THE POSTINDUSTRIAL PARADIGM

Although the term “post-industrial” was coined sometime between 1913 and 1914 by art historian Ananda K. Coomaraswamy and later developed by architect Arthur J. Penty (Rose Reference Rose1991, 21–4), the context for the mid-century surge in theories of “post-industrial society” began to form in the late 1940s, a period distinguished by breakthroughs in computer and machine technology that, in the business realm, resulted in new solutions to pressing problems of automatic control in production (Leaver and Brown Reference Leaver and Brown1946) and, in the intellectual sphere, helped to propel the cybernetics (Wiener [1948] Reference Wiener1961) and automation (Diebold Reference Diebold1952) movements, which captivated the minds of such influential mid-century thinkers as Martin Heidegger and Hannah Arendt (Simbirski Reference Simbirski2016). Beginning in the late 1950s, as the development of automation continued apace, a range of social scientists turned their attention to a plurality of transformations that ran the gamut from the economy and culture to politics and technology as a means of unraveling the distinctive character of an emerging society set apart by staggering affluence and a sweeping occupational shift from industrial manufacturing to informational services (Brick Reference Brick1992, 350–62; Kumar [1995] Reference Kumar2005). Whether they chose to describe this new social formation through the qualifier “leisure” like David Riesman (Reference Riesman, Larrabee and Meyersohn1958), “affluent” like Kenneth Galbraith (Reference Galbraith1958) and Gunnar Myrdal ([1963] Reference Myrdal1965), “post-industrial” like Peter Drucker (Reference Drucker1969) and Daniel Bell ([1973] Reference Bell1999), “technetronic” like Zbigniew Brzezinski (Reference Brzezinski1970), or “informational” like Manuel Castells (Reference Castells1989), they all agreed that this new society—whatever it was called—ushered in “an age of discontinuity, marking an end to the industrial era” (Jones Reference Jones1982, 11).

Among these theoretical exercises in social forecasting, Daniel Bell’s The Coming of Post-Industrial Society was arguably the most representative and comprehensive treatment of what I call the “postindustrial paradigm.” His rigorous discernment of the social transformations that defined this impending society indexed a pervasive feeling among liberal and Marxist thinkers of being caught up “in the midst of a vast historical change” ([1973] Reference Bell1999, 37) characterized by a dramatic departure from industrial capitalism in terms of the presiding occupational structure of society as well as the prevailing regimes and general qualities of work within affluent economies structured by knowledge, information, and technology. Specifically, then, my argument builds on two central views espoused by liberal and Marxist theorists of postindustrial capitalism alike: first, that the organization of production and the character of work in a postindustrial society are diametrically opposed to the disciplinary management regimes, manual work routines, and fragmented tasks characteristic of Fordism; and second, that immaterial forms of production based on knowledge, information, and technology are the dominant sources of value in postindustrial economies, while a skilled, technical, scientifically trained, and culturally fluent intellectual workforce permeates and reshapes the class structure.Footnote ² To be sure, most adherents to each of these frameworks were careful to stress that postindustrial society did not displace factories, manual labor, and disciplinary management practices entirely, but fundamentally altered their foregoing circumstances by circumscribing their socioeconomic dominance wherever knowledge and information took command.

According to Bell, the factory had lost its pervasiveness in postindustrial society while the void left by the dislocation of manufacturing and manual toil gave way to the communicative, social, and affective encounters between skilled individuals who mediate knowledge-based services (Bell [1973] Reference Bell1999, 15–7, 126–8, 162–3, 488). And although explanations for the erosion of capitalism’s industrial structure vary considerably, from the widespread outsourcing of manufacturing to the Global South and the displacement of assembly-line labor by automation to the neoliberal assault on organized labor, recent liberal (Cohen [2006] Reference Cohen and McCuaig2009, 33–4, 87; Kelly Reference Kelly1999) and Marxist (Berardi Reference Berardi, Cadel and Mecchia2009, 88–94, 192–3) scholars have upheld Bell’s assessment that the new, postindustrial economy represents a paradigmatic rupture with Fordism in terms of the prevalent nonhierarchical organization of production and the predominantly informational character of work. For Bell, the main catalyst for this “revolution in the class structure of society” (Bell [1973] Reference Bell1999, 125–6) lay in the preeminence of a new kind of theoretical knowledge, which replaced “trial-and-error-tinkering” and empiricism (487), as a governing source of power and capital that spawned innovations in technology, employment, and social relations alike (18–20, 343–4). To the extent that theoretical knowledge could be codified and embodied in automatic machines, a new form of intellectual technology brought to bear the prospect of replacing human judgment by algorithms capable of solving complex problems through automated decision rules (29–31). While such a division of labor between humans and machines exemplifies the fragmentation of microwork today, for postindustrial thinkers, the promise of automation was rather to relieve workers from the drudgery of routine and repetitive tasks in clerical and manual work (Castells Reference Castells2010, 257–8). In short, even if postindustrial thinkers acknowledge that factories are as large and important as ever across numerous industries and regions of the world, they agree that the forms of work and regimes of control characteristic of the Taylorist assembly line have become “an historical relic” in postindustrial societies in general, and especially within the knowledge-based, information-driven, and technological sectors of advanced capitalist economies, such as those engaged in the production of digital commodities (258).

