Human actions sold for $13/hour to train robots, costing $118/hour for robots

A growing trend sees companies paying individuals to record their everyday actions, from household chores to complex professional tasks, to train artificial intelligence and robots. U.S. company Micro1, for instance, has been acquiring first-person perspective videos of domestic labor and daily activities submitted by thousands of people in over 70 countries. The company recently recruited "gig workers" in the U.S. to sell videos of their household chores for $0.21 per minute, or $13 per hour, under the title "remote household task video contributor." Micro1 promotes this data as "high-quality human data" that drives advancements in robotics for factories, homes, and warehouses. The company uses AI agents to review the video quality, rejecting submissions that don't meet standards. This model allows Micro1 to rapidly grow by converting human work into valuable robot training data for sale to other corporations. Beyond domestic tasks, companies are also hiring human experts worldwide for remote work to refine AI models' reasoning abilities. These roles span hundreds of fields, including skilled trades like warehouse workers, truck drivers, and electricians, as well as service professionals, artists, bilingual specialists, philosophers, historians, and even doctors. Rates can reach hundreds of dollars per hour, with former detectives or investigators, for example, being paid $50-$100 per hour to analyze case files and provide insights on investigative techniques for AI models in the legal and justice sectors. Robot developers like Boston Dynamics are also pushing the boundaries of AI training. Their humanoid robot, Atlas, undergoes millions of hours of parallel training in virtual environments, simulating countless repetitions of tasks such as lifting and carrying heavy objects like refrigerators. This extensive virtual training allows Atlas to learn optimal movement and energy efficiency. The next phase involves physical testing at Hyundai Motor Group's Metaplant America in Georgia, where robots will practice and learn tasks in a production line environment. These robots are slated for deployment in automotive production starting in 2028, eventually handling assembly by 2030. The vast amount of operational data generated by these robots will then feed back into further AI training, creating a continuous loop of improvement. Experts predict that the market for such real-world robot action data will be substantial, with Hyundai potentially generating significant profits from it by 2030.

“

High-quality human data is the driving force behind advancing robotics in factories, homes, and warehouses.