Hidden labor behind humanoid robots

The chief executive of Nvidia, Jensen Huang, recently argued that artificial intelligence is entering a new phase. According to Huang, the next frontier is “physical AI”, in which intelligent systems move beyond language models and chatbots into machines capable of acting in the real world. Demonstrations of humanoid robots loading dishwashers or assembling products reinforce this narrative, suggesting a transition from traditional industrial robot arms toward machines that learn and adapt more like humans. Yet the technological spectacle conceals an important reality. Many of these systems still rely heavily on human labour for training and operation, a fact that remains largely invisible to the public and therefore contributes to inflated perceptions of machine autonomy.

The emerging ecosystem around robotics reveals new forms of work that resemble science fiction scenarios. In one reported case in Shanghai, a worker spent days wearing a virtual-reality headset and exoskeleton while repeatedly opening and closing a microwave door to generate training data for a robot. Similar initiatives appear in North America, where the robotics start-up Figure AI has announced plans with Brookfield Corporation to collect large volumes of data from real household environments. At the same time, companies increasingly rely on tele-operation, allowing human workers to remotely control robots when tasks prove too complex for automation. Firms such as 1X Technologies are preparing to deploy humanoid robots in homes, yet human operators may intervene through the machine’s cameras when autonomy fails. As robotics expands into everyday environments, the central challenge will therefore not only involve technical capability but also transparency. Without greater clarity about the human labour embedded in these systems, society may continue to mistake concealed human effort for genuine machine intelligence.