For a cohort of Marxist thinkers, the dominance of emerging postindustrial forms of work, workers, and workplaces set the stage for a renewed critical theory of capitalism that uprooted the conventional focus of classical Marxism on factories and industrial workers in favor of the discovery and analysis of new working-class subjects, new spheres of work, new forms of domination, and new possibilities for collective political action within cultural and intellectual cycles of production (Deleuze Reference Deleuze1992; Marcuse [1964] Reference Marcuse1991, 12, 35; Mills Reference Mills and Horowitz1963, 236).Footnote ³ For C. Wright Mills ([1959] Reference Mills and Horowitz1963), in particular, a critical theory of postindustrial capitalism required dislocating Marxism from its industrial footing in the factory and anchoring it instead in what he called the “cultural apparatus,” an immaterial labor process scattered across the realms of art, science, learning, and entertainment (406). Echoing Bell’s ([1973] Reference Bell1999) claim that the structural features of postindustrial society were no longer labor and capital but knowledge and information (xci–xcii), Alain Touraine ([1969] Reference Touraine and Mayhew1971) and André Gorz ([1980] Reference Gorz and Sonenscher1982) took up Mill’s project by centering cultural and intellectual workers as the backbone of an evolving class formation, while Serge Mallet’s ([1963] Reference Mallet, Shepherd and Shepherd1975) study of the labor process in France’s technological and chemical sectors confirmed the postindustrial hypothesis that knowledge, cooperation, and affective bonds had become standard attributes of work even in the least skilled occupations, thus displacing the disciplinary management regimes, manual toil, and fragmentation that typified the Fordist factory (59–60, 66–7, 113).

Another influential contribution to a critical theory of postindustrial capitalism came from the Italian autonomist thinker Antonio Negri. Through an imaginative reading of Marx’s Grundrisse, Negri challenged the most fundamental categories of Marxist thought—from the labor process to the wage form—while offering an expansive account of social relations, processes of valorization, mechanisms of exploitation, and practices of resistance beyond factories, industrial workers, and material production (Negri Reference Negri and Negri1988; Reference Negri and Newell1989, 215; Wright [2002] Reference Wright2017, 86–92, 150–3). The widespread uptake of autonomist ideas in the late 1990s and early 2000s, especially by Anglo-American Marxists, coincided with the extensive outsourcing of manufacturing, the escalation of venture capital in the technological sector, and the rise of the internet (Virno and Hardt Reference Virno and Hardt1996). Largely prompted by Negri’s collaborations with Michael Hardt, this current of critical thought inaugurated a conceptual grammar fit for understanding and critiquing “cognitive capitalism” as an unmitigated departure from Fordism marked by the erosion of its “industrial and disciplinary regime” (Reference Hardt and Negri2017, 171) and the rise of immaterial production, which had become “hegemonic over all the other valorization processes” (Reference Hardt and Negri2009, 25).Footnote ⁴ This autonomist critique of cognitive capitalism can be summed up in three related claims. First, the ascendancy of immaterial labor as a biopolitical transformation of capitalist production characterized by linguistic, affective, intellectual, and cultural forms of work generative of subjectivities, capacities, and knowledge (Hardt Reference Hardt2006; Lazzarato Reference Lazzarato, Virno, Hardt, Colilli and Emery1996). Second, the proliferation of a collective stock of intellectual assets, such as linguistic, affective, and cultural competencies, known as “the general intellect” (Hardt and Negri Reference Hardt and Negri2017, 117; Spence and Carter Reference Spence and Carter2011). And third, a shift in the planning of production away from capitalist centralization and toward indirect forms of value extraction—what Marx dubbed “formal” subsumption—in which capital is released from the onus of directly organizing, controlling, and disciplining the labor process (Hardt and Negri Reference Hardt and Negri2017, 133, 143, 171–2, 222; Vercellone Reference Vercellone2007). These qualities of cognitive capitalism lay bare the intimacies between liberal and Marxist theories of postindustrial capitalism in terms of their corresponding views about the organization of production, the character of work, and the primacy of knowledge in cycles of valorization, which my theory of the factory model and the protocols of digital production call into question.

