New: A weekly newsletter about the trends shaping cities

Stories

The AI sharecroppers

Slaves working in a digital field
Illustration: Aïda Amer/Axios

Invisible to most of us, an underclass of labor has evolved behind the artificial intelligence revolution — thousands of low-wage workers in the U.S. and across the globe who painstakingly inventory millions of pieces of data and images, and give power to AI programs. Critics call them the new "sharecroppers."

Why it matters: These workers — people who affix labels to data so computers can understand what it is — are starting to attract the interest of social scientists and other experts. They say labelers may at least in part explain the nagging conundrum of American income inequality — and perhaps how to fix it.

Background: We think that AI is all-knowing, but actually it's only partly so. When it comes to the AI behind driverless technology, for instance, sensors can take fantastically granular pictures of streets and hazards of all types, and AI can be fed with the experience of every type of driving situation. But autonomous technology companies still need humans to inform the AI what it's looking at — to circle things like trees, stop signs and crosswalks.

  • Without human labeling, AI is dumb: It doesn’t know a skyscraper from a spider.
  • But that doesn't mean companies pay them well. Instead, they are compensated like bottom-of-the-barrel workers.
  • In the U.S., companies say they are paying such workers $7 to $15 an hour, but that may be the top of the pay scale: Labelers also take on piecework from crowdsourcing platforms. In Malaysia, for example, the pay can be around $2.50 an hour.

The big picture: The winners are AI companies, which are mostly in the U.S., Europe, and China. The losers are workers in both rich and relatively poor countries, who are paid little.

How the companies are managing the labelers: Nathaniel Gates, CEO of Alegion, a Texas-based crowdsourcing platform, said his firm intentionally reduces the job of labeling to the simplest, most routine task possible. While this narrows a worker's chance to move up the skills — and wage — ladder, Gates argues that he is at least "opening new doors that were never available to them."

  • “We are creating digital jobs that didn’t exist before. Often times the folks doing this work are coming from farms and agriculture or factories that dried up because of automation,” Gates tells Axios.

But some experts say such practices builds inequality into the AI economy.

  • In a new book called "Ghost Work," Microsoft's Research's Mary Gray and Siddharth Suri argue that workers such as labelers are a significant part of one of the most dynamic parts of the economy.
  • "Economists don't have a handle on how to price the market," Gray tells Axios. "We've been pricing this labor as a durable good, but it's the collective intelligence that's the value proposition."

James Cham, a partner with Bloomberg Beta, the venture capital firm, thinks that AI companies are gaming the difference between the low pay to the labelers and the immense, long-term revenue from the eventual product.

  • "The companies derive benefit over a long time, while workers are paid just once. They are paid like sharecroppers, making subsistence wages. The landowners get all the returns because of how the system is set up," Cham tells Axios.
  • "It's one big arbitrage."

What's next: Gray said the market may not be able to raise the wages of labelers by itself.

  • In an age where the old political and economic rules don't hold, and societies are fraying, experts need to figure out what is a living wage for such workers.
  • What people are paid "is a moral question, not just an economic one," she said.

Go deeper: Labeling will be a billion-dollar market by 2023