Skip to Content
Policy

Desperate Venezuelans are making money by training AI for self-driving cars

Growing competition to develop self-driving cars—and the high stakes of getting things right—have created new crowdworking platforms that could be a lifeline for desperate workers.
August 22, 2019
Animated gif of the Venezuelan flag and autonomous vehicles in place of the flag's stars
Animated gif of the Venezuelan flag and autonomous vehicles in place of the flag's stars

Next time you hear about the wonderful future of self-driving cars, picture this: Venezuelans, living in crisis conditions after their economy collapsed, sitting at laptops and outlining pictures of trees and bikes so that the robotic vehicles don’t crash.

That was the situation in 2018, according to Florian A. Schmidt, a crowdwork expert and professor of design at HTW Dresden. “You have these formerly middle-class, well-educated, well-connected people with good internet infrastructure who suddenly dropped into poverty,” says Schmidt, who wrote a paper on this new labor market for the Hans Böckler Foundation, a research arm of the German trade union federation. (The paper was released in English this week.) Desperate for work, Venezuelans came across a new group of online crowdworking platforms. These companies—including Mighty AI, Playment, Hive, and Scale—cater to the autonomous-vehicle industry and could be a new battleground in the debate over whether gig workers should be considered employees. 

Hundreds of  thousands of workers from Venezuela signed up to work for these companies last year, in some cases making up as much as 75% of a firm’s workforce. Even today, 75% of search traffic to Mighty AI comes from a site advertising jobs in Venezuela. The companies don’t pay more for data labeling than a platform like Amazon Mechanical Turk, but they do offer a steadier source of income, providing a measure of security for those in a country where inflation recently hit 10 million percent. (Mighty AI did not respond to a request for comment.)

No pixel left unmarked

It’s no secret that AI relies on poorly paid humans to label massive amounts of data. Humans do everything from transcribing voice recordings to identifying NSFW images. These new crowdwork companies are the result of the growing competition to develop self-driving cars and the high stakes of training the vehicles to see and navigate properly. 

Other data-labeling tasks, like building an algorithm for search results, have more room for error. “If you make a query on a search engine and three out of 10 results are crap, it doesn’t really matter,” Schmidt says. “But a level of 30% wrong answers would be totally intolerable under conditions of traffic.” The work itself can be more demanding, too. The cars’ onboard cameras record vast amounts of visual information, and labelers must outline every single object in a picture or video footage. 

As a result, platforms like Mighty AI handle the entire process of finding, training, and managing workers so that their clients—companies busy building and testing autonomous cars—never have contact with them. In fact, many of these companies have two different names for the two sides of their businesses. Mighty AI’s worker-facing name is Spare5 (as in, spare five minutes to do some work); Scale’s is Remotask. 

For the workers from Venezuela, these more centralized platforms were an improvement because “[the workers] are treated more like humans and the work is more valued,” says Schmidt. Many of the Venezuelan workers recruited friends and family to do this work. And they came to rely on their pay, as opposed to the Italian and Brazilian crowdworkers that Schmidt interviewed, who saw the work as a hobby that provided some extra cash. “[The Venezuelans] were aware that, on one level, it’s exploitation and they have to do it because everything else failed them,” Schmidt adds. But they were also happy to have found the work and to have a steady flow of income.

This influx was a surprise for the companies too. Many data-labeling businesses deliberately set up shop in developing countries, but all these enterprises did was translate their websites into Spanish. 

New frontier in the gig-work debate?

Around the world, independent contractors are fighting to be classified as employees. The outcomes have big implications because contractors don’t receive insurance, pensions, and other workplace protections. The issue is relevant for Venezuelan workers too, because many who have left for neighboring countries started doing gig work as bicycle couriers or drivers. The debate so far has focused on these in-person workers, but data labelers for Mighty AI and Playment might have a case as well. Because these companies handle so much of the training and job assignments, they act far more like a traditional employer than a platform like Mechanical Turk. 

But companies can classify workers as employees in one country and independent contractors in another even if they do the same job, according to Valerio de Stefano, an expert in platforms and employment law at KU-Leuven in Belgium. For example, the food delivery platform Foodora classified its workers as employees in Germany but independent contractors in Italy. So even if data labelers in, say, Spain became employees, those working for the same company in Venezuela might not have the same rights. For digital companies, there is also the risk that they will shift their workforce to countries with weaker labor protections. 

Back in 2015, the site CrowdFlower settled a case that accused it of, among other things, misclassifying employees as independent contractors. There haven’t been major lawsuits since then, but as in-person gig workers are starting to gain more protections, crowdworkers might be better poised to try again. For workers in economically impoverished areas, it could be a real boon. For companies, it’s simply “part of the cost of business to comply with the different rules,” de Stefano says. “And if they are not sustainable by complying with the rules, they probably shouldn’t be there in the first place.” 

Deep Dive

Policy

Is there anything more fascinating than a hidden world?

Some hidden worlds--whether in space, deep in the ocean, or in the form of waves or microbes--remain stubbornly unseen. Here's how technology is being used to reveal them.

Africa’s push to regulate AI starts now        

AI is expanding across the continent and new policies are taking shape. But poor digital infrastructure and regulatory bottlenecks could slow adoption.

Yes, remote learning can work for preschoolers

The largest-ever humanitarian intervention in early childhood education shows that remote learning can produce results comparable to a year of in-person teaching.

Stay connected

Illustration by Rose Wong

Get the latest updates from
MIT Technology Review

Discover special offers, top stories, upcoming events, and more.

Thank you for submitting your email!

Explore more newsletters

It looks like something went wrong.

We’re having trouble saving your preferences. Try refreshing this page and updating them one more time. If you continue to get this message, reach out to us at customer-service@technologyreview.com with a list of newsletters you’d like to receive.