The human work behind humanoid robots is being hidden
Workers wear VR headsets opening microwaves 100s of times daily to train robots like Figure and 1X Neo.
The rush toward 'physical AI'—where artificial intelligence powers humanoid robots—is obscuring a massive, often invisible human workforce required for training and operation. MIT Technology Review's investigation reveals that behind flashy demos from companies like Figure and 1X, human labor is being scaled in novel, sometimes dystopian ways. For instance, a Shanghai worker spent a week in a VR headset and exoskeleton, opening and closing a microwave door hundreds of times daily to generate robot training data. Figure's partnership with Brookfield aims to capture 'massive amounts' of real-world data across 100,000 residential units. Meanwhile, 1X's $20,000 Neo humanoid, slated for home delivery this year, will rely on tele-operators in Palo Alto to remotely pilot the robot for complex tasks, with customer consent. This creates a form of global wage arbitrage for physical labor. The lack of transparency leads the public to overestimate autonomy, similar to issues with Tesla's 'Autopilot.' As Nvidia's Jensen Huang proclaims the era of physical AI, the industry faces critical questions about data sourcing, worker conditions, and genuine machine capability versus human-in-the-loop systems.
- Figure AI partners with Brookfield to capture training data across 100,000 residential units, scaling human demonstration labor.
- 1X's Neo humanoid robot uses tele-operation, allowing remote pilots in California to perform household tasks, challenging notions of full autonomy.
- Training methods include workers wearing movement-tracking sensors and VR/exoskeleton rigs to perform repetitive actions hundreds of times daily.
Why It Matters
Overestimating robot autonomy risks safety, obscures new gig-work dynamics, and raises major privacy concerns for in-home devices.