The Battle for Humanity’s Last Dataset

What do a Nigerian student strapping an iPhone to his forehead to film himself making his bed, an Indian seamstress wearing an action camera between her eyebrows in a noisy garment factory, and a 20-year-old Chinese man in a VR headset who opens and closes the same microwave door a hundred times a day have in common?

The Battle for Humanity’s Last Dataset They are all foot soldiers in the same booming industry — and that industry is now fully grown up. One of its biggest players is called micro1.

Founded in Palo Alto, micro1 operates a global gig platform that hands out micro-tasks to roughly 4,000 “robotic universalists” across 70 countries. Every month this small army uploads nearly 200,000 hours of first-person video straight into the system.

Robot companies — Tesla, Figure AI, Agility Robotics and others — are already paying micro1 and its competitors hundreds of millions of dollars a year for this footage. Yet CEO Ali Ansari is refreshingly blunt: “What we actually need is billions of hours.”

And they’re getting them.

The Battle for Humanity’s Last Dataset DoorDash now pays its delivery drivers extra to keep cameras rolling for every step they take and every chore they do at home. Scale AI and Encord are building their own armies of data collectors.

In China there are more than 40 specialized “robot schools” where trainers wearing exoskeletons physically guide humanoid robots through everyday tasks — teaching them how to wipe a table the way only a human body knows how.

If this sounds like a dystopian scene in which flesh-and-blood humans are training their own mechanical replacements, that’s because it is.

But the story goes deeper.

The Battle for Humanity’s Last Dataset In 1958 the philosopher Michael Polanyi introduced the concept of tacit knowledge — the kind of knowing that lives in the body rather than in words.

You can ride a bicycle perfectly, yet writing down the instructions is maddeningly difficult.

A seamstress can feel the exact weight of fabric between her fingers and flick her wrist at precisely the right microsecond — but if you ask her to explain it, she’ll just shrug and say, “Like this.”

For years, AI feasted on the easy stuff: neatly digitized text, labeled images, and carefully curated internet scrapes. That was the “low-hanging fruit” dataset. Now the industry has reached the final frontier — the last dataset that can only be harvested one way: by paying real human beings to live their real lives with cameras glued to their heads.

The Battle for Humanity’s Last Dataset And notice who’s wearing the cameras. Not office workers in San Francisco or Berlin. It’s people in Lagos, in Dhaka, in provincial Chinese cities. The mundane daily life of the Global South is being turned into raw material for the algorithms of the “golden billion.” Yesterday it was rubber and sugarcane. Today it’s videos of opening microwaves.

Critics have already coined the term: data colonialism.

You can smirk at the image of Indian seamstresses selling their irreplaceable tacit knowledge for $100–200 a month. But before you laugh too hard, remember this: every keystroke I typed in 2022 was also silently harvested, packaged, and fed into large language models. I received exactly zero dollars for it.

So I’m not entirely sure who’s supposed to be laughing at whom.

The battle for the last human dataset is not coming. It is already here — and it is being won one forehead-mounted iPhone at a time.