The human job behind AI training: meet the startup paying doctors and lawyers to grade chatbots

Why AI companies need people like this

AI chatbots don't just learn on their own — they need to be trained and checked by real people, especially for difficult, specialized knowledge. If you want a chatbot to give good medical or legal advice, someone with real expertise has to review the answers and point out mistakes. This work is often called "data labeling."

A startup called Micro1 makes money by hiring doctors, lawyers, scientists, and other experts on a contract basis to do exactly that: grade AI answers, help build training material, and even record videos of everyday tasks (like picking up objects) so that robots can learn by watching humans.

How fast it's growing

Micro1's annual revenue run rate jumped from around $100 million to $500 million in just eight months. After costs, the company actually keeps about 60-70% of that money. It's not the biggest player in this space — competitors Mercor and Handshake are larger — but the fact that multiple companies can all grow this fast at the same time shows just how much demand there currently is for this kind of human-reviewed training data.

A notable stance from the founder

Micro1's founder says the company deliberately chooses not to sell its data to Chinese AI labs, framing this as avoiding helping a geopolitical rival. That contrasts with some competitors in the same market who are less selective about their client list.

The bigger picture

Some researchers now think AI companies may eventually spend just as much money on this kind of human-generated training data as they do on the computing hardware itself. If that turns out to be true, companies like Micro1 could keep growing for a long time to come.