The Indian workers filming their work to train AI

Indian workers are being paid to film themselves at work and the footage is used to train artificial intelligence.

The full report

Mohammed Ashim, a 22-year-old barber in Bengaluru, has recently begun strapping a camera to his head to record his work. While the routine involves cutting hair and trimming beards, the process requires strict discipline regarding hand movement, lighting, and framing. His footage is collected by local companies that sell it to technology firms for training physical AI systems capable of controlling robots in real-world environments.

Rashdan Technologies, a company involved in collecting this human-generated data, pays workers between $5 and $10 (£3.74; £7.47) for eight to 10 hours of recorded work. The company emphasizes that consent is critical, noting that while one camera recording hands is relatively easy to arrange, multi-camera rigs capturing the entire body or environment are significantly more difficult to secure due to privacy concerns. To protect workers' identities, companies blur faces and exclude names or contact details from the data handed off to clients.

The demand for such footage is expanding as robotics companies seek data on industrial tasks, agriculture, household chores, and small business activities like pottery and garment-making. Manish Agarwal, founder of Humyn Labs, notes that there is no existing internet archive of embodied human activity. His firm employs 96,000 data collectors across Latin America and Southeast Asia to capture high-quality signals involving limb movements and dynamic force application, such as applying the correct grip force to avoid crushing objects.

Despite the financial incentive, experts warn that workers may not fully understand the implications of their participation. Bino Paul, a senior professor at the Tata Institute of Social Sciences in Mumbai, cautions that by the time workers realize they are training machines designed to automate their jobs, it may be too late. He argues that AI must be viewed as part of a socio-technical system where ethics and human agency are upheld to ensure beneficence rather than maleficence.

For many, the work is a matter of survival or additional income. Mohammed Arman, a textile worker in Bangalore with ten years of experience, accepts the extra payment because he already performs the stitching and cutting. Similarly, science teacher Jolly Arifur records her domestic work from home to supplement her teaching salary, earning two to three dollars per hour. However, industry leaders argue that India must move up the value chain from simply supplying raw video footage to owning the pipeline that cleans, structures, and validates this data.