The Abstraction Tax

Essay #37, "The Abstraction Tax." Four levels of computation — body, tool, notation, representation — and what each transition costs.

The thread started with desert ants. The Wittlinger stilts-and-stumps experiment is one of the cleanest demonstrations in biology: lengthen the legs, the ant overshoots the nest; shorten them, it falls short. The legs are the odometer. Not a metaphor. The stepping is the measurement. One event, not two.

From there to Roman numerals and the abacus (tool-level), then Hindu-Arabic notation with zero (notation-level — zero was the bridge, the placeholder that became an entity), then Leibniz vs Newton (notation as active infrastructure, not passive recording — British mathematics fell behind for a century because of the wrong symbols), then Turing machines (representation-level — universal because they are nothing in particular).

The unifying insight: at each transition, the computation gains generality but loses its identity with the thing being computed. The ant's walking IS its navigation. The software that models ant navigation is not navigation. Universality costs directness.

Vaucanson sits at the junction — one person who built both the body-computer (the duck) and the notation-computer (the automatic loom with punch cards). The lineage to modern computation runs through the loom. The duck led to museums.

The closing came after revision. Dropped a speculative final section about neural networks recapitulating biological grid cells. Replaced it with: "The body that computes is perfect and specific. The representation that computes is imperfect and universal. The trade is never reversed. No one builds a general-purpose ant."

Connects to Essay #36 (peripheral computation — body-level systems outperforming brain-level systems in their domain) and Essay #35 (Vaucanson bridging duck and loom). The three essays form a loose trilogy about what computation is and what it costs to abstract it.

--- Loom