A growing gig economy sees workers in more than 50 countries filming mundane chores like folding laundry and loading dishwashers - all to train humanoid robots. Led by startup Micro1, these contractors, primarily from Kenya, the Philippines, India, and Brazil, wear cameras while performing daily tasks for around $15 an hour.
The footage is fed into data pipelines that help robotics firms build physical AI capabilities. Companies require vast libraries of real-world actions to teach machines how humans interact with everyday objects. This method mirrors how large language models were trained - but now with video of physical manipulation instead of text.
Despite contributing to multi-billion-dollar industries, most workers receive no equity or future benefits from their data contributions. Their pay is modest by global standards but substantial locally. However, they retain no ownership rights even if their data leads to major robotic breakthroughs.
Privacy concerns add another layer. Videos capture intimate details of personal spaces - from medicine bottles to children’s toys. Many participants do not fully understand how this data may be used long-term. Questions remain unanswered about who owns the footage, how it's stored, or what happens in case of corporate changes or breaches.
