Perception Is Not Computation

In the AI Story

Cézanne's paintings of Mont Sainte-Victoire serve as Merleau-Ponty's paradigmatic demonstration. The paintings do not record the mountain as a camera would. They depict the experience of perceiving the mountain — the ambiguous outlines, the interpenetrating spatial relationships, the colors that vibrate against each other. Cézanne captured not the optical data but the lived perceptual encounter, which is categorically different from any representation data could generate.

The distinction bears directly on what AI systems do when they 'see' or 'understand.' A computer vision system processes pixel arrays through neural networks to produce labels. The process may be accurate, useful, sophisticated. But it is not perception in Merleau-Ponty's sense. The system has no body approaching the object from different angles, no motor anticipation of hidden faces, no felt pull toward surfaces that solicit further exploration.

The depth of a cube, for embodied perception, is not a computed distance. It is a felt tension — the bodily anticipation of a surface not yet seen but already oriented toward. The perceiver reaches toward the object, sees its near face, anticipates the far face, shifts position to bring new aspects into view. The object is constituted through this progressive exploration, not discovered as a complete representation.

For understanding AI, the distinction dissolves the common claim that AI systems 'see' or 'understand' in ways comparable to human perception. The outputs may be accurate and useful. The process producing them is not perception but representation-processing — a categorically different activity.

Origin

Merleau-Ponty developed the distinction throughout Phenomenology of Perception, particularly in the chapters on sense experience, the body, and spatiality. His extensive engagement with Cézanne — most fully in the essays 'Cézanne's Doubt' (1945) and 'Eye and Mind' (1961) — provided the artistic demonstration that phenomenological argument alone could not achieve.

The distinction has been extended by contemporary enactivist researchers. Alva Noë's Action in Perception (2004) formalizes the claim that perception is an active achievement of embodied organisms rather than a computational process. Evan Thompson's Mind in Life (2007) extends the argument into biology, grounding cognition in the life of embodied, autopoietic systems.

Key Ideas

Perception is exploration. The perceiver does not receive complete images but actively constitutes objects through bodily engagement — turning, reaching, shifting to bring new aspects into view.

Temporal unfolding. Perception unfolds in time through the body's motile engagement, not as instantaneous computation.

Motor intentionality. Before cognition categorizes, the body reaches, orients, and shapes itself to what is perceived.

Cézanne's demonstration. The painter's work reveals what perception actually is — ambiguous, unfolding, embodied — against the false clarity of the camera.

Categorical difference from AI. Computational systems process representations with extraordinary sophistication. They do not perceive, because perception requires a body engaged with a world.