Meet the hidden human workforce behind the boom in artificial intelligence
Since ChatGPT took the world by storm last fall, people have been in a frenzy debating the impact artificial intelligence and other new automated technology will have on America’s job market. The “robots are taking our jobs” narrative was further boosted by viral videos showing new, “fully automated” McDonald’s and Taco Bell restaurants.
The knee-jerk reaction to these videos is to say that robots are coming for our jobs, but while AI and other kinds of automation have progressed, that doesn’t mean they’re necessarily eliminating jobs. Instead, the new tech is simply changing how we work and what kinds of jobs exist. Automation technology has ushered in a fleet of secret workers behind screens, machines, and smiling robot faces. The robots and chatbots aren’t replacing humans, they’re just keeping the people out of sight and out of mind. And while separating the customer from the workers serving them may be good for the companies, there’s mounting evidence it’s a terrible deal for the employees.
Out of sight, out of mind
When people start prognosticating about AI coming for our jobs, they love to point to videos of sleek robots and shining screens handling mundane tasks. The McDondald’s video shows a machine delivering food at the drive-thru, self-ordering kiosks, and a stark lack of human staffers behind the order counter. At Taco Bell, there are several automated drive-thru lanes. Beyond fast food, impressive-looking robots are working in coffee shops, delivering food, and cleaning floors. On the AI side, tools like ChatGPT have been used to write real articles and take college exams. BuzzFeed recently announced it plans to use AI to help generate content for its site.
But in many cases, these videos and stories — and the fears they drive — lack crucial context. The futuristic McDonald’s is an experimental concept store outside Fort Worth, Texas, aiming to improve service speed and accuracy by effectively severing the relationship between its workers and customers. But that doesn’t mean there are no humans in the store. If you pay close attention to the video, you can see a worker in the back behind a pane of glass. McDonald’s has said the store isn’t “fully automated” by any means, and it employs a similar number of staff as a traditional store — they’re just in the back making the food and keeping things running. And despite the fact most customers will never see a Taco Bell worker at its newfangled store, it has plenty of people working in the kitchen.
Even if these tools seem more sophisticated, this isn’t the first robot-driven freak-out. Less than 10 years ago, new developments in technology spurred a similar fear that robots were coming for us. A 2014 analysis estimated that automation would wipe out 47% of all jobs by 2034 and that self-driving technology would eliminate the need for human taxi and delivery drivers, while long-haul truckers were thought to be on borrowed time. Nearly a decade later, these cataclysmic forecasts haven’t come to pass. Truck drivers are still in high demand, and self-driving technology is nowhere near replacing human jobs. In fact, a more recent 2020 report from the World Economic Forum estimated that while 85 million jobs would be replaced by machines by 2025, an estimated 97 million new jobs would be created to help support this new economy.
The human behind the curtain
Amid the fear-mongering about a robot takeover, people often miss the degree to which machines still require human workers in order to function. Take customer service: For years, businesses have tried to cut costs by replacing human phone calls with chat-based, automated customer-service bots. But instead of replacing customer-service workers, many of these text-based tools still rely on human backups in complex situations and to make customers feel as if they are talking to a real person.
Laura Preston recently wrote about her experience working as one of these “human fallbacks” for a real-estate chatbot called Brenda. When a customer wanted to speak to someone about an apartment listing, they would be connected with Brenda, who could answer basic questions about the listing or give details on the apartment itself from the price of rent to the square footage. But many of Brenda’s answers came across stilted or the system was simply unable to answer more complex questions, so a “human fallback” would step in. Preston and other human workers would take over the conversation and try to help the client, cleaning up stock answers to better address their needs or doing deeper research into housing vouchers and pet policies. According to Preston, employees were trained to use Brenda’s “voice” in the interactions in an attempt to make the conversation appear seamless. And the push to robotically answer a deluge of questions came with a serious mental toll: “Months of impersonating Brenda had depleted my emotional resources,” wrote Preston. “It occurred to me that I wasn’t really training Brenda to think like a human, Brenda was training me to think like a bot, and perhaps that had been the point all along.”
Preston was working from the United States, but in many cases, these services are hiding human employees so they can outsource it to places where the labor is cheaper. Take food-delivery robots: While they’re presented as being fully autonomous, the reality is that they often have remote backup drivers. Tiny Mile, which runs a service in Toronto called Geoffrey, relies on drivers in the Philippines, while Kiwi’s robots, which are used on some US college campuses, have been known to use workers in Colombia making less than $2 an hour to help complete deliveries. Companies claim the remote drivers only take over when the robots can’t navigate a situation, but given how prone these robots are to getting stuck and becoming obstacles for sidewalk users, it’s not clear how often that happens.
