A more precarious on demand-labor market arising around emerging tech businesses: what can we do about it?
OpenAI’s Chat GPT has unleashed a global concern for the future of labor markets, prompting widespread debates on measures to mitigate the job losses of workers displaced by this new technology. Debates revolve around potential solutions, such as taxing robots that will replace humans, or envisioning a society where work is redefined as an option rather than a necessity.
Yet, there is a noticeable lack of attention to how AI companies and other tech businesses are exacerbating the precariousness of the labor market by generating new types of hidden contractual labor. At the end of this article, I advocate for three measures to mitigate and prevent the vulnerabilities arising from these evolving labor structures.
The large quantity of human labor required to fuel AI systems, websites, and apps is staggering. Human input is usually present in tasks such as solving complex programming; labelling, generating, or cleaning data to train AI models; content moderation; and taking over AI chats whenever the system faces challenges to answer. However, this human input is often hidden and invisible, what Mary Gray and Siddharth Suri (2019) have labeled “ghost work”, as they are integrated in a way that seems like a mere function of a computer program.
The apparent motivation behind concealing this labor is to create an illusion of AI omnipotence, attracting attention towards a particular company’s system “superpower” to persuade users to adopt their AI services. Yet, another motivation might be to hide the precarious conditions under which the workers fueling these technologies operate.
Companies typically outsource this labor by plugging a labor pool into their Application Programming Interfaces (APIs) to automatically hire on-demand workers to carry out a punctual task. These labor pools, called online web-based labor platforms, such as Amazon Mechanical Turk, Microsoft’s internal Universal Human Relevance System, and LeadGenius, provide cheap labor without traditional employment benefits. They are different from location-based platforms, where tasks are performed at a specified physical location by individuals, such as taxi drivers and delivery workers.
As a result, a substantial workforce has emerged around tech companies, characterized by heightened income insecurity and adverse working conditions, blurring the responsibility of companies towards their workers, and bypassing standardized labor conditions. This increases the inequality between the individuals at the pinnacle of the company’s hierarchy and those at the bottom. Moreover, as stated by Berg (2019), this arrangement transfers the risks onto workers, who bear the financial burdens associated with changes in demand.
The demand for this type of labor is predominantly driven by companies based in developed countries, and a significant portion of the workforce supply originates from developing countries (ILO, 2021), especially those with low wages and surplus of tech graduates. However, the 2020 Online Labour Index (OLI) by Stephany et al. (2021) unveiled a surprising reality - in fact, the top three countries contributing to the workforce supply are Asian, while the United States and the United Kingdom hold the fourth and fifth positions, as illustrated in Figures 1 and 2. Notably, Romania, Germany and Italy also make their mark among the top 15 nations.
Henceforth, on-demand workers represent a new type of worker that already constitutes a significant portion of the labor market in developed economies. This suggests that the most immediate impact of emerging technologies is the displacement of conventional employment towards “non-standard employment arrangements, including temporary and part-time work, casual or zero-hours contracts, and bogus self-employment” (Berg, 2019).
The complexity of the phenomenon lies in the fact that their temporary status is part of what makes them indispensable for propelling the ‘AI Revolution’. To illustrate, consider the project of Fei-Fei Li, a computer science professor and co-director of the Stanford Human-Centered AI Institute, aimed at training machines to identify the main object in an image. The team tried several methods to carry out the project, such as hiring undergraduates and developing a machine learning algorithm to automatically assign labels to images. However, they were all failed attempts.
Years later, Li’s team discovered Amazon Mechanical Turk, providing access to a vast pool of inexpensive international labor capable of working around the clock. The resulting data set, called ImageNet, became a gold standard for research teams, enabling the development of more sophisticated image recognition algorithms. “MTurk workers are the AI revolution’s unsung heroes” (Gray & Suri, 2019).
Their lack of fixed commitment to any specific enterprise leads to the formation of a common pool of workers accessible to numerous businesses able to tap this well of shared experience, availability and diversity allowing continuous and affordable innovation. At the same time, it decreases transaction costs given that APIs and AI have eliminated costs associated with recruiting, training, and retaining workers. However, these costs are now borne by workers themselves.
The necessity of this type of labor for advancing technological progress underscores the urgency of regulating this new type of work. Presented below are some ideas that could help prevent the deterioration of the labor market and craft a technological transformation that benefits all.
Firstly, there is a need to broaden the scope of labor law provisions by extending rights and benefits to all sorts of workers, regardless of their contractual arrangement or employment status (Gray & Suri, 2019). This is important given that temporary and part-time work often has fewer rights and social protections.
Secondly, the great amount of investment required to undertake large data projects, which are labor-intensive and time-sensitive, hinders companies from immediately hiring workers and/or paying fair compensation for the work. Nevertheless, since datasets and AI services are often sold and resold multiple times, companies can establish a mechanism for tracking workers' contributions and appropriately compensating them each time their work is utilized.
Finally, implementing regulations for digital labor platforms to compel them to assume responsibility for the workers in their platforms. This includes measures such as ensuring that workers are not required to pay fees or subjected to preferential access to tasks by paying a quota, prevent algorithms from discriminating against those who reject work, adopt measures such as the "right to disconnect", and granting avenues to platform workers to obtain a “decent price” for their work. A more comprehensive approach would involve platforms hiring workers as employees, covering their social security, providing a minimum salary for a set number of hours, and allowing additional earnings from completed tasks to supplement their income.
To conclude, placing online labor platforms at the forefront of the political focus is important to guarantee that citizens have decent work opportunities in an post-industrial economy, and technological progress that benefits all.
Bibliography:
Berg, J., Furrer, M., Harmon, E., Rani, U., and Silberman, M.S. (2018) Digital labour platforms and the future of work: Towards decent work in the online world, Geneva, ILO
Farrell, D. and Greig, F. (2017) The Online Platform Economy: Has Growth Peaked?, JPMorgan Chase Institute
Gray, M. and Suri, S. (2019) Ghost work : how to stop Silicon Valley from building a new global underclass
ILO (2021) The role of digital labour platforms in transforming the world of work, World Employment and Social Outlook. Accessible at: https://www.ilo.org/global/research/global-reports/weso/2021/WCMS_771672/lang--en/index.htm
Stephany, F., Kässi, O., Rani, U., & Lehdonvirta, V. (2021) Online Labour Index 2020: New ways to measure the world’s remote freelancing market. Big Data & Society, 8(2). Accessible at: https://doi.org/10.1177/20539517211043240