Virtual sweatshops are a form of crowdsourcing, where companies break large tasks down into several smaller tasks that can then be outsourced online to independent contractors. These smaller tasks require human intelligence that computers, algorithms, and artificial intelligence are unable to easily solve, such as CAPTCHA. While the first company to implement and utilize virtual sweatshops has not been identified, it is known that the trend began in the early 2000s. Most companies that use virtual sweatshops are not transparent about the process. Nonetheless, the phenomenon gained media attention in December of 2014 when workers at Amazon’s Mechanical Turk, a form of a virtual sweatshop, protested against the company’s owner. Virtual sweatshops came under scrutiny in the latter half of the 2010s in regards to ethical concerns - working conditions, policy gaps, and virtual trust decreasing.
- 1 Background
- 2 Applications
- 3 Reception and Awareness
- 4 Virtual Sweatshops and Research
- 5 Ethical Concerns
- 6 See Also
- 7 References
The words “virtual sweatshop” are derived from the term “sweatshop” which was coined in 1884 to signify a shop or factory in which employees work for long hours and low wages. Virtual sweatshops are often likened to sweatshops, except virtual sweatshops are accessible online and workers can be hired on a global scale. Defining attributes of virtual sweatshops include: employees working on tasks outsourced by a company, receiving low compensation for completed tasks (in US currency, wages can start as low as 1 penny), and work that is completed solely online. Workers in virtual sweatshops are tasked with laborious and redundant functions in order to alter, refine, and improve the quality of online data.
The origins of virtual sweatshops have not been traced to a particular instance in history, but its usage and popularity emerged in the early 2000s. In 2013, the online outsourcing industry was estimated to have earned $2 billion in revenue, and expected to increase to $25 billion by 2020. This market, comprising of sites like Clickworker, UpWork, Mechanical Turk, Crowdflower, and more, has grown immensely since the early 2000s and is anticipated to continue growing well into the 2020s. Essentially, any individual with a computer and internet connection can work for a "virtual sweatshop," such as Mechanical Turk. The company will assign thousands of micro-tasks, such as tagging images, moderating content, taking surveys, or completing user experience tests of websites and applications. Certain companies have made it practice to outsource these tasks to third-world countries, where the minimal pay would have higher value, be more desired, and attract a larger low unit cost labor pool. The real value for companies that utilize "virtual sweatshop" labor is not in the tasks themselves, but in the data that is generated from completing the tasks. The data could be used in a variety of ways, such as perfecting an AI's pattern-recognition algorithmic capabilities. 
Companies would outsource small tasks, such as tagging photos, taking surveys and user experience tests, or verifying URLs, that humans can easily complete, but difficult or impossible for computers. Once the "workers" pass a computer skills and reading comprehension test, they are assigned tasks.
Virtual sweatshops are also platforms for companies to challenge scientists in solving complicated problems for financial rewards. This application is the most lucrative form of employment in virtual sweatshops, as these companies are willing to compensate a large payout for solving problems that their immediate could not solve. Virtual sweatshops are also revolutionizing data collection for researchers, as samples in online surveys are increasing in its accuracy of representing the population.
Amazon's Mechanical Turk
Mechanical Turk was launched by the company Amazon in 2005 and is one of the largest virtual sweatshops in the U.S., with over 0.5 million virtual workers. Workers are referred to as “Turkers” and complete HITs (Human Intelligence Tasks) that are posted on Amazon’s site. The Turkers have the ability to choose which tasks to complete and are compensated in credits for Amazon.com. Turkers can complete tasks, but their employers must be located in the United States. Employers are known as Requesters, and they have the option to set the criteria to seek workers and can accept or reject the final product the Turker submits. Employers are typically entrepreneurs or owners of newly founded companies that need to test out a product or website design. This user experience testing process occurs either prior to launching a new idea, or in a revamping stage of the product or website in attempt to gather data on user preferences and advice input to improve the typical user experience. For this reason, Turkers are asked to interact with the product under revision normally while describing their thoughts and steps while completing tasks through the interface aloud so that their experience can be recorded and evaluated.
Mechanical Turk was originally the most successful form of crowdsourcing at the time of its release, but other companies followed suit in the latter half of the 2000s and in the 2010s to harness the utility and power virtual sweatshops can provide. One can argue that this type of access has its disadvantages as "it is expensive, difficult, slow, and not obvious if it produces better outcomes".