Many of the leading companies developing self-driving vehicles, which have threatened to one day replace a host of driving jobs, also rely on a fleet of hidden workers. Obviously, there are the highly paid engineers in the US who help develop the software and tools used to map and guide the car, but that isn’t the full picture. Autonomous-driving tech is reliant on poorly paid workers around the world who label the thousands of data inputs the car’s sensors capture. Without that labeling, the computer wouldn’t be able to identify what the sensors are picking up, which allows the systems to slowly learn and make decisions on how to navigate the road. For instance, the data labeling is supposed to help the cars distinguish whether an obstruction is a child or a traffic cone — though it doesn’t always work out as planned. An MIT Technology Review investigation last April found that self-driving companies, including Tesla, took advantage of the collapse of Venezuela’s economy by getting workers in the country to label self-driving data for an average of just over 90 cents an hour. Last year, Tesla laid off 200 US-based workers it directly employed to do this labeling, suggesting it was automating a majority of those tasks instead — a computer teaching another computer.
It’s not just self-driving tech. A recent Time magazine investigation found that OpenAI, the company behind ChatGPT, relies on Kenyan workers paid less than $2 an hour who have to view content on a range of disturbing topics, including “child sexual abuse, bestiality, murder, suicide, torture, self harm, and incest,” to try to make the tool less toxic. This follows an earlier report about Facebook using the same subcontractor in Kenya for its own content moderation. Other companies have gone so far as to recruit workers in refugee camps, where the opportunities are few and people will accept incredibly low wages, to help train their machine learning and AI tools. Far from the “set it and forget it” tools that require just a few genius coders, many of these supposedly autonomous marvels actually rely on an army of low-paid workers around the world.
The myth of efficiency
If you ask the companies, they’ll say the automation push is about efficiency and better serving customers. McDonald’s claims its concept store will speed up service and lead to fewer wrong orders, while Tesla has claimed that automating data labeling is more efficient. The idea is that these machines or software solutions will allow a job to be done faster or better, making life easier for companies and customers alike. But in reality, these tools aren’t more efficient — they just shift the necessary work away from the end consumer and disconnect people from the effort that is required to deliver them a product.
For one thing, it’s not even clear that all the newfangled tools that companies have built are actually making the economy more efficient. US labor productivity — the measure of how many worker hours are required to produce a certain amount of economic output — has been growing at below its long-run average since 2005. And despite hope that the forced digital transition would turn it around, productivity growth has only gotten worse since the start of the pandemic.
Instead of improving productivity, automation is often focused on increasing the power that employers have over workers. In his book, “Automation and the Future of Work,” the economic historian Aaron Benanav explains that companies aren’t putting money toward tools to make employees’ lives easier, but are pouring money into “technologies allowing for detailed surveillance of those same workers” like computer-monitoring software that tracks the keystrokes of employees or Amazon’s sophisticated algorithmic management tools that evaluate workers’ every movement.
These technologies are often deployed to de-skill work — jobs are broken down into more specific tasks and can be done with less training. As a result, workers are shifted from employee to contractor status. People who once worked stable, middle-class jobs are thrown into a more precarious world where wages are lower and they have less say over the terms of their employment. The data labelers are the tip of this iceberg: A large (and growing) industry of “microworkers” on platforms like Mechanical Turk or Clickworker fuel the supposed automated tech of all these tech companies. Amazon CEO Jeff Bezos went so far as to call the use of workers to make a process seem automated “artificial artificial intelligence.” For those who still hold onto service or warehouse jobs, the specter of automation is wielded like a Sword of Damocles to keep them from pushing for better working conditions or wages. Meanwhile, the technologies that are deployed simply give employers more power to track everything workers do while on the clock — a miserable working reality. Looking at this trend in 2015, the journalist Lauren Smiley wrote that it was creating a world where “you’re either pampered, isolated royalty — or you’re a 21st-century servant.”
There’s no question that some of these technologies provide conveniences for consumers or even free up their time so they can focus on their own work, but that doesn’t make the treatment of the workers they depend on acceptable. Technology could be used to empower workers, for example, by giving them more control over their work so they can use their skills to make more informed decisions, but that data is often not shared by companies.
Instead of using this new technology for good, Phil Jones, a researcher and the author of “Work Without the Worker: Labour in the Age of Platform Capitalism,” argues that companies simply deploy semi-automated tech to make it seem as if executives or the brand itself deserve all the credit for the end product rather than the human employees who made it possible. “Workers disappear in the long shadow of the machine,” Jones writes, and customers and clients don’t have to think as much about how the sausage gets made. All the while, the quality of jobs is degraded.
New technologies like AI are framed as offering us various forms of empowerment and liberation: We’ll be able to work more productively, spend less time doing our chores, and anything we want will be a click or tap away. But those promises never paint an accurate picture of how that tech is transforming the world around us or the true cost of those supposed benefits. Automation may empower some people, but in the process, it’s making things a lot harder for the hidden workers keeping everything moving.
Paris Marx is a tech writer and host of the Tech Won’t Save Us podcast. They are the author of the book Road to Nowhere: What Silicon Valley Gets Wrong about the Future of Transportation.