If robots take over the world, we will remember the humans who helped them get there. Robotaxis are no exception.
They’re not just engineers who built the AI systems that drive Waymos and Zooxes on the road. Every day, thousands of people around the world sift through reams of driving data collected by cars equipped with ungainly sensors.
Workers may be called validators, annotators, labelers, etc., but their purpose is the same. It’s about being able to understand what the AI driver is seeing.
“What they basically do is help the car understand where it is in space and time, and importantly, how the model can safely navigate through any scenario,” Rowan Stone, CEO of Sapien, a data foundry with customers such as Zoox, told Business Insider.
This task can be as simple as enabling AI to identify objects found on the road through sensors such as cameras and lidar. “Is that a cone?” A stop sign? Tumbleweed?
Stone also pointed to scenarios such as police scenes blocking roads and school buses dropping off children — real-world situations in which Waymo’s robotaxis have struggled — and said labelers are providing further guidance on how to appropriately respond.
“Obviously, we need to get humans back there,” Stone said. “Datasets need to be recalibrated, models need to be retrained with additional context, and fixes need to be deployed.”
Niche jobs in major industries
The entire data label industry has the potential to be large. Stone said Sapiens has more than 1 million “contributors” around the world.
Especially for AV systems, the number is much lower. Stone maintains less than 5,000 employees worldwide in its self-driving vehicle business. That number is likely to expand as more robotaxis emerge.
David Alfonso and a team of AI annotators discuss tagging, labeling, and classifying raw data. taskus
Omar Zoubi, vice president of TaskUs, which provides third-party data labeling and remote support agents to companies like Waymo, told Business Insider that the company has just under 2,000 employees across its AV businesses, which could double by the second quarter of this year.
Labeling itself may not be a very attractive job. At Sapien, average hourly rates are often set by the customer or AV operator and range from $3 to $6 an hour, Stone said. TaskUs does not disclose the amount paid to data labelers.
Sapien’s CEO said many of the company’s contributors are based in Germany, Japan and Southeast Asia. Overall, Sapiens’ “contributor” base spans about 100 countries, he said.
AI is taking some jobs away
Artificial intelligence is also doing some of the work of labeling data, but that doesn’t mean humans are no longer needed.
Lucas Grapentine, director of solutions engineering at Sapien, told Business Insider that the AI is “pre-labeling” parts of the raw data set, and the human’s job is to check the AI’s behavior.
Ensuring AI accuracy is especially important when dealing with autonomous vehicle systems where human lives are at risk.
“It all depends on whether you can trust that data,” Grapentine said.
TaskUs’ Zoubi said he expects the role of data labelers to evolve as automakers and AV companies acquire more data, meaning they encounter more complex driving scenarios. AI may be able to handle simpler tasks, but humans will have to take over interpreting more complex scenarios, he said.
“I believe, at least in my personal opinion, that’s where you’ll see things change.” — “We’re not just doing basic annotation and labeling of data, but we’re also identifying more root causes and fine-tuning the data to help the AV operate and navigate certain situations,” Zoubi said.
Stone makes a similar prediction. The idea is that as robotaxis better adapt to new cities and their quirks, their AI models will improve and rely on fewer humans over time.
“I think human need is on the decline, but I don’t think it’s going to go to zero,” he said.
