Indian Workers Film Daily Tasks to Train Artificial Intelligence
Hundreds of thousands globally are paid to record their actions, generating data crucial for developing robots capable of interacting with the real world.
In Bengaluru, India, workers like Mohammed Ashim are strapping cameras to their heads, not for social media, but to train artificial intelligence. Ashim, a barber for four years, now records his hand movements while cutting hair, with clients' faces blurred. His footage, along with that of thousands of others worldwide, is collected and processed by companies like Rashdan Technologies and Humyn Labs. This data is vital for training 'physical AI,' a type of artificial intelligence designed to control robots in real-world scenarios.
Workers are paid between $5 to $10 for eight to 10 hours of recorded work, a sum described as meaningful for many. Companies emphasize obtaining worker consent and ensuring privacy by blurring faces and personal identifiers. However, the demands are increasing, with clients now requesting multi-camera setups that capture a wider field of view, making data collection more complex and requiring higher pay for workers.
Manish Agarwal, founder of Humyn Labs, explains the necessity of real-world data for AI training. "Physical tasks require high-quality signals involving limb movements, fingers, wrists, elbows, shoulders and dynamic force application," he stated. Existing internet archives lack datasets of embodied human activity, such as cleaning a kitchen or assembling a car with precise force application. Companies like Humyn Labs employ 96,000 data collectors across Latin America and Southeast Asia to capture this information. Agarwal stresses the importance of sourcing data from diverse environments to ensure AI deployments are suitable for various global contexts, such as U.S. factories or homes.
Mohammed Arman, a garment stitcher with ten years of experience, views this work as a means of survival. "Every rupee I send back matters. If someone offers me extra money to wear a camera while I do the exact same work I've been doing for 10 years, I'm not going to say no," he said. Similarly, Jolly Arifur, a science teacher, supplements her income by recording domestic chores like cooking and cleaning, finding it a manageable way to earn extra money around her teaching schedule.
Concerns linger about whether this data collection could eventually lead to job displacement. Bino Paul, a professor at the Tata Institute of Social Sciences, suggests that while AI might augment some jobs, it could automate others. He advocates for government regulation to ensure AI development is guided by ethical principles and human agency, aiming for beneficence rather than maleficence. "By the time workers realise what they have participated in, things may have gone too far," Paul cautioned.
India currently serves primarily as a supplier of raw video data. However, Agarwal believes India must move up the value chain, from simple recording to processing and structuring data, setting standards for quality. As AI development continues to attract significant investment, the demand for this type of data is expected to grow, providing opportunities for workers in regions like India, provided ethical considerations and worker protections keep pace with technological advancement.