Abstract
Progress in enabling machines to perform an expanding range of manual tasks depends on task-level data that encode how those tasks are carried out. Generating these data requires workers to record their actions as they perform their jobs. Workers thus supply a key input into the creation of labor-substituting capital, potentially facilitating their own displacement and shaping how the gains from automation are divided between labor and capital. We study how workers make this decision, and whether it is shaped by information asymmetry over the downstream use of their data and by weak worker bargaining power. In a survey experiment with factory workers in Delhi NCR, India, drawn from textiles, electronics, and other manufacturing, we elicit the minimum additional daily payment workers require to make first-person recordings of their work, cross-randomizing whether workers are informed that the resulting footage will be used to train a machine that could replace them. Each worker prices two counterparties in randomized order, their own employer and an outside research organisation, which identifies what the employment relationship does to the price of consent.