Are fortune cookies Chinese food? Google’s artificial intelligence system is hungry to find out. The machine learning algorithms that power online search rely on vast amounts of organised data to operate effectively. But the answer might depend on where you live: fortune cookies, ubiquitous in North American restaurants, were actually invented in California, inspired by a Japanese cookie.
So US searchers might label fortune cookies Chinese, but Chinese searchers would not. That’s why Google, Facebook and other corporations depend on a large, global force of remote freelance workers enlisted by third-party contractors to tell the machines how locals make sense of things. Some of the biggest companies in the industry, such as Appen, Leapforce and Lionsbridge, have as many as a million worker accounts each. These roles and similar jobs, like content moderation, are often referred to as ghost work: the invisible human labour that keeps AI systems running.
But unlike platforms such as Fiverr, these third-party companies offer what’s sometimes termed microwork — a series of small tasks that can be completed by strangers over the internet — with no opportunity for career advancement. There’s no reputation to build, no clients to court. There’s no contact with managers either, because the process is highly automated: the machines are doing the hiring. Companies send freelancers time-limited, project-based work for specific clients. Freelancers must qualify for each project, so there’s no guarantee of future work, and they can be ejected from the website at any time. Many of the projects require local knowledge, meaning workers must certify that they’re residents of a specific country and fluent in the language — and that people in lower-income countries can’t out-compete other freelancers on price. Residents of the Middle East or South Asia might receive just a few dollars per hour, while those in the UK, Europe and the US can make four times as much.
So instead, some freelancers have decided to game the system. Remote workers in poorer countries are evading residency requirements to present themselves to machines as European residents. These independent contractors are crossing borders using their laptops, making the boundaries imposed by the company redundant.
On a video call, Jake* sits at his desk in an unnamed city in Africa, wearing a royal blue t-shirt. He shows me his work setup on his laptop screen. “So right now, we’re in France.” Jake switches tabs. “Now we’re in Spain.”
Jake does well with the France-based projects. He passes the periodic audits, he says, despite no knowledge of French: he keeps Google Translate open as he works. In addition to his day job, Jake works about a dozen hours per week between the two European markets and makes the equivalent of an average full-time salary where he lives.
People will always try to game automated systems when there’s an incentive to do so, says Rajshree Agarwal, an economist who has studied machine learning-assisted hiring. That’s why any automated system needs human input — whether it’s for job applications or search engines. But machines need to be pointed in the right direction. “The human element is critical,” she says.
Having foreign workers on projects that require local knowledge might create some noise in the data, says Alan Black, a computer scientist at Carnegie Mellon University. But that’s the advantage of machine learning: the vast quantities of data mean Jake’s responses are just a tiny drop in the bucket, and analysts use a range of techniques to sort out bad answers — whether they’re caused by a twitchy mouse finger or an unqualified worker.
It’s difficult to say how prevalent residency gaming is, but researchers agree that if it becomes significant enough to isolate in the data, companies tend to take further steps to curb it. The contractor companies do appear to be aware of the phenomenon, though: workers said their circumvention methods have failed at times, leaving them unable to access the platform.
“Tech companies have long utilised contracting and outsourcing — from offshoring hardware production to Asia starting in the 60s all the way to content moderation today. Such work was and is the giant submerged iceberg of the tech labour force,” says Margaret O’Mara, a historian at the University of Washington. “Given the limited pay and benefits afforded these workers, it is not surprising that some make efforts to work around the system to give themselves a less precarious place in it.”
Google did not respond to a request for comment.
In online microwork’s early days, Amazon’s Mechanical Turk required workers to nominally certify their US residency, without much oversight. But as foreign workers began posing as US residents, the platform started requiring a Social Security number. Back then, workers around the world were attempting to adjust to the US market — but there’s been a reversal: now, tech companies require people who speak other languages to help feed their machines.
“Reviews in Hindi are much more likely to be from the heart, while reviews written by native Hindi speakers in English are much more likely to be polite,” says Black, whose research focus is computer speech synthesis. “So these companies now want to deal better with Hindi, too.”
Kevin*, a German resident who previously worked on freelance project platforms, known as crowdwork, before focusing on microwork, says that the flat rate of the microwork sites ensure he gets a decent wage, no matter what the project. But there’s no way to advance. “With crowdwork, you can earn more money and reputation,” he says. But with microwork, “even if you’re very good at one project, that doesn’t mean you’ll get more projects or more money — and sometimes there’s no work at all,” he says.
Some observers, like Thomas Friedman, previously predicted that the flattening of labour markets — the ability to source workers from anywhere in the world — would cause different wages around the world to converge somewhere in the middle. That hasn’t happened yet, says Hernan Galperin, a digital inequality expert at the University of Southern California, because the online job market still favours local workers. And a few interlopers like Jake and Kevin are unlikely to tip the balance. “It would need to happen at scale, which does not seem to be the case thus far,” Galperin says.
A new work-from-home push that has emerged since the pandemic in Silicon Valley has also stirred debate about how to compensate newly remote employees for cost of living. Twitter and Facebook both announced permanent work-from-home options last month, and Facebook said half of its employees may be remote within five to ten years. In a recent interview, Facebook founder Mark Zuckerberg said the move could attract skilled workers who’d been unwilling to move to a big city. He also said employees who move to cheaper areas could see a pay cut. These companies are already accruing similar benefits from their freelance workforce, Agarwal argues.
The demand for remote freelancers to work on these databases is likely to grow, researchers say: algorithms can always be refined and categories spring up and change all the time. More and more companies are also using crowdsourcing platforms to accomplish tasks: as of 2016, the tech industry accounted for less than half of all crowdsourced tasks. The World Bank estimates that the entire online outsourcing industry will generate up to $25 billion in revenue this year — up from $4.4bn in 2016. Richard Heeks, a professor of development informatics at the University of Manchester, estimates that as of 2019, there were ten million digital gig workers across Africa, Asia and Latin America. Malaysia has even made boosting its ranks of online freelancers a policy goal to capture more of the market. “Because it’s very border independent, it’s a very easy thing for companies to do across the planet,” Black says.
Unlike an undocumented immigrant, who in the United States can be prosecuted for crossing a border, these workers may be guilty of nothing more than breaching their work contracts by lying about where they live — a civil offence, not a criminal one, says Charles Pastor, in-house counsel for cloud recruiting platform iCIMS. Even if the companies in question took such a case to court, it may be difficult to take action against a worker who lives abroad.
But Agarwal says protectionist immigration policies are increasingly out of step with the new reality of widespread remote work — such as the recent suspension of visas that allow foreigners to work in the US.
“If you can be abroad and work for a US company, you don’t need a work visa,” Agarwal says. “The H-1B visa policy is missing the point: in a remote work world, the jobs are going to follow the market.”
Jake says he can only work about ten hours per week on the website — but he tries to save as much as he can. “I might want to move to France in a few years,” he says. “And I’ll need a few thousand euros in my bank account for the visa.”
*Names have been changed. Details of their work and locations have also been omitted to protect their identities
Looking to make your next big career move? Check out WIRED Hired for the latest tech jobs
More great stories from WIRED
🦆 Google got rich from your data. DuckDuckGo is fighting back
💰 The Animal Crossing fans running in-game businesses
🤑 Inside the ‘bullshit’ get-rich-quick world of dropshipping
🎵 The secret behind the success of Apple’s AirPods
🔒 The UK’s lockdown rules, explained