Thousands of Humans Teach Robotaxis to Read Stop Signs for Under $6 an Hour

KEY POINTS

•Rowan Stone, CEO of Sapien, explained that thousands of data labelers help robotaxis understand real-world objects and scenarios they struggle with.
•TaskUs currently employs just under 2,000 labelers worldwide for autonomous vehicle operations, with plans to double that by mid-2026.
•Labelers often earn between $3 to $6 per hour, working remotely from about 100 countries to annotate sensor data from clunky cameras and lidars.

Behind every Waymo and Zoox rolling gracefully—or maybe not—are a few thousand humans labeled as annotators, validators, and the ever-glamorous labelers, painstakingly telling AI what a cone is versus a tumbleweed. This inspiring workforce, capped under 5,000 globally but spread through a whopping 100 countries including Germany, Japan, and Southeast Asia, earns a dream rate set between $3 to $6 per hour, as revealed by CEOs Rowan Stone of Sapien and Omar Zoubi of TaskUs. These underpaid legends meticulously clean data from clunky sensors and cameras, battling confusion over police scenes and school buses, to prevent robotaxis from committing digital jaywalking. AI’s doing some ‘‘pre-labeling’’ homework but humans must still approve or the car might think a school bus is a spaceship. As automation gets smarter, Stone predicts fewer humans but isn’t fool enough to see them reaching zero, acknowledging that even this robot revolution has the IRL equivalent of unpaid interns.

Share the Story

(1 of 3)

Mockingbird News

@MockingbirdNews · now

Robotaxis can’t drive themselves without humans playing thousands-of-hours of ‘Is that a cone or just junk?’ for $3-$6 an hour. Meanwhile, AI takes coffee breaks.

179/280 charactersReady to post