Why Are Indian Workers Filming Their Jobs to Train AI?

Indian workers are strapping cameras to their heads to record their everyday tasks because global technology companies need real-world data to teach robots how to move and manipulate physical objects. By recording barbers, garment stitchers, and domestic workers, data brokers are creating training datasets for physical artificial intelligence that cannot be generated inside computer simulations.

Why are technology companies paying workers to record their movements?

Technology companies and frontier artificial intelligence laboratories need human movement data to develop physical AI, which enables robotic systems to operate in real-world environments. While text and digital images are widely available on the internet, there is no massive digital archive documenting embodied physical labor. Autonomous machines cannot learn how to wipe a counter, fold fabric, or assemble manufactured parts purely through virtual modeling.

Manish Agarwal, founder of data collection firm Humyn Labs, told BBC News that robots cannot navigate reality simply by training within computer simulations. Tasks in the real world require nuanced biological signals from the shoulders, elbows, wrists, and fingers, alongside dynamic force control. For example, a machine needs to know precisely how much grip force to apply so it does not crush a fragile plastic bottle while handling it differently than a rigid steel glass. Because datasets of people cleaning kitchens or assembling cars with tactile precision do not exist online, companies must record this physical activity from scratch across diverse working environments.

Agarwal's firm, which operates bases in Bangalore and San Francisco, works with 96,000 data collectors across Southeast Asia and Latin America. Humyn Labs sifts through thousands of hours of submitted video, filters out unusable clips, labels the actions so machine models can understand the imagery, and formats the files for corporate clients. Agarwal noted that geographic context also matters; algorithms destined for American factories or homes require data that serves as a proxy for those spaces, meaning standard Indian residential layouts may not always match the target deployment environment.

How does the video recording and data collection process work?

Workers record their work by mounting small cameras to their heads or bodies, and in some instances wearing sensory hardware on their fingers to track physical touch and pressure. Data collection brokers provide strict guidelines to ensure the footage meets the technical standards required by AI engineers. Camera angles must keep the worker's hands clearly visible within the frame, lighting must remain bright and sharp, and workers are instructed to avoid sudden, jerky head movements while performing their jobs.

Mohammed Ashim, a 22-year-old barber in Bengaluru with four years of experience, records his shifts using a head-mounted camera. Ashim told BBC News that the routine requires careful physical discipline, as the operations team instructed him to maintain consistent hand positions within the lens frame without tilting his head abruptly. The recording setup focuses entirely on his hands and fingers while automatically blurring the faces of his clients. Ashim explained that while the money is helpful, he is excited by the futuristic nature of the work, having never expected his barbering and razor skills to be purchased by international technology corporations.

The technical requirements from artificial intelligence developers are also escalating rapidly. Shahil Baig, co-founder and chief executive of data processing firm Rashdan Technologies, stated that clients are no longer satisfied with single-angle recordings. AI researchers are increasingly asking for multi-camera rigs featuring four to six cameras positioned around a person's head to capture an entire 360-degree scene rather than just hand movements. Baig noted that persuading people to wear heavy five-camera headsets is significantly more difficult than asking them to wear a single small camera, prompting his firm to offer higher compensation for multi-camera data collection.

How much are workers paid, and who is taking these jobs?

Data collection agencies typically pay workers between $5 and $10, or roughly £3.74 to £7.47, for eight to ten hours of recorded footage. For many people working in informal or manual trades across India, these payments provide a substantial supplement to their regular daily earnings. The low barrier to entry means brokers have found willing participants across a wide variety of occupations.

Rashdan Technologies gathers data across diverse industries, dispatching recording gear to agricultural sites, manufacturing facilities, and craft workshops. Baig explained that AI labs order footage of specific tasks, including pottery, candle-making, garment manufacturing, embroidery, and general household chores. For workers balancing precarious incomes, getting paid extra to perform their everyday labor is a welcome opportunity.

Mohammed Arman, who has spent 10 years working in Bangalore's textile sector cutting and stitching garments, recently began wearing a camera on his head while running sewing machines. Arman told BBC News that the extra money represents simple survival for his family back in Bihar, emphasizing that every rupee sent home matters. Because his hands operate automatically through a decade of muscle memory with needles, thread, and fabric, wearing the camera adds no friction to his daily labor while bringing in extra pay.

The work has also attracted individuals seeking flexible income at home. Jolly Arifur, a science teacher, spent seven months wearing a rig-mounted camera on her head while carrying out daily domestic chores, such as sweeping tile floors, scrubbing, cooking, and folding laundry. The arrangement allowed her to earn money on a flexible schedule while caring for her young son.

What privacy measures protect workers and customers?

Data brokers enforce privacy safeguards to ensure personal identities are removed before footage is packaged and delivered to AI labs. Because data collectors often work in public or client-facing environments like hair salons and commercial shops, preserving privacy is critical for both the worker and the public.

Firms primarily focus recording lenses on the hands, arms, and work surfaces rather than facial features. When wider rigs capture broader physical spaces, the processing pipeline blurs all bystander and client faces. Baig emphasized that Rashdan Technologies strips away names, contact numbers, and all personally identifiable information before handing datasets over to technology clients, ensuring that workers give clear consent and remain comfortable with the collection process.

Could recording manual work eventually automate these jobs?

Labour researchers warn that workers participating in these video collection programs might be training the robotic systems that will eventually replace them. While technology companies present physical AI as a tool to assist human labor, advanced robotics could eventually eliminate manual and service roles altogether.

Bino Paul, a senior professor at the Tata Institute of Social Sciences School of Management and Labour Studies in Mumbai, told BBC News that while AI could serve as a helpful aid for certain workers, it threatens to automate others entirely. Paul warned that workers may not realize what they have contributed to until the transition has progressed too far to reverse. He argued that AI should not be handled solely as a technical system, but rather as a socio-technical framework where ethical standards and human agency ensure the technology delivers positive outcomes rather than harm.

At the same time, Indian business leaders warn about the country's economic position within the global AI supply chain. Agarwal observed that India currently functions largely as a cheap supplier of raw material, where workers strap on recording hardware to produce footage that is exported to foreign research laboratories. With dozens of competitor agencies expected to launch across India this year, Agarwal warned that the financial value of raw video collection will rapidly erode. To protect its economic standing, he argued that India must move up the value chain by building systems that clean, structure, label, validate, and set quality standards for AI training data rather than merely recording footage.

Syed Aqeel — verified this story.