New artificial intelligence (AI) technologies require a huge amount of tags to process the information and differentiate the content on which they operate; a titanic task that is only possible thanks to the work of hundreds of thousands of people who perform such tagging. However, far from being a worthy source of employment, this activity has become a new form of labor exploitation.
Why did the data tagging industry become a new form of slavery and colonialism?
Por: Gabriel E. Levy B.
An in-depth investigation developed by the Massachusetts Institute of Technology (MIT) on the colonialism of artificial intelligence, published through the MIT Technology Review, and supported by the MIT Knight Science Journalism fellowship program and the Pulitzer Center, revealed how the data tagging industry for artificial intelligence systems takes advantage of poverty in countries such as Venezuela or Kenya to obtain practically free labor and, in this way, build a global emporium.
The issue in context
Most of the algorithms operating in e-commerce platforms, voice assistants (such as Alexa or Google), AI systems or automated systems in autonomous cars are based on deep learning, an AI technique that necessarily requires a large number of labeled examples to work; that is, a person needs to be asked to teach an algorithm to distinguish, for example, between a photo and a video, a pair of pants and a T-shirt, or a stop sign and a no-turn sign.
This is achieved through a platform with labels that indicate the data the AI system needs to have on the other side of the screen.
According to figures from MIT, the market value of this “ghost work,” as anthropologist Mary Gray and social computer scientist Siddharth Suri called it, is expected to reach €12.85 billion by 2030.
The MIT Technology Review research also reveals how digital companies engaged in this work, such as Appen, among a dozen other companies that provide hidden data tagging services for the artificial intelligence industry, use a large amount of cheap labor to tag and classify videos, photos and manually transcribed audio.
The new AI generation
As deep learning became an industry in the early 2010s, a new generation of more specialized collaborative work AI web portals emerged, ensuring greater accuracy by providing customers with a more hands-on approach. When automakers joined the program in 2017, they wanted not only better performance, but also 99% or better accuracy, which is how some platforms had to reinvent themselves while others disappeared.
One of the most prominent of these new professional services is Scale AI. Founded in 2016 by 19-year-old Alexandr Wang, a student at MIT. Scale AI quickly amassed tens of thousands of tagging workers and has high-profile clients such as Toyota Research, Lyft and OpenAI.
Scale AI is currently valued at EUR 6.85 billion. In February it was selected among other companies to provide services to the U.S. Department of Defense under a total contract worth up to €234 million.
Scale AI’s initial growth was based on its ability to provide high-quality labeled data quickly and cost-effectively, largely thanks to the army of humans it hires at low cost for these tasks.
In 2017, this system launched a worker-oriented platform called Remotasks to create a global library of cheap contractors. Initially, it sought employees in the Philippines and Kenya, but at about the same time, competitors such as Appen, Hive Micro and Mighty AI’s Spare5, began to notice a sharp increase in registrations in Venezuela. By mid-2018, approximately 200 000 Venezuelans were registered with Hive Micro and Spare5, representing 75 % of their respective workforces.
Building on the success of its competitors, in 2019, Scale began to intensively recruit Venezuelan workers, using referral codes and social media marketing campaigns to convince people that they could earn a lot of money..
Although all these platforms have become an excellent opportunity to generate economic resources for hundreds of thousands of people around the world, the conditions are increasingly precarious and variable, so that at the beginning of the business the payment for the tasks of a tagger who dedicated 12 hours a day could reach up to 60 dollars, according to data collected by researchers; while at present, for the same work, a tagger does not even receive 10 dollars, since for each transcription made, he is only paid a few cents that are added to his account, demanding more and more time and effort with a lower remuneration, without any regulation to put a stop to this type of exploitation.
According to the MIT report, in Venezuela it is common to find families that take turns working 24 hours a day in order to obtain a higher profit. Thus, today, in order to receive the same income that a single person received for 12 hours several years ago, a whole family is required, doing the same activity 24 hours a day, and even so, the income is lower.
In conclusion, although the labeling industry has become an excellent employment opportunity for hundreds of thousands of people around the world, in parallel and with the aim of maximizing profits, companies like Escale AI, Appen, Hive Micro and Mighty AI’s Spare5, far from becoming examples of innovation or digital transformation, are a reference of the new labor exploitation and colonialism of the 21st century.
 Hao, K. and Hernández, P. (April 28, 2022). How the AI industry took advantage of Venezuela’s economic collapse. In MIT Technology Review, (Trans. Milutinovic, A.). Available at https://www.technologyreview.es/s/14171/como-la-industria-de-la-ia-aprovecho-el-colapso-economico-de-venezuela