The newest hype around automation is caused by the chatbot, ChatGPT, a software based on a natural language processing (NLP) model developed by the company OpenAI. After the company opened a test version to the public, millions of users started to play with it, letting it write university exams, journalistic articles, poems, and screenplays. Inevitably, the launch of the software has also led to speculations about which jobs and professions it would substitute. “Professors, programmers, and journalists could all be out of a job in just a few years,” The Guardian reported, only one example of many. Leaving aside the real abilities and limitations of the software, it is such formulations that sound familiar and invite skepticism.
The proliferation of digital technologies throughout the last few decades has brought renewed speculations about rising and imminent technological unemployment. According to these predictions, digital technology and automation will be doing away with human labor. While the predictions one or two decades ago focused more on menial and routinized forms of labor, newer developments in AI, lead articles such as the one cited above to predict the dawn of unemployment in professions such as journalism or education. Such automation hypes and discourses, however, are at least as old as capitalism itself. In fact, these discourses have a tendency to return periodically: in the 1930s, 1950s, 1980s, and again in recent years, as for example, Aaron Benanav points in his book Automation and the Future of Work. All of these upticks in the discourse on automation were bound up with real processes of automation that replaced jobs and made workers redundant. After each wave of automation, however, more people than ever before had entered into wage labor.
The Impact of Automation Technologies in the Present
This text is not concerned with the abilities and limitations of ChatGPT and even less so with predictions of which jobs are endangered by such software. Rather, it directs its attention to a few ways by which AI, automation, and digital technologies are impacting labor in the present. Automation in its various forms does indeed replace human workers in the present and has the potential to extinguish even more jobs in the future. However, jobs that are “automated away” often have a curious tendency to reappear at other places, although, often looking very different and taking place in new and hidden production geographies, done by new workforces.
We may think of the digital homeworkers on online gig economy platforms who are crucial for the development, training, and maintenance of self-driving cars. Or about Colombian students remotely steering “autonomous” delivery robots who get stuck while delivering food on the campus of US universities. Or about the tens of thousands of content moderators for social media platforms like Facebook working day and night to enforce the platforms’ regulations against violence, nudity, or hate speech. Behind automation technologies there is often more human labor than expected.
The development of AI applications is based on huge, categorized training datasets, the production of which necessitates large amounts of human labor.
Thorough analysis of the impact of digital technology accordingly demands to go beyond the smokescreen of technological unemployment and into the ways digital technologies are re-ordering both the global division of labor and the ways labor is done every day. Here, we can see new geographies of labor produced by digital technologies and infrastructures, as well as new forms of digital and automated management that amount to a digital rebirth of many of the features associated with classical Taylorism.
AI’s Geography of Hidden Production: Digital Labor Platforms
Today, the online gig economy, often also referred to as cloud or crowdwork, encompasses thousands of platforms such as Amazon Mechanical Turk, Fiverr, and Appen. Millions of digital home-based workers across the globe log into these platforms from their kitchens or living rooms to earn money from the tasks these platforms provide. Working from their personal computers, they constitute a hyperflexible, on-demand workforce that can be accessed and let go in seconds.
These online labor platforms enact new forms of control and flexibility and serve as decentralized sites of digital production that are crucial to many nodes of the global economy, most notably the production and training of artificial intelligence (AI). These workers categorize pictures, test software, transcribe audio recordings, or optimize search engine results. Often hidden from view and dispersed around the globe, these workers nonetheless form a growing component of the digital working class, as well as the political economy of the internet more generally.
Among other things, the development of AI applications is based on huge, categorized training datasets, the production of which necessitates large amounts of human labor. Today, crowdwork platforms are providing millions of hours of hidden labor that is necessary for algorithms that power self-driving cars or allow devices to understand human language. OpenAI, the company that developed ChatGPT, also used crowdworkers to train a bot to play the video game Minecraft. A number of crowdwork platforms have focused their business solely on the booming sector of training data for AI applications, and the great amount of human labor necessary for the training and optimization of AI has become the main factor pushing the dynamics of the online gig economy in recent years. “Training data isn’t labeled or collected on its own. Human intelligence is required to create and annotate reliable training data,” advertises the Australian platform Appen on its website. “Our platform collects and labels images, text, speech, audio, video, and sensor data to help you build, train, and continuously improve the most innovative artificial intelligence systems.” This labor is done by the globally distributed workforce of the platform’s more than one million registered independent contractors working from their personal computers.
