A new gig economy is emerging in living rooms and kitchens across more than 50 countries: workers filming themselves doing laundry, washing dishes, and folding clothes — not for social media, but to teach humanoid robots how to navigate the physical world. The company orchestrating much of this effort is Micro1, a startup that recruits contractors in countries including Kenya, the Philippines, India, and Brazil to strap on cameras and record hours of domestic tasks — loading dishwashers, sorting socks, wiping counters — for around $15 an hour. That rate is competitive in Nairobi or Manila. It is a rounding error against the billions flowing into the robotics companies that will monetise the footage. And that gap — between the value extracted and the compensation offered — is the defining story of how humanoid robots are actually being built.
Photo by Ehaan Deva on Pexels
The data pipeline behind humanoid robots
As MIT Technology Review reports, Micro1 has recruited gig workers internationally to record themselves performing household tasks — folding towels, opening refrigerators, stacking plates, mopping floors. The footage feeds the same data-hungry approach that powered large language models, except instead of training AI on text scraped from the internet, robotics companies need massive libraries of real-world physical manipulation: human hands gripping, twisting, lifting, and placing objects in the cluttered, unpredictable geometry of actual homes.
The economics reveal the asymmetry. Investors have poured billions into humanoid robots, and robotics companies are spending heavily on real-world training data from providers in this space. Scale AI alone has collected more than 100,000 hours of training footage. Meanwhile, companies like DoorDash have begun letting delivery drivers contribute training data as a side hustle. The structural pattern is now unmistakable: companies in wealthy markets outsource the labour-intensive foundation of AI development to lower-cost workers globally, capturing most of the value at the top of the chain.
$15 per hour — from workers who have no stake in what they’re building
Fifteen dollars an hour is significant income in many of the developing economies where these platforms recruit. But that competitive local rate obscures a deeper imbalance. The workers filming their kitchens earn a flat hourly fee. They hold no equity, receive no royalties, and retain no rights over the data they produce. If the footage they generate helps a robotics company achieve a breakthrough valued at billions, the worker who filmed herself folding laundry in Lagos sees nothing beyond the original payment.
This is the same architecture that built the large language model industry. The workers who labelled images for computer vision, moderated content for social platforms, and annotated text for LLMs operated under identical conditions — competitive local pay, limited transparency about end use, and zero leverage over the companies purchasing their output. As robotics firms attract significant venture funding, the gap between capital flowing to robot companies and the protections extended to the workers generating their foundational data is not merely a pattern worth watching. It is the business model.
The privacy cost no one is pricing in
The labour exploitation would be concerning enough on its own. But the data being collected carries a distinct and unprecedented character. These are not warehouse environments or public streets — they are people’s homes. The footage captures kitchen layouts, family photos on walls, children’s toys on floors, medication on counters, the interior geography of private life. And according to industry reporting, many workers lack a clear understanding of how their intimate home videos will be used, stored, or shared downstream.
Researchers studying human-centred computing have noted the importance of ensuring workers engaging in this type of data collection are informed by the companies themselves about how this technology might develop and how that might affect them longer term. But informing workers after they have already filmed the inside of their homes is not consent — it is notification. And the robotics companies purchasing this footage have disclosed little about their data retention policies, whether footage is anonymised, or what happens to a video of a worker’s living room once the training run is complete.
The question is not abstract. If 100,000 hours of footage exist showing the interiors of homes in 50-plus countries — the homes of workers who were paid $15 an hour and clicked through a terms-of-service agreement — who owns that archive? Who can access it? And what happens when a robotics company is acquired, goes bankrupt, or suffers a data breach? The workers filming themselves loading a washing machine in Bangalore are not just training robots. They are building a visual database of domestic life across the developing world, and they are being paid a fraction of what that database is worth while bearing all of the privacy risk.
Photo by cottonbro studio on Pexels
Feature image by Tima Miroshnichenko on Pexels
Originally written by: Daniel Voss
Source: Silicon Canals
Published on: 2 April 2026
Link to original article: Gig workers in 50+ countries are filming themselves doing chores to train humanoid robots for $15 an hour