More recently, a growing scholarly literature on the digital labor process has offered a robust set of empirical and theoretical resources for critically re-examining the central postulates of the postindustrial paradigm (Altenried Reference Altenried2020; Betancourt Reference Betancourt2016; Gray and Suri Reference Gray and Suri2019; Jones Reference Jones2021; Munn Reference Munn2022). By contrast to the speculative theorization and social forecasting prevalent among theorists of “post-industrial society” and “cognitive capitalism,” the general orientation in recent accounts of digital capitalism has been to anchor theory and critique on rigorous studies of existing social relations, working conditions, and lived experiences of both workplace domination and collective resistance (Christiaens Reference Christiaens2022) across an ample realm of productive activities within the internet economy, from logistical operations (Allison and Reese Reference Allison and Reese2023) and content moderation (Roberts Reference Roberts2019) to gaming (Woodcock Reference Woodcock2019) and data work (Miceli and Posada Reference Miceli and Posada2022; Muldoon et al. Reference Muldoon, Cant, Wu and Graham2024). The result is a far-reaching rendering of economic life—of domination and agency—under digital capitalism starkly at odds with earlier theories of postindustrial society as much in regards to the toilsome and precarious character of digital labor as to its estrangement from the cultural, intellectual, and informational content of digital production. In the last two sections, I build on key insights from this literature as a means of grounding my theory of the factory model and the protocols of production on the social realities of the digital labor process.

PROTOCOLS OF PRODUCTION: A THEORY OF THE FACTORY MODEL

Traditionally, the factory system has been defined as a large, unified production unit where workers are assembled and supervised by a common authority and work is commanded by a central source of mechanical power, which most economic and social historians agree only came into vogue in eighteenth-century Britain (Freeman Reference Freeman2018; Landes Reference Landes1998, 186). While manufactories—where the labor process was similarly combined and controlled—predate the factory system by centuries (Safley and Rosenband Reference Safley and Rosenband1993), these workplaces are often distinguished from factories insofar as they did not employ a central system of power-driven machines in production (Landes Reference Landes1986, 603). This association between technology and the factory system is a widely accepted rendering of the meaning and periodization of factories that continues to hold sway among critics who recognize the factory’s fundamental organizing power in digital production (Altenried Reference Altenried2022).Footnote ⁵ But if we consider the long-running history of large-scale, consolidated industrial production, which goes back at least to the sixteenth century (Peacock Reference Peacock2006), it is clear that the use of automated or “self-acting” technologies is a relatively recent development. In this sense, the factory system—as it is conventionally understood—is one particular historical instantiation of unified production delineated by its use of centralized mechanical or, more recently, electronic and algorithmic power.

