(Bloomberg Businessweek) -- Behind every self-driving car or cashier-less Amazon Go convenience store sit thousands of humans whose job it is to train computers to see. These people look at pictures and identify what’s in the footage, labeling something as a truck or a bag of Doritos. Their observations are fed back into artificial intelligence software that then learns how to do the same thing over time. It’s the drudgery behind the magic.

A three-year-old startup called Scale AI Inc. has been trying to improve this process on behalf of both man and machine. It’s built a set of software tools that take a first pass at marking up pictures before handing them off to a network of some 30,000 contract workers, who then perform the finishing touches. Scale has attracted big-name customers in the self-driving car field, including Alphabet Inc.’s Waymo, General Motors Co.’s Cruise, and Uber Technologies Inc.

Now, Scale is looking to sell its wares to just about any company developing AI technology. Several high-profile venture capitalists are sold on the mission. On Aug. 5, Scale plans to reveal an investment that values the business at more than US$1 billion. “It takes billions or tens of billions of examples to get AI systems to human-level performance,” says Alexandr Wang, Scale’s co-founder and chief executive officer. “There is a really big gap between the handful of giant companies that can afford to do all this training and the many that can’t.”

Even by Silicon Valley standards, Wang is something of a phenom. He grew up in New Mexico, the son of two physicists. During his teenage years, Wang excelled at coding competitions and got job offers from tech companies as a high schooler. This put him on a path to graduate early, work in Silicon Valley, and then start Scale by the time he was 19. Now, at the ripe old age of 22, Wang has a fresh $100 million from investors, including Mike Volpi, a general partner at Index Ventures. “When we signed the term sheet and went out to dinner, I ordered a nice bottle of wine to celebrate,” says Volpi, “and then had to ask him if I’m breaking the law.” (Wang was indeed of legal drinking age by then.)

As companies race to build AI systems on par with the likes of Google’s or Facebook Inc.’s, they face two major challenges. One is getting enough data to train the machines. The other is making sure the data and results are good. And while machines can do much of this work, it really takes people to interpret photos, text, and video to point the computers in the right direction.

In the autonomous-car industry, companies spend millions of dollars each year hiring people to label the pictures gathered from cameras in their vehicles. Typically, a worker sees an image pop up on a computer screen and uses a mouse to trace the outline of all the cars and categorize them in the software. Then come buildings, parking spaces, pedestrians, traffic lights, and so on. It can take anywhere from 10 minutes to a couple of hours for a person to go point by point over every object in a single photo, and there are millions of images to scour. That data is then fed back into an AI system, so the cars can learn what things are in the world around them.

Scale has built software that looks over the images first. In many cases, it’s able to label most of the objects automatically. Workers are then asked to review the images. If they need to intervene, the system lets them click once somewhere, say, in the middle of a car, and it traces the object for them. “Tasks that used to take hours end up taking just a couple of minutes,” says Wang.

Scale has about 100 employees working at its San Francisco headquarters, in addition to an army of contractors scattered all over the world handling the image labeling work. The contractors receive detailed instructions from Scale on what they’re meant to look for. The company is also developing software to identify the best labelers. Wang wouldn’t say exactly where the workers are contracted from or how much they earn, but he insists the pay is good. “We are not trying to optimize the human cost,” he says. “They earn in the 60th to 70th percentile of wages in their geography.”

Newer customers of Scale include OpenAI, a research company that uses the service for language processing, and Standard Cognition, which is building software to automate the checkout process at retailers similar to Amazon Go. Standard Cognition has a test store in San Francisco, as well as research centers where people pick objects off shelves under video surveillance. “The ultimate question we have is, ‘Is this ketchup, or is it mustard?’ ” says Jordan Fisher, the CEO. “And if it’s ketchup, we need to know if it’s 12.6 ounces of Heinz ketchup, so we can get you the right receipt.”

There’s plenty of competition for this type of work. In June, Uber acquired the labeling automation startup Mighty AI. Amazon.com Inc. offers automated data labeling services as part of its cloud product, and startups such as Hive and Alegion also do similar stuff. Hive CEO Kevin Guo says data labeling is just a low-end part of his company’s business and that the real money is in building actual AI models for customers in different industries. “I did not center our company around labeling because I don’t think it will be that big of a business,” he says. “There’s quite a lot of these companies these days, and honestly I don’t think there’s much differentiation between them.” Scale investors, which include Accel and Peter Thiel’s Founders Fund, say Wang’s tools are more advanced and can label data faster and more cheaply.

As for the human cost behind the work, Volpi of Index Ventures defends the drudgery as inevitable. “If you could be pulling a rickshaw or labeling data in an air-conditioned internet café, the latter is a better job,” he says. “You’re paid better, and it’s not as stressful to your body.” And if someday automation comes for venture capitalists? “I can make peace with that,” he says. “I’ll probably have to do something that is higher value to society.”