Pro-Worker Artificial Intelligence: Reshaping the Direction of Technology

In April 2026, a video showing hundreds of workers at a garment factory in Gurugram went viral. Seated at their sewing machines, with head-mounted cameras, workers laboured on. The cameras recorded every stitch, finger calibration and hand movement. Some of them work twelve-hour shifts, six days a week, with little or no extra pay for the first-person footage of the tasks that define their work.“They are recording our work to prepare and train a robot, which is making us worried…it will mean job losses,” said Ravi Kumar, a factory worker from Delhi. Another worker from Nagpur stated, “I’m working in my own grave.”

Real human labour, contributing to large behaviour models (LBMs) and helping train AI-powered humanoid robots that could one day replace them. This is not a one-off occurrence but a lived reality for thousands of blue-collar workers across India. Investigative reports conducted by The Guardian, DW, and the Indian Express highlight at least five states where the daily routines of workers are captured from factory floors. Workers from Delhi, Gurugram, Nagpur, Bengaluru, and Chennai are all contributors to an ever-growing pool of egocentric data. Data, critical for creating LBMs and developing humanoids.

The prize?

A market size of an astronomic 5 trillion dollars for humanoids by 2050, according to one Morgan Stanley report.  One billion humanoid robots could be in use by 2050. Most importantly, one could avail it for as low as $15,000 by 2050. However, terms and conditions apply. The technology must advance at an accelerated pace, production volumes must increase, and supply chains should be predominantly Chinese.

The costs? 

Several, but in this particular scenario, physical labour, the intellectual property of informal workers, privacy, consent, collective harm and possible job loss.  Added costs include constant replacement anxiety, fears of workplace surveillance, and extreme pressure from real-time monitoring.

All of this, however, begs one simple policy question. 

Is the direction of technology a choice? 

Nobel laureates Daron Acemoglu and Simon Johnson stipulate that, "Better technology offers the potential of better lives for many people, but only if it creates sufficient new tasks requiring human expertise and workers have enough power to demand a matching increase in compensation.” 

In their book, Power and Progress: Our Thousand-Year Struggle over Technology and Prosperity, the writers highlight that throughout history, there have been several episodes of technological progress that produced “no improvements” and led to “even significant deterioration”. According to them, if the market were to be left to its own devices, it would deliver great wealth to aggressive investors and executives and leave out the rest.

In the past four and a half decades, much of the industrial world and the US have witnessed rising earning inequality, stagnation in earnings among non-college workers, and an ever-shrinking share of labour in the national income. Data from the manufacturing sector, reflecting both US and global trends, show the precipitous drop in labour’s share of value added from 74 per cent in 1981 to 46 per cent in 2016. A comparable contraction, though less severe, has been observed in the share of labour in national income both globally and in the US. 

According to Acemoglu, Johnson and David Autor, if the current tech sector efforts on “even more automation” hold, the pain for workers “will be magnified". They caution, "Really bad outcomes are possible when deluded technology leaders are able to impose their messianic visions on society.”

The direction of technology is a choice. Striking a balance between automation and technologies that create new tasks is critical not only for stable wages and job creation but also for managing the cascading web of social consequences and economic disruptions it can trigger.

“Pro-worker Artificial Intelligence (AI)” and technologies "that make human skills and expertise more valuable by expanding worker capabilities" can provide a framework for intervention. 

What is pro-worker AI?

By acting as a catalyst for new task creation, “Pro-worker AI” cultivates demand for novel manual proficiency. Consequently, such an approach can mitigate the wicked social challenges posed by job loss, rising inequality, and premature mortality.

Pro-worker AI acts as a force multiplier by combining human skills with technical expertise. They can enable workers to carry out “more sophisticated tasks related to their jobs”, “tackle new tasks” and also “acquire new expertise”. AI’s ability to learn from examples by harnessing knowledge from unstructured information makes it a critical tool for augmenting worker capabilities. 

Technologies are generally classified under five expansive and economically discrete categories as provided by Acemoglu, Autor and Johnson (2026). They can be labour-augmenting, capital-augmenting, automation-orientated, expertise-levellers or new-task creators. “New task-creating technologies” are unambiguously good for workers. They have the potential to expand what humans can do by creating entirely new tasks that did not exist a generation earlier, thus generating a demand for novel human expertise. 

History offers a parallel precedent for pro-worker technologies. In the 19th century, as the industrial revolution lowered labour share in the textile industry, the impact of automation was counterbalanced by expanding roles in manufacturing, railways, clerical and accounting roles. This trajectory manifested once more during the four decades post-World War II.

Pro-Worker AI in Practice

The “electrician’s assistant (EA)" is a tangible example of pro-worker AI. The tool leverages the expertise of electricians while providing real-time assistance from a large language model. By working collaboratively with the technician, it ensures efficiency with reliability, simultaneously allowing them to accomplish more challenging tasks. The tool not only increases the effective supply of workers but also boosts technical proficiency, meeting the definition of pro-worker AI.

“Teacher’s AI-aid" is another example of pro-worker AI. The support tool, trained on curriculum material, student preferences and real-time inputs from the class, can work collaboratively with the educator. Such collaborative paradigms can be extended to support workers in different domains like construction, plumbing, healthcare, team leadership, training and supervision. 

However, despite its obvious potential for good, pro-worker AI has not been a mainstream phenomenon. This is largely driven by, first, the commitment to the overarching ideological vision of artificial general intelligence (AGI) and, most critically, market failures for firms deploying AI as well as its developers.

The Way Forward

Moving towards pro-worker AI calls for a recalibration of the pro-automation and pro-AGI mindset that currently dominates the tech market. Acemoglu, Autor and Johnson,  through their research, provide policy levers that can significantly shift the needle towards pro-worker AI. Within India’s blue-collar workers in the textile sector, two become absolutely critical. 

Discouraging the theft of expertise by firms that use their own employees to train AI models through legal frameworks and amplifying the perspective of these workers through institutional mechanisms will help prevent a future that is inherently anti-worker. 

Imposing limits on a firm’s ability to harvest their employee’s human expertise, by providing workers a decisive say over how their data is used, will help limit the commodification of human knowledge. 

Finally, at the institutional level, civil society organisations, especially labour unions, should champion voices of frontline workers and advocate for protective strategies at the local, state and central levels.

Striking a balance between automation and collaboration, coupling human judgement and AI capabilities will be critical as the technology matures.