GoPro Workers Training the Robots That Replace Them · Issue #150

Opening

Dear reader, picture a textile factory in Karur, a small town in Tamil Nadu, India. It’s an ordinary workshop—affixing labels, ironing cloth bags, folding clothes. But there’s one strange thing: about 8 workers are doing this work with GoPros strapped to their foreheads or smart glasses on their faces. This is a scene an AFP reporter witnessed firsthand.¹

One woman working there is 28 and quit her job as a teacher after giving birth. What she does now is simply film her daily routine—folding clothes, washing dishes, tidying her child’s toys—while wearing a camera for 6 to 7 hours a day. In exchange, she earns an income. But this data will one day be used to train a robot that replaces the very work she’s doing.

Watching this scene, I found myself thinking of China’s brain-computer interface (BCI)² work, which I covered in the last issue. Let me cut to the conclusion: the two countries are climbing the same mountain—AI—but they’re gripping completely different layers of it. China is trying to have the state own humanity’s innermost layer, the brain, while India has 1.5 billion people building humanity’s eyes, only to watch that ownership flow outward. Today, let’s talk about why this “location of ownership” is the decisive variable separating exploitation from liberation.

China Decided the State Would Own the ‘Brain’

As I discussed in the last issue, China’s choice is clear: it’s targeting the innermost layer—the human neural signal itself.

In July 2025, China’s government—seven ministries, including the Ministry of Science and Technology—jointly released a blueprint for cultivating the BCI industry. It’s a state-led, unified playbook that bundles industrial planning, medical regulation, and research oversight into a single document. The goals are concrete, too: achieve breakthroughs in electrodes, neural chips, and signal-decoding algorithms by 2027, and cultivate 2 to 3 “world-class” companies by 2030. In December 2025, China also set up a brain-science fund worth ¥11.6 billion (about $165 million).³

What I find notable here is what China is trying to own. Electrodes, chips, algorithms—all hardware and intellectual property. Once built, they remain assets of domestic firms, things the state can control. China is nailing down the deepest, most defensible layer of the AI stack as a national asset.

Of course, this path has its limits. Invasive technology that implants electrodes in the human brain faces high clinical and ethical barriers, and the market is opening slowly. China’s BCI market is estimated at only around $500 million even in 2025. But the direction is clear: inward, and into the hands of the state.

India Has 1.5 Billion People Building the ‘Eyes’

India’s path is the exact opposite: eyes instead of a brain, secondhand GoPros and smartphones instead of invasive surgery, purchase orders from foreign companies instead of state funds.

The company collecting data in Karur is Objectways. Founded in 2019 as a data-annotation⁴ firm, it has since expanded into robotics data. It has offices in India and the United States, counts Fortune 500 companies as clients, and works with machine-learning platforms like Amazon SageMaker. Its CEO, in their 50s, is originally from Tamil Nadu but now based in the US.¹

What they’re collecting is “egocentric data”⁵—in plain terms, first-person footage of ordinary human life. Folding clothes, making coffee, washing dishes, even cleaning the bathroom. This full A-to-Z of hand movements and field of view becomes the basic raw material for training humanoid robots. For a robot to learn “this is a cup,” it’s not an engineer in San Francisco who needs to show it thousands of times—it’s someone in a small Indian town. Morgan Stanley has projected that more than 1 billion humanoid robots will be in use worldwide by 2050.

As a fact-checker, let me flag one thing here. The footage claims the industry expects “500 million hours of data a day.” I find this figure hard to take at face value—the source is unclear, and physically producing 500 million hours of footage a day would require tens of millions of people filming simultaneously. Read it as rhetoric revealing the industry’s ambition, not as a verified fact. What is confirmed, though, is this: according to AFP’s reporting, workers earn about ₹250 (about $2.6) per hour of footage.¹ Another worker in the footage said she receives a separate ₹10,000 monthly stipend just for wearing the camera.

Here’s the core point: Indians “produce” the data, but they don’t “own” it. The eyes that get built become the property of the foreign company that placed the order.