My own interpretation of the factory resides in the consistent sociopolitical factors shared by these models of amalgamated production, irrespective of the technology they employ and the commodities they produce. I therefore center my definition of the factory on what I take to be the constitutive elements of its production model: capital, labor, and control. I argue that the factory’s particularity inheres in its historically stable ability to place the labor process under capital’s control through an integrated apparatus of discipline, supervision, and fragmentation that has been a persistent feature of consolidated production since the early modern period, from workhouses (Pinheiro Reference Pinheiro2022) and manufactories (Berg Reference Berg1994) to dockyards (Cooper Reference Cooper1984) and plantations (Mintz Reference Mintz1985). Indeed, from the perspective of capital and control, “the fully developed factory was really not different from the ‘protofactory,’ except for the intensification of [its] crucial elements” (Freudenberger and Redlich Reference Freudenberger and Redlich1964, 382). And while factories vary significantly across periods, regions, cultures, industries, and states of technological development, the factory model’s distinctive contribution to the organization of capitalist production has been, and remains, to control the labor process through what I call “protocols of production”: formal and objective directives for standardized procedures and behaviors that are mediated and enacted by a combination of diverse spatial arrangements, technologies, and management regimes in order to discipline workers, fragment tasks, and supervise the labor process. As much in physical as in digital form, the factory is a workplace in which workers are directly organized by a superior authority to perform tasks that generate value for their employer, whether the end product is a material good, a digital commodity, or a logistical task. Situated at the bottom of the factory’s hierarchal chain of command, workers are required to follow a script and obey the rules laid out by management: time and productivity are assiduously tracked, quantified, and monetized; physical movements and cognitive operations are determined and bound by discrete, fragmented tasks; compliance to protocols is ordained by regimes of discipline and supervision; and production is technically standardized. The particular organization of the labor process in a factory is a means to the end of capital accumulation, which presses the factory to constantly adapt its form by applying technological innovations, scientific improvements, and more efficient spatial configurations to production while constantly prying open new sources of exploitable, precarious labor. But whatever the factory produces, however it produces it, its model of extracting and generating value has historically entailed capital’s consolidated control of labor through techniques of discipline, supervision, and fragmentation that reproduce alienation and precarity as concrete social qualities of its labor process.

Borrowing the concept of “protocol” from networked computing, I use the term “protocols of production” to theorize the ways in which the factory model purveys capital’s control over labor. According to media theorist Alexander Galloway (Reference Galloway2004), a computer protocol is a “distributed management system that allows control to exist within a heterogeneous material milieu” (8). Based on a given set of rules that “outline specific technical standards” (6), protocols “govern the set of possible behavior patterns” and “enact an agreed-upon standard of action” (7). Protocols therefore manage, regulate, and control; they encode information, execute standardized operations, and “create the core set of rules from which all other decisions descend” (245). In the context of a factory, protocols are implemented to order production—space, people, and things—through a synthesis of variegated technological systems, spatial arrangements, and management regimes specifically combined to administer control over the labor process. And if protocols “always operate at the level of coding” (8), then they are, as Eugene Thacker remarks of all computer codes, “always enacted” (xii). The factory’s protocols of production also operate at the level of coding insofar as its own codes define “a set of procedures, actions, and practices designed in particular ways to achieve particular ends in particular contexts” (xii). “In the factory code,” as Marx wrote in Capital ([1867] Reference Marx and Fowkes1990), “the capitalist formulates his autocratic power over his workers like a private legislator,” as a “factory Lycurgus” whose factory code is a pure “emanation of his own will” (549–50). “This code,” Marx continues, “is merely the capitalist caricature of the social regulation of the labour process which becomes necessary in co-operation on a large scale and in the employment in common of instruments of labour, and especially of machinery” (550). Like a computer code, the factory code is a performative text that commands, authorizes, and restrains; it imposes demands on the reader to perform precise, formal, and standardized courses of action in pursuit of a specified goal: the accumulation of capital through the extraction of value. As a technique of controlling production by discipline, supervision, and fragmentation, the factory code is the essence of the factory’s protocols. Marx’s allegorical rendering of the capitalist as a “factory Lycurgus”—the Spartan lawgiver—captures capital’s autocratic authority over production embodied in the factory’s protocols, the enactment of which is mediated by technology, enforced by management, and finally performed by workers.

