Gig worker Crystal Kaufman said she spent a weekday looking at images captured by a camera placed on a baby's head, labeling objects that came into her field of vision. In another job she examined images of paws, she said, and in another she marked aerial photographs of animals.
For nearly a decade, Kaufman has performed thousands of small tasks that help companies collect huge data sets used to train artificial intelligence (AI), she said.
“These products are supposed to look like magic,” Kaufman, who performs tasks on the platform Amazon Mechanical Turk and advocates for workers as the lead organizer of the group Turkopticon, told ABC News. “People don't know that behind all of this there is a workforce, a human workforce.”
AI has reshaped everything from medical diagnoses to wedding vows to stock market profits, but this technology wouldn't be possible without gig workers like Kaufman around the world.
But analysts and advocates say workers working on AI training are often given no knowledge about the end products they are helping develop or the companies behind them. ing. You also run the risk of having your work rejected after it is completed, which could mean you won't get paid or have any way to collect your money.
“If we want to build a better society, we cannot ignore the tens of millions of people who are doing this work,” Sonam Jindal told ABC News. Jindal is director of AI, labor and economics at Partnership on AI, a coalition of AI organizations. “If they're being overlooked and they're in a precarious situation, that's a problem,” he said.
To mimic human discernment, AI products typically use algorithms that respond to queries based on lessons learned from scanning large amounts of text, images, or video. For example, an AI tool that helps doctors diagnose cancer could be trained using digital copies of CT scans.
But training materials often have to be first curated by human workers so that AI models can read the content, Jindal said.
“The AI model doesn't know on its own how to tell a cat from a dog, or whether someone has cancer, or whether it's a stop sign,” Jindal explained. “People are deeply involved in building these datasets.”
According to a 2021 report by Open Research Europe, the number of gig workers around the world started increasing over the past decade, in part to perform these AI-related tasks. According to the study, about 14 million workers get jobs through online platforms such as Amazon Mechanical Turk and Upwork, which act as intermediaries between freelance workers and tech companies.
Many of these global workers live in the United States, with about 96% of Amazon Mechanical Turk employees, for example, logging in from the United States, according to the data site MTurk Tracker.
Although online gig workers in the U.S. maintain flexible schedules, their jobs include many of the key characteristics of “bad jobs,” says Matt, an assistant professor in the technology management program at the University of California, Santa Barbara. Bean told ABC News.
“Bad jobs are basically jobs that don't give you a lot of autonomy over what you do,” Bean says. “In other words, they don't feel like there's a meaningful connection between what they're doing and any worthwhile outcomes in the world.”
The lack of meaning is partly due to the routine nature of the work and the lack of information provided to freelance workers about the products they are developing and the companies that will manufacture them, Jindal said. Stated.
“Transparency is a big issue,” Jindal said. “This is partly about a very pragmatic approach to building AI models. People say, 'All I need is data.'”
“Information is being passed to someone who may not know the full context,” Jindal added.
In addition to uncertainty about the end product, AI gig workers are at risk of what they call “mass rejection.” This is when a company rejects a completed body of work.
In such cases, workers lose wages and have no way to appeal the ruling, but the company retains the data they create, Kaufman said. She added that companies offering work on Amazon Mechanical Turk may deny data for no reason and change usernames as a way to avoid accountability.
As a result, workers not only lose immediate income, Kaufman said, but their approval ratings on the platforms, which determine the quality of jobs offered to workers, also take a hit.
“So the more rejections you get, the worse your approval ratings are,” Kaufman explained. “Things like that can take away a person's entire life.”
In response to a request for comment from ABC News, Amazon said Mechanical Turk monitors large volumes of rejections and takes appropriate action, including suspension, when it encounters rejections.
Amazon added that the average rejection rate on its platform is less than 1%. Additionally, the company said it has a participation agreement and terms of service to ensure there is no market misconduct by those requesting or agreeing to perform work.
In her work at Turkopticon, Kaufman and other employees pressured Amazon to improve workforce conditions for AI-related jobs, she said. The explosive growth in popularity of AI products has increased public attention to the challenges faced by these workers, she added.
“I feel like I have more power and more awareness,” Kaufman said. “It's an incredible feeling.”