The Intersection: Ownership Sits Inside, Labor Sits Outside

Now let’s overlay the two countries. When I’ve built go-to-market strategies, I’ve always asked the same question: “Who holds the defensible asset in this value chain?”

Viewed through this question, the difference between China and India comes into sharp relief.

China has the state own the innermost layer of the stack—the brain-and-chip layer, which is hard to replicate and faces high regulatory barriers. High barriers to entry mean that once you’re ahead, others can’t easily catch up. India, by contrast, holds the outermost layer of the stack: raw data flowing in through the eyes. This can be produced cheaply at the overwhelming scale of 1.5 billion people, but at the same time, anyone can replace it, and ownership leaks outward.

The same logic is repeating itself in India’s IT services industry. India’s software services exports exceed $230 billion as of FY25,⁶ an achievement built up over 25 years of serving as the world’s “back office.” But that very repetitive work is precisely what large language models (LLMs) are devouring first. Because India provided only labor without owning a defensible asset, it’s the first to wobble when the rules of the game change.

The market is reading this structure coldly. In 2026, foreign capital has pulled more than $30 billion out of Indian equities—an all-time record.⁷ That money is flowing toward Taiwan and South Korea, which directly manufacture chips and hardware. One investor has bluntly called India an “anti-AI play.” Of course, there’s a counterargument. India has more than 1,800 Global Capability Centers (GCCs)⁸ employing about 2 million people, and 80% of new GCCs prioritize AI and machine learning. This view holds that India is a hub for fine-tuning AI applications for enterprises. But even this counterargument sits within the grammar of labor, not ownership, given that it amounts to “tuning someone else’s model.”

Expanding the Lens: Why the Same Camera Is Both Liberation and Exploitation

But here’s a twist. Looking only at the ownership structure, India’s picture looks bleak. Yet if you ask the woman in Karur what this work means to her, the story isn’t so simple.

India is a socially conservative country where even girls’ education is sometimes abandoned because it’s seen as “a waste of the dowry fund.” In a place like this, a job that lets you earn an income without leaving home changes an entire life. One woman in the footage introduced herself as the first in her family to attend college, saying her salary covers her parents’ health on one side and her child’s education on the other. There’s an Indian proverb: “Educate a boy, and you educate a family; educate a girl, and you educate an entire village.” Here, that’s not a metaphor—it’s reality. This is unmistakable liberation.

At the same time, this is a supply chain designed for extraction. Workers are training the very technology that will replace them, and that data can be used against them at any moment. In some regions, there’s a risk that employers force workers to wear cameras with no additional pay at all. The claim that “if someone willingly does it for pay, it’s a fair trade” and the question “aren’t we still being exploited?” coexist within the same workshop.

The key to resolving this paradox is the first axis: ownership. Without owning the data, liberation is temporary and exploitation is structural. The moment the model finishes learning and the robot becomes smart enough, this job’s “safety window” closes. The woman in the footage knew exactly this. “Until the model is trained,” she said, “the robot can’t take our jobs.” The shelf life of liberation is tied to the location of ownership.

Oz’s Lens

Honestly, I’m cautious about India’s narrative that “population itself is competitiveness.”

There’s a pattern I’ve often encountered doing data analysis: scale, on its own, isn’t value. For scale to become value, someone has to own the asset that scale creates and channel it into reinvestment. Scale without ownership is just cheap supply. Even if 1.5 billion people build the eyes, if ownership of those eyes sits elsewhere, all that remains in India is wages—and even those wages come under threat the moment the robot finishes learning.

This doesn’t mean China’s path is the right answer. A model where the state owns the brain carries its own problems of control and ethics, and concerns that AI is inherently a power-concentrating technology are, if anything, even stronger on China’s side. What I want to offer is a frame: when you look at any country’s AI strategy going forward, ask not “how much are they doing” but “which layer do they own.” That question will show you far more quickly whether a country stands on the side of liberation or exploitation.

Closing

Let me compress today’s argument into three lines.

