Inside the race to train robots on human work
Thousands of workers are recording everyday tasks to create training data for robots that could staff factories, clean homes, and even fight wars.
On a Friday evening in summer, a private chef named Aliser Demir presented a three-course meal to two friends at reporter Tyler Le's apartment. Demir, who runs a private chef business, was participating in Shift, the data-collection arm of German startup Microagi. Shift offered to pay Demir his usual rate of $100 to $150 per person in exchange for recording footage of his work. Demir expressed that he has always enjoyed serving people and now also feeds AI.
This story is part of a series titled "The AI Data Grab," which explores how tech giants are increasingly turning to workers, users, and the physical world to gather data for AI training. Demir is involved in a global effort to collect data on human interactions with the physical world, aiming to create machines capable of working in factories, homes, and even in warfare.
Elon Musk, who believes Tesla's future lies in humanoid robot Optimus, envisions a future where AI and robots make human labor largely unnecessary. However, a significant bottleneck in this endeavor is the scarcity of training data. Kyle Chan, a Brookings Institution fellow, agrees that data is the main obstacle in the robotics world.
Shift collects data in over 15 countries, paying more than 35,000 people to record various physical tasks, from cooking to repairing engines. Contract workers in Mexico and the Philippines control robots in American warehouses, recording their movements to enable machines to eventually perform these tasks independently. In the Bay Area, a company called Encord offers a "data factory" where workers remotely control robots to generate training data.
The demand for robot training data has surged, with Encord's customers requesting 10 times more data than a year ago. Robotics companies have attracted significant investments, with $33.4 billion raised globally in the first half of the year alone. However, despite impressive robot demos, the challenge remains in creating reliable machines capable of working in an eight-hour factory shift.
To overcome this hurdle, robotics companies are seeking more training data, turning to suppliers such as Scale AI, Instawork, and Uber, as well as specialized firms like Mecka AI.
Written by urgent.news from Business Insider's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
