Tech companies desperately want to film you doing chores
Original reporting by The Verge
An AI training startup called Shift recently made an enticing offer: free home cleaning for New Yorkers, with plans to expand to London and other major cities. The proposition is simple and appealing – imagine your chores disappearing without a cost. But there’s a catch, as there so often is in the world of data. In exchange for spotless floors and gleaming countertops, Shift requests comprehensive video footage of its cleaners at work, capturing every detail of the domestic labor we all wish we could outsource.
The Data Scramble
This isn’t about surveillance for its own sake. Robotics companies are racing to teach machines how to navigate and interact with the complex, unpredictable physical world – a monumental challenge that chatbots and image generators don't face. Tasks like folding laundry or picking up an apple, intuitive for humans, are maddeningly difficult for robots to master. Unlike digital data, which was easily scraped from the internet, high-quality physical world data is scarce and expensive. To overcome this bottleneck, companies like Shift are employing novel strategies. They join a growing trend, with some startups reportedly recording inside clients’ homes (with opt-in), others paying individuals to wear camera hats to collect "egocentric" data, and still others creating controlled "data farms" where workers perform repetitive tasks for cameras. The ultimate goal: to gather the rich, real-world information needed to power the next generation of automated domestic helpers.
The scramble for physical world data, exemplified by initiatives like Shift’s free cleaning offer and Pronto’s in-home recording, underscores a critical bottleneck in AI’s evolution. Unlike the vast, easily accessible digital datasets that fueled earlier AI breakthroughs, real-world interactions and movements are proving far more elusive and expensive to capture. This creative — and sometimes controversial — data acquisition phase is shaping the trajectory of robotic development, moving it beyond virtual environments into the messy, unpredictable reality of our homes and workplaces.
The Data Imperative
This intensifying pursuit of physical data carries profound implications. On one hand, it represents the engine driving the next generation of AI, promising to unlock truly autonomous robots capable of performing complex tasks in our environments. The allure of outsourcing domestic chores to an intelligent machine is immense, potentially freeing up human time and energy. However, this progress is inherently tied to novel challenges concerning privacy, consent, and the commodification of private life. As companies devise increasingly ingenious ways to record human activity, the line between innovation and intrusion blurs, necessitating robust ethical frameworks and transparent practices. The exchange of services for access to our physical world data is not merely a transaction; it’s a foundational shift in how value is created and extracted in the burgeoning age of embodied AI, with consumers inevitably facing a future where their routines and environments become invaluable inputs for machine learning. The debate over who owns this data, and how it is used, will define the integration of robots into daily life.