open the original interactive work ↗
loading trajectories…

Hands enter, repeat a gesture, drift out of unison, and meet again. Each movement is both an action and a record of that action. Robots now learn from records like these. My guess, and it is only a guess: the work of human hands earns something like fifteen trillion dollars a year. Recordings of that work are bought for a few dollars an hour, sometimes for nothing, then copied into every machine that follows. Paid once and kept forever, the know-how of the hand is going for pennies on the dollar.

The estimate: about a quarter of world labour income goes to manual work, roughly 16 trillion dollars a year (range 12–22 trillion), from ILO and World Bank data. Reported pay for filming hand work to train robots in 2026 runs from nothing to about 15 dollars an hour (Guardian, TechCrunch, MIT Technology Review). Rough figures; corrections welcome. Hand motion: ARCTIC (Fan et al., CVPR 2023), CC BY-NC-SA 4.0. Hand model: MANO (Romero et al.), MPI, non-commercial. This adaptation uses a reduced set of nine notebook trajectories.