A growing workforce of gig workers across the globe is recording themselves doing household chores to train humanoid robots, and the arrangement raises serious questions about data privacy, safety, and labor transparency. MIT Tech Review reports on how company Micro1 is recruiting thousands of these workers to capture video data of everyday tasks in their homes.
The setup sounds straightforward: workers film themselves cooking, cleaning, and doing laundry. That footage feeds into training datasets for robotics companies building humanoid robots that need to learn how humans move through domestic spaces. But the details get complicated fast.
The Privacy Problem Nobody Solved
Workers are navigating real privacy challenges with zero institutional support. One father struggles to keep his two-year-old daughter out of frame. A Nigerian worker tiptoes around her shared compound to avoid recording neighbors who stare in confusion. These aren’t edge cases. They’re the daily reality of turning your home into a data collection studio.
What’s more concerning: none of the workers interviewed by MIT Tech Review know how their data will be stored, shared, or sold. Micro1 doesn’t name its robotics clients or tell workers what specific projects their footage supports. The company’s defense? “People are opting into doing this. They could stop the work at any time,” says CEO Ali Ansari.
That framing misses the point entirely. Yasmine Kotturi, a professor of human-centered computing at the University of Maryland, puts it directly: workers should be informed about “where this kind of technology might go and how that might affect them longer term.”
Can Messy Home Data Actually Train Robots?
Beyond privacy, there’s a fundamental quality question. Roboticists are skeptical that chaotic, uncontrolled home recordings can produce reliable training data.
“How we conduct our lives in our homes is not always right from a safety point of view,” says Aaron Prather, a roboticist at ASTM International. “If those folks are teaching those bad habits that could lead to an incident, then that’s not good data.”
Micro1 says it rejects unsafe videos and argues that even clumsy movements help teach robots what not to do. That’s a reasonable point, but the sheer volume of footage makes quality control a serious bottleneck. When thousands of workers record differently in different environments, consistency becomes nearly impossible to enforce.
UC Berkeley’s Ken Goldberg offers a sobering timeline check: “It’s going to take longer than people think.”
Why This Matters Right Now
This story sits at the intersection of three major AI trends:
- The humanoid robot race is accelerating. Companies need massive amounts of real-world movement data, and they need it fast. That demand is creating new gig economy categories that existing labor protections don’t cover.
- Data sourcing is getting messier. As AI companies exhaust clean, curated datasets, they’re turning to distributed, uncontrolled collection methods. The trade-off between scale and quality is real.
- Informed consent in AI data work remains broken. Workers contributing to AI systems routinely don’t know what they’re building, who benefits, or what happens to their data. This pattern repeats across content moderation, RLHF labeling, and now robotics training.
What Should AI Practitioners Take Away?
If you’re building robots or buying training data, the transparency gap here is a liability. Regulations are coming. The EU AI Act already requires documentation of training data provenance. Companies that can’t explain where their data came from and how consent was obtained will face problems.
For the robotics field specifically, the home-recording approach highlights a deeper tension: you can collect massive datasets cheaply, but cheap data isn’t always good data. The companies that win the humanoid robot race will likely be the ones that figure out quality control at scale, not just volume.
The full story is worth reading over at MIT Tech Review for the worker interviews and additional technical context.