First, China has the state own the inner layer—the brain (BCI and chips)—while India has 1.5 billion people building the outer layer—the eyes (vision data)—without owning it. Second, camera labor in Karur is both liberation, in the form of independence without leaving home, and exploitation, as the tail end of an extractive supply chain training the very robot that will replace the worker. Third, what ultimately determines the weight of this paradox is who owns the data.

So the next time you see news about any country’s “AI rise,” I’d suggest checking the location of ownership instead of just scale and speed. The picture will look different.

If you were in the shoes of that woman in Karur, would you see this work as “liberation” or as “exploitation”? And what conditions would tip that judgment one way or the other? Let me know in the comments which of the two axes you weight more heavily—I’ll factor it into how I choose material for the next issue.

💬 Is camera labor in Karur liberation or exploitation? Leave a short comment on what conditions tip your judgment · 📨 If this perspective was useful, please pass it on to a colleague

References & Further Reading

Primary sources

AFP, “The Indian workers training AI robots to take their jobs”, June 2026. Published on Al Jazeera ··· The primary source on GoPro labor at the Karur factory, the ₹250 hourly wage, and information about Objectways.
China’s seven ministries, “Implementation Opinions on Promoting Innovation and Development of the Brain-Computer Interface Industry”, July 2025. TechCrunch analysis ··· Good background for understanding why China wants the state to hold the “inner layer.”
Nasscom, “Technology Sector Strategic Review 2025”. India Business Trade summary ··· The basis for the scale of India’s software services exports.

Background

Ruchir Sharma, “India is a loser in the AI race” interview, May 2026. Business Today ··· Offers an investor’s view on why foreign capital is leaving India for Taiwan and South Korea.

Related past issues worth reading

OZ Talking: the China BCI edition ··· Connects directly to today’s “China owns the brain” axis. Reading both together sharpens the contrast in ownership structures.

📝 Glossary

The author, Kwangseob Ahn, is a professor of business administration at Sejong University and lead consultant at OBF (Oswarld Boutique Consulting Firm). He teaches statistics and data analysis — business data management and business analytics — while leading GTM and AI strategy consulting in the field, designing the seam between technology and business. He has published academic research on a memory architecture for AI dialogue systems (HEMA) and runs Daily Arxiv, a daily curation of global AI papers. He holds a master's from Korea University's Graduate School of Technology Management and a KMBA. He is the author of Homo Brainless: The People Who Outsource Their Thinking.

Footnotes

Objectways / Camera Labor in Karur: A case AFP reported firsthand at a textile factory in Karur, Tamil Nadu, in June 2026. Workers wear GoPros or smart glasses on their foreheads to film their everyday movements, and that footage is sold as training data for humanoid robots. ↩ ↩² ↩³
BCI (Brain-Computer Interface): Technology that reads neural signals from the brain and connects them to computers or machines. It splits into invasive types, which implant electrodes in the brain, and non-invasive types, which measure signals from outside. ↩
¥11.6 Billion Brain-Science Fund: A roughly $165 million fund China established in December 2025 to support the BCI industry. It’s state capital backing domestic firms from research through commercialization. ↩
Data Annotation: The work of labeling data—images, video, audio—with tags like “this is a truck, this is a person” so AI can learn from it. This is also the work behind self-driving cars’ ability to distinguish objects on the road. ↩
Egocentric Data: Data shot from a “first-person point of view.” Filmed from the height of a person’s head and eyes, it’s training material that helps robots directly mimic human hand movements and field of vision. ↩
Software Services Exports (FY25): For India’s fiscal year (April 2024–March 2025), software services exports exceeded $230 billion. The US (50%) and Europe (31%) are the largest markets. ↩
$30 Billion Capital Outflow: The amount foreign investors withdrew from Indian equities in 2026, already surpassing 2025’s full-year record—an all-time high. ↩
GCC (Global Capability Center): An in-house operational hub multinational companies have established in countries like India. It’s a form that brings work once outsourced (BPO) back in-house; India has more than 1,800 of them, employing about 2 million people. ↩