What we see behind many sophisticated AI applications is not the end of manual and repetitive labor but rather its global re-organization by help of digital platforms.
Digital technologies and infrastructures such as the means of algorithmic management employed by the platforms allow for a globally distributed and very heterogenous workforce to be inserted in a tightly controlled and coordinated digital production line. What we see behind many sophisticated AI applications is not the end of manual and repetitive labor but rather its global re-organization by help of digital platforms. These tightly controlled and automatically organized labor process can then be understood as a digital re-instantiation of the Taylorist factory.
A New Taylorism in the Digital Factory
It is not just in the gig economy, both in form of cloud work and in form of location-based services such as ride-hailing or food delivery, where we gain a glance at the importance of the algorithmically organized, and often completely automatic organization, control, and measurement of labor. We may think of the app-based labor of Uber drivers or Deliveroo riders but also, say, about the way Amazon warehouses employ digital technologies for the tight organization and granular control of the labor process.
Throughout the world of work, we can observe how digital technology allows for new modes of standardization, decomposition, quantification, and surveillance of labor— often through forms of automated management. A development that could be described as the emergence of a new form of digital Taylorism. The digital renaissance of a managerial logic that was originally tied to the traditional factory and that gains traction exactly at the moment many analysts predict the end of routinized labor in an age of automation, robotics, and immaterial production.
Even if digital technology allows for the rise of classical elements of Taylorism such as rationalization, standardization, decomposition, and deskilling, as well as the precise surveillance and measurement of the labor process, this is, of course, not a simple return of Taylorism. Rather, the phenomenon has emerged in novel ways. Thus, invoking Taylor does not mean to argue for a rebirth of Taylorism in its classical form but rather seek to emphasize how digital technology allows for the rise of classical elements of Taylorism in often unexpected ways. One difference of many is, for example, the temporality of management: While Frederick W. Taylor, Frank and Lillian Gilbreth, and others faced a back-and-forth between their studies and improvements in the production process, digital Taylorism’s horizon is a system of real-time control, feedback, and correction.
Furthermore, networked devices, sensors, and apps have moved Taylor’s time and motion studies outside the enclosed spaces of factories and into the streets and living rooms of workers. These forms of algorithmic management and control of the labor process allow for new forms of control and coordination, and accordingly new logics and geographies of digital labor. In many ways, digital technology is able to take on the spatial and disciplinary functions of the traditional factory and develop new forms of coordination and control that can reach out onto streets to Uber drivers or into the private homes of cloud workers busy with training AI algorithms.
This allows the quick and flexible inclusion of a diverse set of workers in very different situations. Returning to the advertisement of the gig platform Appen, it describes its workforce as “An amazingly diverse group of individuals—students and professionals, mothers and veterans, teachers and gamers—each with a unique perspective to contribute to our global workforce.” The new quality of digital technology and remote algorithmic management allows for the efficient remote cooperation and tight surveillance of a huge number of distributed home-based workers. Digital Taylorism is hence no longer bound to the disciplinary architecture of the factory. Indeed, today’s digital factory can take many different forms, it might be an automotive plant, an Amazon warehouse, or it can also be a gig economy platform.
The Automation of Management Instead of the Automation of Labor?
It is important to acknowledge that the tendency described here as digital Taylorism is not the only or the dominant form of digital labor in the present. Digital technology, especially in its most advanced forms, creates a set of very different labor situations, in which a new digital Taylorism exists alongside other— markedly distinct— labor regimes. The focus on such sites of labor, however, opens a particular perspective on the transformation of labor and capitalism in the digital age as well as fostering a critical view on the questions of AI and automation. Such a view is less about questioning the deep impact of such technologies on the contemporary economy, but rather sharpening our understanding of these impacts. Rather than simply automating human labor away, these technologies are implicated in its global restructuring and re-composition.
Digital technology, especially in its most advanced forms, creates a set of very different labor situations, in which a new digital Taylorism exists alongside other— markedly distinct— labor regimes.
Hereby, what is automated is in many cases management instead of labor, and with this we see new forms of automated organization and control. In the context of the current crises, investment into software for algorithmic management often serves as a cheaper alternative to automation technologies. Here we can see again that automation in the context of globalized capitalism is much less the linear process that many predictions suggest. Rather, it is a turbulent, uneven, and crisis-ridden process where sometimes labor is automated at one point only to reappear at another, often recomposed in geographic and social terms and hidden behind new infrastructures of code and concrete.