Cambridge Analytica was a company that developed out of the SCL Group in 2013. Its main purpose was to collect data on people’s personalities so that an algorithm could be constructed for campaigns to more effectively target viewers. The company played an important role in Donald Trump’s U.S. presidential campaign and in Britain’s EU membership referendum Leave campaign. More specifically, it's known that the data analytics firm helped Donald Trump to victory. 
The case of Cambridge Analytica is an instance in which virtual sweatshops were used for analytical and research purposes. To acquire personal data, the company utilized Amazon’s Mechanical Turk and Qualtrics, a survey tool, to hire Turkers to take personality quizzes. The quiz paid Turkers $1 to $2 to complete the survey and provide access to their personal Facebook data. Around 320,000 Turkers participated in the survey, and unknowingly gave Cambridge Analytica access to at least 160 Facebook profiles per Turker. This data was then used by the company in developing their algorithm.
KolotiBablo is a virtual sweatshop that specializes in solving CAPTCHAs, a small puzzle many websites use to ensure their users are humans and not bots. CAPTCHAs are in place to prevent spammers from mass creating accounts and comments on websites. Although KolotiBablo claims their business does not support spammers, the service is commonly advertised among those who want to automate activity using bots . KolotiBablo capitalizes on cheap labor in countries like China, India and Pakistan . The service uses an application programming interface to send CAPTCHAs to KolotiBablo workers. The workers then solve the CAPTCHA and send it back to the client . KolotiBablo pays their workers $1 or less for every 1000 CAPTCHAs they solve. Most KolotiBablo workers make around $2 a day. For many KolotiBablo workers this is their full time job and only income . KolotiBablo is criticized for exploiting workers and providing a service that is used to bypass security measures on many websites 
Another version of the virtual sweatshop is setup in the warehouse basements of China for outsourced gaming. In 12 to 18 hour shifts, young people in China "play" computer games by killing onscreen monsters and winning battles, harvesting artificial gold coins and other virtual goods as rewards that, as it turns out, can be transformed into real cash. The virtual sweatshop is run like that of a stock exchange where websites like Ucdao.com and Ebay are used as platforms to buy and sell menial tasks to the gamers also known as gold farmers. These "gold farmers" are called in because from "Seoul to San Francisco" affluent online gamers who lack the time and patience to work their way up to the higher levels of gamedom are willing to pay the young Chinese to play the early rounds for them.  All of this work is done for less than a quarter an hour. The gaming business is quite a lucrative one, especially for the owners of the companies that run these virtual sweatshops, however, that money is quickly pocketed before it is seen by any of the workers. Some workers concede that the $75 a month is comparable to what would be made elsewhere in China, and that the free room, board, and computer game play makes it enough for them to stay in the virtual sweatshops. 
Reception and Awareness
Virtual sweatshops remain relatively unknown in the media and the general public’s knowledge. Many companies that use virtual sweatshops ask workers not to disclose that they have worked for them. This form of labor can allow workers more flexibility in their schedules, as the workers are the ones choosing which tasks to complete and when. Since these virtual sweatshops are entirely online by nature, this allows employees to work from home. Flexibility, location, and simpleness of tasks are the main motivations as to why employees work for virtual sweatshops.
A Christmas email campaign, initiated in December 2014 on behalf of Amazon’s Turkers, began protesting against the terms of virtual labor. The campaign, directed at Amazon’s CEO Jeff Bezos, demanded better representation and recognition for workers and for more regulation. It protested against the lack of legal consequences, such as employers not paying their employees and requested for minimum payment rates for work. In response to these types of working conditions, online forums were created for virtual workers to share tips and evaluate sources of work.
Virtual Sweatshops and Research
Crowdsourcing is very popular when it comes to research as it allows access to multiple people and gains results quicker. For example, Amazon's Mechanical Turk is well known in regards to research. From the researchers' perspectives, they can easily access a wider variety of "subjects" or "participants" at any time and/or day.  Thus the sweatshops may serve to advance education and human knowledge. However, this raises the question of whether these studies are ethical due to the poor conditions of workers who provide the data for the studies.
While virtual sweatshops provide a pristine example of how to use the internet to find and direct brainpower in order to complete mundane tasks as well as foster a new market for human labor, there are many ethical concerns that arise due to this new phenomenon.
