TechCrunch's mobility desk this week surfaced the question the autonomous vehicle industry has been carefully managing the optics around: the hidden human cost of robotaxis. The phrase is typically deployed to mean pedestrian safety or displaced cab drivers. The less visible version of the cost is the workforce of remote safety monitors, data labelers, and edge-case reviewers whose labor makes AI driving systems look more autonomous than they are. It is a structural feature of the current AI deployment moment, not a bug that will be patched out: systems that appear autonomous are almost always hybrid, with human judgment inserted at the moments the model is least confident.
China's Robot Race and the Labor Question It Obscures
The same week, China's World Humanoid Robot Games put humanoid robots competing in athletic events at the National Speed Skating Oval in Beijing. The spectacle is real and the robotics capability is advancing quickly. What the spectacle does not show is the same hidden workforce: the engineers who reset the robots between attempts, the trainers whose motion capture data the models learned from, the quality assurance staff reviewing footage. A 2024 paper in Work, Employment and Society by Ursula Huws found that across AI industries, the ratio of visible AI output to invisible human labor input is systematically misrepresented in both marketing and journalism, with significant consequences for how labor policy responds.
The Fast Company Survey That Complicates the Narrative
A Fast Company report on AI's surprising job effects found that some workers are capturing genuine productivity gains from AI tools while others face displacement. The pattern is familiar from every automation wave: the people closest to the automated task lose, the people one abstraction level up from it often gain, temporarily. Robotaxis and humanoid robots are not replacing all human labor. They are replacing the visible, credited kind and expanding the invisible, uncredited kind. The question the industry does not want centered is who owns the productivity gains from that substitution. Soleio's thinking on AI as a taste system is useful here: when AI handles execution, the value migrates to whoever controls the decisions about what to execute. In robotaxi terms, that means the platform, not the driver, and certainly not the remote monitor in a call center watching a feed.