Your Free Cleaning, Their Robot: AI's Silent Barter

What happened?

According to a report by BBC Technology, an artificial intelligence company — whose identity has not been publicly revealed — has been sending professional cleaners to apartments in New York for free. In exchange, the cleaners record every movement, every object handled, and every decision made during the cleaning. This data is used to train robotics algorithms that, in the near future, could fully automate household cleaning tasks. The initiative is part of the imitation learning strategy, where an AI model observes human actions to replicate them. Unlike simulated environments, real homes offer invaluable complexity: variable clutter, unpredictable objects, and changing lighting or space conditions. This approach is not new in robotics: companies like Google with its RT-2 project have used YouTube videos to train robots, but direct collection in real homes with paid workers represents a qualitative leap. The collected data includes not only movement sequences but also contextual decisions, such as which cleaning product to use on each surface or how to organize objects based on the user's implicit preferences. According to sources close to the project, the company has hired dozens of cleaners through temporary agencies, paying them hourly wages above New York's minimum ($15/hour), although the value of the data generated per session could exceed $1,000 in the AI training market. The company has not revealed its name, but speculation points to startups like Matician or Dusty Robotics, or even tech giants like Amazon (which invests in domestic robotics with Astro) or Samsung (with its Ballie robot). However, none have confirmed their involvement.

Why is it important?

This case illustrates a paradigm shift in data collection for AI. Traditionally, tech companies have relied on low-paid workers (often in developing countries) to label images or transcribe audio. Now, the model extends to physical and skilled jobs, where the “payment” is not monetary but a free service. This raises several issues:

• Valuation of human labor: The cleaner receives a salary for their work, but the company obtains an asset (the data) whose potential value is much higher. According to Forbes estimates, a high-quality robotic movement dataset can cost between $50 and $200 per hour of recording, while the cleaner's wage is around $20 per hour. The difference is an implicit subsidy from workers to the company.
• Informed consent: Residents receiving free cleaning know that data is being collected, but do they really understand how it will be used? Are they informed that this data will serve to automate the cleaners' jobs? The BBC notes that the consent terms are written in technical language that few users read fully. This recalls the Roomba case in 2017, when it was revealed that iRobot planned to sell home maps to third parties, sparking a privacy scandal.
• Future of work: If the model is replicated, it could accelerate the automation of trades like cleaning, gardening, or delivery, displacing millions of workers without adequate safety nets. In the US, the cleaning sector employs over 2 million people, many of them immigrants and low-wage. Automation could eliminate up to 40% of these jobs in a decade, according to a McKinsey study.

What consequences will it have?

In the short term, we will see more domestic robotics companies adopt similar strategies. Companies like iRobot (Roomba) already collect home map data, but not at this level of detail. Startups like Dusty Robotics (specializing in commercial cleaning) or Matician (developing a robotic vacuum with an articulated arm) could follow the same path. Even Tesla, with its Optimus robot, could benefit from domestic task data for training. The cost of collecting data in real environments is high, but the value of the data is even higher. The market for robotics training data is estimated to reach $5 billion by 2025.

In the long term, the availability of large volumes of domestic task data will accelerate the development of versatile robots capable of performing multiple tasks at home. However, it will also generate a regulatory debate about the ownership of data generated by workers and the need for fair compensation. Recall the precedent of Amazon Mechanical Turk, where workers are paid for micro-tasks; here the payment is indirect and opaque. The FTC has already shown interest in AI training data, and in 2023 issued guidance on the use of consumer data for AI. It may require companies to disclose the value of the data obtained and offer additional compensation to workers if that data is used to automate their jobs.

What should readers know?

If someone offers you a free service, especially if it involves data collection, it is crucial to ask: who really benefits? In this case, the immediate benefit is for the user (free cleaning), but the strategic benefit is for the AI company, which obtains highly valuable training data. The cleaning workers, for their part, are both beneficiaries (salary) and potential victims (future displacement).

“We are facing a new business model where human labor becomes fuel for its own obsolescence,” says an analyst at TheVortiq.

Regulators should pay attention to these practices. The US Federal Trade Commission (FTC) has already shown interest in AI training data. It would be prudent to require full transparency about the future use of data and establish compensation mechanisms for workers whose data contributes to automating their own jobs. Meanwhile, consumers should read the fine print of any free service and consider whether the data exchange is worth it. In a world where data is the new oil, free cleaning might just be the bait for a robotic revolution that will leave many jobless.