Policy Gaps and Job Exploitation
Virtual sweatshops are not under the same legislation and regulation as physical human labor. Many virtual sweatshop employers are not required or enforced to file forms for payroll taxes, set minimum wages, compensate overtime, or offer benefits to its workers. Job security is not ensured and companies can cancel or refuse to pay any work that they requested. This lack of regulation is due to the policy gaps that form as new technologies emerge.
Unfortunately, the computer revolution led to fast-paced changes in technology which outpaced ethical policies. Virtual sweatshops were not anticipated so there weren't guidelines or laws in place to handle the virtualization of labor. Moor's Law defines a linear relationship between social impact and ethical problems, in which technological revolutions that have a greater social impact also create more ethical problems. Virtual sweatshops are in accordance with Moor’s Law meaning that the more widespread they became, the more ethical problems they tend to generate. This phenomenon can be proven by Cambridge Analytica’s exploitation of virtual sweatshops, as they used virtual sweatshops (which already faced ethical concerns regarding payment and job security) to access other users’ Facebook data, thus creating more ethical ambiguity and concern regarding privacy and accessibility.
There have been debates in the 2000s and 2010s in regards to how to facilitate ethics in online discourse, among virtual laborers. After almost two decades of existence, there has been a minimal improvement. However, virtual sweatshops still maintain their unethical practices with little improvement. The lack of improvement is due to the lack of transparency in how companies work — companies that use virtual sweatshops often require that their employees refrain from mentioning their affiliation and taking measures to avoid drawing public attention, thereby obscuring the processes in which they are obtaining their labor force. The lack of transparency in the form of information invisibility enables companies in their unethical actions, and hindering the formation of tighter regulation.
Without regulating this online labor force to the extent that physical jobs are regulated, this policy gap continues to exist in virtual sweatshops, through movements like Amazon’s Turkers email campaign advocate for change.
Governance and Decentralization
Virtual sweatshops typically do not adhere to formal law since they operate almost exclusively online across multiple countries but can be regulated by company guidelines, the market, and the architectural structure embedded in the technology. Wikipedia has undergone a phenomenon in which its governance has increasingly decentralized as a result of the freedom and flexibility in its technological design and organization. Companies have been able to self-prescribe the mechanisms, expectations, and work expectations for the "workers". Since virtual sweatshops solely exist online, companies have the option to forfeit bureaucratic hierarchical work structures in favor of software programming schemes that organize the virtual work environment for them.
This restructuring of governance has allowed virtual sweatshops to operate decentrally, which aids company growth and production. Yet, it presents ethical problems by dispersing accountability. Without the presence of formal law, virtual sweatshops can only be held accountable to the standards that the company establishes for themselves.The globalization and decentralization of virtual sweatshops made it challenging for those seeking more regulation because they must confront the different legal frameworks depending on the workers' location. Virtual sweatshops can outsource their labor to many nations and organizations that do not adhere to the same forms of government, and thus creating loopholes against opponents of virtual sweatshops. This lack of uniform regulation across different domains is an ethical concern since it allows the room for human rights abuses such as discrimination, job insecurity, and low wages less than typical minimum wages, along with avoiding otherwise applicable fees through fraud and tax evasion.
Virtual sweatshops have the option to maintain a steady workforce, despite not offering workers the same benefits and stableness that physical jobs typically provide. Virtual trust plays a large role in attracting workers to virtual sweatshops, as it is expected that their employers will compensate them for their work. Without this trust, virtual sweatshops could not thrive. Many theorists who study ethics debate whether virtual trust is a myth or not and virtual sweatshops show that sometimes it can be trusted.
Virtual sweatshops demonstrate the concept of “verkeersbordvrij”. Verkeersbordvrij is a Dutch term that means “free of traffic signs” which is used to characterize a scientific study of traffic management in the Dutch city of Drachten. The study removed all street signs to see if drivers would still operate ethically and with the safety of others in mind. What the study proved was exactly what has been occurring in virtual sweatshops — that despite the lack of regulation, transactions between workers and employers can remain ethical, and demonstrate trust. This virtual trust can be invested in the actual people behind the screens or the processes themselves, but nonetheless functions to maintain company production and a steady workforce. In other words, virtual trust acts as an unspoken contract that binds participants into an employer-employee relationship to ensure payment and reliability. Virtual trust, however, is not legally contractual which can easily be broken. Due to this ethical uncertainty, many workers prefer to have regulation and laws in addition to virtual trust to guarantee that their work is compensated.
