Tech companies desperately want to film you doing chores
This week, an AI training startup named Shift announced it would provide free home cleaning services to residents of New York, with ambitions to extend this initiative to other cities, including London. Looking around the typical household, it’s easy to see why this idea has traction.
However, the offer comes with a significant condition. Shift requires video footage of its cleaners performing various domestic tasks such as scrubbing dishes, wiping counters, dusting surfaces, and mopping floors. Essentially, the startup seeks comprehensive visual data capturing everyday household chores—work that many would gladly delegate if given the option. Robotics firms are actively competing to develop machines capable of performing these tasks, with plans to commercialize such automation.
Teaching robots to master routine household activities remains an immense challenge. Unlike software AI applications such as chatbots or image generators, physical robots interact with a complex, unpredictable environment where factors like three-dimensional space, motion, tactile feedback, and varied materials come into play. Simple human actions—folding clothes, picking up objects, pouring liquids—prove extraordinarily difficult to program effectively, requiring sophisticated understanding and data.
The development process demands vast amounts of data, particularly visual and motion data. While text and image data can be harvested on a massive scale from online sources, physical world data is much harder to acquire discreetly or cheaply. Access to high-quality footage and recordings of real-world physical activities represents a major bottleneck for companies building physical AI systems, prompting startups such as Shift to innovate in data gathering strategies.
This approach is not isolated. In India, reports have surfaced about the platform Pronto utilizing footage from clients’ homes to train AI for tasks like cooking, cleaning, and laundry, only recording with customer consent. This practice sparked backlash, with competing startups distancing themselves by claiming they neither conduct nor plan such recordings inside homes.
Other ventures seek to scale data collection efforts differently. Silicon Valley’s Human Archive aims to collaborate with companies to outfit gig workers with wearable camera caps that record first-person perspectives of daily activities—precisely the egocentric viewpoints necessary for training robots on spatial navigation and task execution. Similarly, Shift engages directly with consumers worldwide, reportedly compensating tens of thousands across numerous countries for activity recordings via its app.
Certain organizations abandon naturalistic settings altogether, instead orchestrating controlled sessions where workers repetitively perform specific physical actions while being filmed by cameras and sensors. These “data farms” produce consistent, high-quality AI training material justifying financial incentives to participants.
Additionally, some data is now generated by robots already in operation in homes, though fully autonomous robotics are still distant prospects. Companies release products while simultaneously collecting data from users to refine performance and incorporate remote human aid when automation falters, feeding that information back into machine learning cycles.
The concept of exchanging personal data for benefits is longstanding, visible in loyalty programs, cookies, telematics insurance, and smart devices that monetize user information. What distinguishes this emerging trend is the willingness to trade access for complex, physical task data. For the moment, this might mean welcoming a human cleaner equipped with cutting-edge recording technology to your home at no cost, ultimately paving the way for selling robotic replacements to do the same work in the future.
