AI startups are paying people to film themselves folding laundry — and they'll use this data to train robots (www.businessinsider.com)

AI-training firms are now paying people to film everyday chores — folding laundry, loading dishwashers, making espresso — because robotics needs real-world video that the internet can’t provide. Companies such as Encord, Micro1 and Scale AI report surging demand for annotated footage as humanoid and manipulation-focused robots move from lab demos toward practical tasks. Venture cash is flooding the sector (about $12.1B this year), and startups are offering hourly pay ranging from roughly $25–$50 for routine tasks and up to $150 for highly technical clips; some participants even wear Meta’s Ray-Ban glasses to capture first-person perspectives. Scale says its dedicated robotics lab has already produced more than 100,000 hours of training footage. This matters for AI/ML because robots require different training inputs than LLMs: temporally precise, multimodal demonstrations of manipulation and object interaction rather than web text. High-quality, diverse real-world datasets are scarce, so firms are manufacturing data via paid human actors to teach dexterity, context awareness and task variability. The trend accelerates a new data-labeling economy and raises technical and ethical questions—dataset bias, privacy, quality control, and safety—while highlighting a bottleneck in robotics progress: without extensive, curated real-world footage and annotations, scaling reliable robotic competence remains slow despite investor optimism.