Machine Learning Bias
A large usage of virtual sweatshops is the production of large amounts of data for machine learning algorithms. In machine learning, programs learn mathematical methods to classify data, based on human generated labels. Thus the quality of machine classification is limited by the quality of the human classification. Since conditions are so poor and there is pressure to complete tasks rapidly this degrades the quality of the data. Additionally, biases of the virtual sweatshop workers are subsequently learned by the algorithm. For example, Amazon used an ai algorithm to filter job applications based on past hiring data. The algorithm weighted female applications down, as there are fewer females in the field of software engineering and fewer and gotten hired in the past by the company. So the algorithm inadvertently reinforced pre-existing gender biases.
In order to recruit people to be in the virtual sweatshops, companies will target areas where workers are more likely to accept such a position. Typically, this means turning toward low-income countries where people need jobs and will sacrifice their happiness for the money . Due to the nature of virtual sweatshops, people are only going to be willing to take the jobs if they are in more desparate situations. Some people will say this exploitation of certain countries or people for the gain of others, while the comapnies will claiom they are offering a job that is not physically demanding and people will be getting paid. This plays in the old dilemma of outsourcing and whether companies using workers where they do not have to meet ethical workplace environments and wage standards is proper. This has been an issue for years with actual sweatshops but how does this apply to a digital sweatshop where people are simply on a computer.
- Cherry, M., “A Taxonomy of Virtual Work”, 2011, p. 962-972
- Harris, M., “Amazon's Mechanical Turk workers protest: 'I am a human being, not an algorithm'”, The Guardian, 2014
- ”sweatshop”, Merriam-Webster.com, Retrieved March 10th, 2019
- Kavanaugh, S., “Virtual Sweatshops Paint A Bleak Picture Of The Future Of Work”, 2017
- "The Data Factories and Virtual Sweatshops that make the Internet run smoothly," Inevitable Human, 11/19/18, https://inevitablehuman.com/the-data-factories-and-virtual-sweatshops-that-make-the-internet-run-smoothly/
- Zittrain, J., “The Internet Creates a New Kind of Sweatshop”, Newsweek, 2009
- Samuel, A., “Amazon’s Mechanical Turk has Reinvented Research”, 2018
- Cadwalladr, Carole. “'I Made Steve Bannon's Psychological Warfare Tool': Meet the Data War Whistleblower.” The Guardian, Guardian News and Media, 18 Mar. 2018, www.theguardian.com/news/2018/mar/17/data-war-whistleblower-christopher-wylie-faceook-nix-bannon-trump.
- Weissman, C., “How Amazon Helped Cambridge Analytica Harvest Americans’ Facebook Data”, 2018
- "Krebs on Security.” Brian Krebs, krebsonsecurity.com/2012/01/virtual-sweatshops-defeat-bot-or-not-tests/.
- Barboza, David. “Ogre to Slay? Outsource It to Chinese.” The New York Times, The New York Times, 9 Dec. 2005, www.nytimes.com/2005/12/09/technology/ogre-to-slay-outsource-it-to-chinese.html.
- Urban, Kylie. “Amazon Mechanical Turk: a Way to Recruit Study Participants?”, Crowdsourcing Human Research Subjects | Michigan Health Lab, 21 July 2016.
- Moor, J. “Why we need better ethics for emerging technologies”, 2005, p. 111-119
- Floridi, L. & Turilli, M., “The ethics of information transparency”, 2009, p. 105-112
- Forte, A., Larco, V. & Bruckman, A., “Decentralization in Wikipedia Governance”, Journal of Management Information Systems, 26:1, p. 49-72
- Aneesh, A., “Global Labor: Algocratic Modes of Organization”, 2009, p. 348-370
- de Laat, P., “Trusting virtual trust”, 2005, p. 167-180
- Zittrain, J., “The Lessons of Wikipedia”, The Future of the Internet, 2008, p. 127-146
- Jeffrey Dastin (2018) Reuters "Amazon scraps secret AI recruiting tool that showed bias against women" https://www.reuters.com/article/us-amazon-com-jobs-automation-insight/amazon-scraps-secret-ai-recruiting-tool-that-showed-bias-against-women-idUSKCN1MK08G
- Ong, Jonathan Corpus. “Digital Sweatshops in Disaster Zones: Who Pays the Real Price for Innovation?” The Guardian, Guardian News and Media, 11 Oct. 2016, www.theguardian.com/.