24-year-old Anjali Sardana has founded a startup called Pronto, which sends domestic workers, such as house cleaners, into customers’ homes wearing head-mounted cameras. The recordings are said to be used to collect AI training data for AI companies in the United States.
Customers have the ability to opt into a program that allows household jobs to be recorded using a head-mounted camera worn by the workers inside your home.
As AI companies face increasing demand for more diverse and specialized training, rather than publicly available texts and images, they need new data to train AI on. The proposed avenue is collecting real-world human behaviors, including how people move, clean, and interact with their environment. This information is used and studied by artificial intelligence for development.
Because homes vary widely in layout, objects, size, and location, these environments provide a wide range of environments for AI training and deliver less predictable settings for AI to explore.
Since AI and robots don’t naturally understand tasks such as loading a dishwasher, folding laundry, and navigating around furniture, these videos would create learning opportunities to teach AI how to perform these everyday tasks.
This, however, is bigger than just training AI chatbots like ChatGPT. This type of data could eventually support the development of physical AI systems, including humanoid robots, and provide them the ability to perform tasks such as aiding in home assistance, helping with warehouse-like tasks, and more. This is training for the future of robotics.
The demand for this type of data comes from a broader shift in artificial intelligence focus. Early AI models were trained primarily on information already publicly available, such as images, videos, and other digital information. However, companies developing physical AI systems are tasked with teaching these machines how to understand and interact with the real-world.
An investor at Pronto stated that the videos do not capture audio and stated that the collected footage gets deleted within two days; however, the company has confirmed that certain datasets are retained from the videos, such as hand and joint movements that are tracked.
Customers who opt into this program pay an extra 29 Rupees for the service, and the workers are also paid more for recording the work; however, it has sparked controversy, as increased worries regarding AI using or selling personal and identifiable data persist. Though Pronto claims they delete footage after 48 hours, any personal information, family photos, valuables, security systems, and more can be stored within the derived datasets that remain.
For domestic workers, the program represents an opportunity, but also many potential challenges regarding their privacy. Additional compensation provides a new source of income, especially for workers in industries where wages can be scarce. This can raise ethical concerns regarding whether workers are freely choosing to participate, or whether financial pressures could influence their decision, ultimately forcing them to participate.
Others feel satisfied, as the program allows homeowners to watch unfamiliar workers in their homes, another precautionary measure in case items are broken or stolen within the home.
Some homeowners view this program as an additional layer of transparency and security. Because this system saves videos for 48 hours, it provides customers with security when allowing workers into their homes.
This footage can provide documentation of household tasks being completed as well as help resolve disputes if belongings are damaged, misplaced, or questions arise about what work was performed throughout the working period. Supporters argue that when used with clear consent and privacy protections, this technology provides benefits to both the homeworker and the worker, and helps companies expand artificial intelligence.
