Indian Workers Are Filming Their Jobs to Train AI Robots That Could Replace Them

Indian factory workers are increasingly being asked to record themselves performing routine physical tasks as artificial intelligence companies collect the video data needed to train a new generation of robots.

The recordings capture workers carrying out activities such as sewing, handling materials and other factory tasks from a first-person perspective. The footage is intended to help machines learn how humans perform physical work.

The data is known as egocentric video because it records activities from the worker's viewpoint. Developers believe such footage can help AI systems understand human movements and eventually allow robots to reproduce those actions in real-world environments.

In India's southern state of Tamil Nadu, workers have been among those participating in the emerging data-collection industry. Some wear cameras attached to their heads or use smart glasses while carrying out their normal duties.

The demand for this type of footage has created a new category of work around AI training. Data companies are recruiting workers to generate large volumes of recordings covering different human activities and working environments.

Objectways, a data-annotation company operating in India, has been involved in collecting this type of information. Its India operations include workers recording activities in factories and controlled environments for use in AI development.

The work illustrates a growing requirement in robotics that goes beyond the large language models behind chatbots. Robots operating in factories need to understand physical environments, recognise objects and learn how people move their hands and bodies while completing tasks.

For industrial applications, that could include activities such as stitching, handling materials and welding. Unlike text-based AI systems, robots must translate what they learn from data into physical movements involving tools, machinery and objects.

The development is creating an unusual relationship between workers and the technology being developed. People are being paid to generate data that could eventually allow machines to perform some of the same tasks.

For workers, the immediate activity provides an additional source of income. But the longer-term implications are less certain because the same recordings could help companies automate tasks currently performed by people.

The issue is particularly relevant in India, where manufacturing remains an important part of the government's plans for expanding formal employment. Automation could create demand for workers with robotics and technology skills while reducing the need for labour in some repetitive tasks.

The technology is not yet capable of simply replacing every factory worker. Training robots to perform physical tasks reliably remains considerably more difficult than teaching AI systems to process text, images or other digital information.

Robots also need to operate safely around people and cope with variations in materials, tools, lighting and working environments. A human worker can make rapid adjustments when a piece of fabric moves unexpectedly or a component is positioned differently.

That makes large collections of real-world human activity valuable to robotics developers. The more varied the data, the more situations an AI system can potentially encounter during training.

Companies developing humanoid and other AI-powered robots are therefore competing for access to this kind of physical-world data. Workers in India have become part of that supply chain, recording the movements that developers hope will make machines more capable.

The emerging model also raises questions about who benefits when human labour becomes the source of training data for machines. Workers receive payment for the recordings, while technology companies gain datasets that may eventually support automated production systems.

For now, the cameras and smart glasses being worn by Indian workers represent an early stage in the race to teach robots how humans work. The same footage that provides income today could help determine how much human labour factories require in the future.