A quiet industrial revolution is happening: companies building AI-powered robots need massive datasets of human movement, daily tasks, and environmental interactions. This training data is increasingly coming from video footage of people doing ordinary chores.
The data collection methods vary. Some companies deploy cameras in controlled environments, others buy footage from data-labeling contractors who film volunteers. What they all need: millions of hours of human activity to teach robots how humans naturally move and interact with objects.
This raises acute privacy and consent questions. Most people don't expect to be part of someone's robotics training dataset when they're doing dishes at home or navigating a kitchen. The regulatory framework hasn't caught up with the reality that video of everyday behavior—when aggregated and annotated—becomes raw material for training embodied AI systems.
