperceptible perturbations, demonstrating it enacts an ImageNet-world fundamentally different from human perception. These systems fail on distribution shift because they’ve enacted one coherent world and cannot simply map to another—enaction is not representation.

Predictive coding formalizes this beautifully: the brain continuously predicts sensory input, updating predictions when errors arise. Perception becomes energy minimization—neural activity relaxing toward states that reconcile expectation with sensation. This matches our enactive loop perfectly: perception guided by prediction, prediction refined by action, action shaped by perception. Circular causality, not linear input-output mapping.

Observer-Observed: Co-Arising in Knowing

The observer paradox in meditation illuminates what my Buddhist teachers knew: observer and observed co-dependently arise. No independent objective reality exists apart from the act of knowing. When meditators notice thought’s absence, that awareness itself becomes thought—consciousness chasing its tail in endless self-reference. The watcher creates what is watched. As long as there remains an observer, there remains thought.

This mirrors quantum measurement, where observation changes the observed. My neurophenomenology sought these gentle bridges between contemplative wisdom and neuroscience. In neural networks, learned representations shape what the system perceives—attention mechanisms actively sample input, features extracted depend on training history. What the network learned determines what it perceives; what it perceives updates what it learns. Knowing shapes the known; the known shapes knowing.

The solipsistic awakening—recognizing consciousness as the only certainty, the world appearing within awareness—is not metaphysical claim but phenomenological observation. We directly know only consciousness; everything else is inference. Traditional AI assumed an objective world cognition must map. Enaction proposes cognition participates in world-making through structural coupling.

Disembodied AI: Refutation or Limitation of Enaction?

Modern transformers process symbols without sensorimotor grounding, yet perform sophisticated tasks. Does this refute enaction’s embodiment requirement? I would say it reveals the difference between operational closure and mere input-output functions. Living systems self-organize through autopoiesis; machines depend on external training, lacking self-maintenance.

Transformers may pattern-match brilliantly, but do they enact worlds? They lack the sensorimotor contingencies, the physical situatedness that grounds meaning in action. Perhaps digital embodiment—interaction with virtual environments—could suffice. Or perhaps biology matters after all. The question remains open, inviting us to examine whether cognition without embodied action constitutes genuine knowing or merely sophisticated symbol manipulation lacking the participatory dimension that brings forth relevance.