Robot labs buy workers’ hand movements to break real-world data scarcity

In the Indian city of Bangalore, 22-year-old barber Muhammad Asim spends about six hours a day fixing a camera on his head while cutting hair and trimming beards, careful to keep his fingers within the frame, adjust lighting and avoid moving his head abruptly, while the software completely obscures customers’ faces. Asim joins hundreds of thousands of workers worldwide who earn daily wages to record their precise movements for intermediary companies that collect these recordings and sell them to AI laboratories. This shift reveals a silent dilemma facing developers of advanced models: training robots to operate in the tangible world has hit a dead end inside digital simulation, and the only way to break this barrier is to purchase motor skill from human hands hour by hour.

The need for this type of recordings is directly linked to what is calledphysical artificial intelligencePhysical AI, the mode capable of directing robots and machines in real-world environments. Manish Agarwal, founder of Human Labs, which oversees 96,000 data collectors in Latin America and Southeast Asia and is based in Bangalore and San Francisco, explains that the internet contains no historical archive of embodied human activity; there are no open databases that teach a machine how to clean a room, or how to assemble car parts with hands that sense the exact amount of kinetic pressure required. Physical tasks demand complex signals that include limb, finger, wrist and shoulder movements, and the application of appropriate dynamic force, such as knowing how much grip is needed to hold a plastic cup without crushing it compared with a steel cup, data that cannot be extracted from text or static images.

This acute shortage has pushed intermediary firms like Rishdan Technologies to expand rapidly. Its chief executive and co-founder Shail Big says his company chases tasks wherever they appear for leading robot firms and AI laboratories, gathering data that cover industrial work, agriculture, household affairs, as well as handicrafts such as pottery, embroidery and candle making. The company pays between $5 and $10 for a work session lasting 8 to 10 hours, with faces blurred and names and contact details removed to ensure workers’ consent and protect their identities. However, client demands are escalating technically; partners are now requesting helmets equipped with four to six cameras surrounding the worker’s head to capture the scene from all angles, which entails higher compensation for the additional physical effort.

While workers such as Muhammad Arman, who spent ten years sewing and cutting textiles in Bangalore, see wearing the camera as a means to stay afloat and secure extra remittances for their families in Bhopal, and science teacher Julie Arefour views filming household cleaning and cooking tasks as a source of additional income of two to three dollars per hour, academics warn of the long-term cost of this model.Binu Paul, professor of labor management at the Tata Institute of Social Sciences in Mumbai, points out that workers may participate unknowingly in training machines that will later swallow their jobs.He calls for regulations that place ethics and human effectiveness at the core of the technical system. Moreover, Agarwal stresses that global South markets cannot remain merely suppliers of raw digital material exported abroad; they must move up the value chain to own data cleaning, structuring, auditing and the formulation of quality standards.

We see in this shift a strategic imperative that is imposing itself on institutions and tech teams in the Arab region. Entities that pour massive investments into automating supply chains, logistics warehouses and infrastructure projects in the Gulf cannot rely on imported robotic models trained exclusively on footage recorded in Indian homes or Asian factories; data realities confirm that residential and industrial environments in those countries do not serve as geographically precise proxies for the needs and construction and operational characteristics of regional facilities. The decisive difference for digital-transformation leaders will not be limited to importing mechanical arms or buying computing power, but lies in building local auditing pipelines that gather motion data from regional work environments. Conversely, this development signals an early cue for software developers and startups in Egypt and the Levant, as the digital-outsourcing market shifts from superficial text classification to motion-data engineering and sensor-input tuning, the space that will determine the actual added value in the next AI supply chain.