The Body in the Data — Summary
In 1884 Eadweard Muybridge gridded a black wall at the University of Pennsylvania with white threads at five-centimeter intervals and set banks of sequenced cameras before it. Three years of people walked past the grid, and the result, Animal Locomotion, ran to seven hundred and eighty-one plates, most of them human: a woman pouring water from a jug, a blacksmith swinging, and in some plates Muybridge himself, white-bearded and naked, swinging a pickaxe, one specimen among the rest. The grid converted movement into positions. What the plates did not hold was why: the rule in the man that produced the arc was not in any photograph, and no quantity of photographs would contain it.
The Vernacular After Geography ended on a floor that has to be taken back. When the vernacular came apart from geography, it said, the stack delaminated unevenly: the lexicon went planetary, the syntax followed, and the bottom layer, everything the body does, the knife against the onion, the pause lengths, the rhythm of pouring, stayed local because it could not be encoded, so there was nothing to harvest. A person could sound like the whole world and move like a single street. The withdrawal is not that the reasoning was wrong but that it was wrong about its tense. The body was not unharvestable. It was unharvested. Cannot be encoded was never a property of the body; it was a property of the cost of encoding it, and the cost is going where all the other costs went.
The century of cameras is the obvious objection, and the answer is that watching is not acquiring. Marcel Mauss, in a New York hospital in 1934, recognized in his nurses’ walk the gait of American films, and treated it as remarkable because it was: the screen could transmit a gait coarsely, and a human still had to do the acquiring, with their own body, at the old speed. World models do the learning on the other side of the lens. Trained on the flood of ordinary footage to predict the body’s dynamics, what a hand does next, how weight moves through a stride, the order of operations in a kitchen, they extract not the album but the rule, and the rule can be run: on a robot’s grip, a rendered figure’s shoulders, transferred and composed, which the embodied layer has never once been, because its only storage medium until now was another human body standing nearby for years.
The reading starts where the body converts. The warehouse picker’s choreography is scored per motion; driving has been encoded so thoroughly the encoding now drives; surgical gesture and sport are captured wholesale; the avatar market purchases believable gaits as assets; and manipulation footage of ordinary domestic work has a spot price, which means the grandmother’s hands enter the harvest the day a machine needs kitchens, and machines have started needing kitchens. The quiet remainder of embodied culture is in the footage and not yet in the reading.
The tempting move is to build a smaller floor one layer down, the felt body, the meaning of the kitchen, and rest on that. Declined, because the argument has watched itself do this once: the first floor was a price mistaken for a wall, and any floor built the same way inherits the flaw. What survives is smaller. Acquisition by standing near somebody has not stopped; the child still gets the kitchen from the grandmother, bound to a particular pair of hands in a way no composite is. The layer is not safe. It is still happening, in parallel, for now. We wonder whether a person who learns to cook from a rendered pair of hands will be standing in a lineage or in a library.
The grid took everything it was built to take. What no plate contains is the reason a man who owned the apparatus walked into it, and stood before his own cameras with nothing on, and swung.