Despite the overwhelming association of factories with assembly lines and heavy machinery, technology is an ancillary instrument of the factory’s protocols and therefore insufficient to delineate its distinctive mark. Rather than displacing or radically altering the factory model’s means and goals, automated technologies—from mechanical machines to algorithmic infrastructures—only amplified its tendency to integrate previously distinct nodes in the supply chain while perfecting and rationalizing its propensity to subsume the labor process under its control. The historical triumph of the factory over competing organizations of production was therefore not primarily a product of its competence in increasing efficiency by dint of technological innovation (Marglin Reference Marglin1974, 62, 84), but of its efficacy in commanding the labor process through discipline, fragmentation, and supervision. Even Andrew Ure (Reference Ure1835), one the most ardent advocates of the factory’s claim to technological progress, conceded that the invention responsible for inaugurating the modern factory system was not Richard Arkwright’s pioneering self-acting spinning frame, but his “code of factory discipline,” which succeeded in solving the “main difficulty” of factory production: “training human beings to renounce their desultory habits of work, and to identify themselves with the unvarying regularity of the complex automaton” (15). Arkwright’s groundbreaking contribution to the factory system, then, lay in devising protocols of production that rendered his technological invention useful in subduing “the refractory tempers of work-people accustomed to irregular paroxysms of diligence” (16). Capital’s control over labor in the factory was therefore, as Ure concluded, not the result of technology but of “obedience to Arkwright’s polity.” Only under the command of protocols that discipline, supervise, and fragment production would further technological innovations be useful in perfecting capital’s control over the labor process. By using automatic machinery as a medium for its protocols, the industrial factory was able to rule out human judgment from production and consequently deskill, standardize, and devalue labor by supplanting the worker’s allegedly irregular hand and intractable will with a machine “so self-regulating, that a child may superintend it” (19). Indeed, deskilling through the decomposition of tasks and the annihilation of human judgment was the lauded effect of automated technologies that, “in embodying handicraft dexterity and intelligence in a machine” succeeded in “substituting cheap and docile labour for what is dear, and sometimes refractory” (150). What Ure deemed the progress of the factory system (x), then, was owed not to self-acting machines alone, but to their application of the factory’s protocols of production.

Similarly, the emblematic contributions to the modern factory system by Samuel Bentham’s panopticon in the late eighteenth century and Frederick Taylor’s system of scientific management in the early twentieth century were based less on technological innovations than on a rational reorganization of the protocols of production. That is, both Bentham and Taylor applied existing technologies to amplify capital’s control over the labor process. While the novelty of Bentham’s panopticon inhered in a more efficient spatial arrangement for supervision (Ashworth Reference Ashworth1998), Taylor’s improvement consisted in a forensic approach to task management that dressed labor discipline and fragmentation in the garbs of scientific reason (Reference Taylor1911b, 139–40). Indeed, Taylor’s theory of scientific management, in particular, enunciated the basic feature of the factory’s protocols—control—by systematizing and quantifying production into fragmented, simplified, and regular tasks (69–73, 96–7, 121, 146, 181) while also eliminating human judgment from task-work (Taylor Reference Taylor1911a, 28, 72–9, 102–4, 125–33). Combined, these strategies resulted in the heightened alienation of workers from their work while reproducing the social elements of precarity in the workforce upon which Taylor’s system depended (Aitken Reference Aitken1960, 12). Historically, the social effects of scientific management were palpably evident in Ford’s auto plants, where protocols of production were mediated by a series of technological innovations, from the assembly line at Highland Park in 1910 to the automatic engine at the Cleveland Foundry in the 1950s—where the term “automation” was coined by the Company’s Vice President of Manufacturing (Diebold Reference Diebold1952, ix). Indeed, Ford’s strides in automated manufacturing were technical applications of protocols that exacted absolute conformity and obedience from line workers, consequently uprooting their sphere of autonomy in production by circumscribing the requirement for tasks to the most basic human capacities and movements (Biggs Reference Biggs1996, 130–4). Socially, Taylor’s and Ford’s contributions to the factory model produced familiar results: an expansive precarious workforce, high turnover rates, the active recruitment of untrained immigrants, and a constant pool of surplus workers standing by at the gates (Biggs Reference Biggs1996, 134; Meyer Reference Meyer1981, 67–94). From Arkwright and Bentham to Taylor and Ford, all the way up to Google and Amazon, the discipline, supervision, and fragmentation provided by the technological, spatial, and administrative implementation of the protocols of production have allowed the factory to fulfill its social function of capital accumulation by placing the labor process under its hierarchical control while reproducing a precarious workforce that becomes increasingly alienated from the materials, activities, and content of their work to the extent that the labor process becomes more fragmented, supervised, and disciplined. As we come to grips with the social relations of production inside data centers and microwork platforms, we will find ourselves in the midst of a digital labor process at once thoroughly anchored in and utterly dependent upon the factory model’s protocols of production.

ABSENT FACTORIES: INSIDE THE HUMAN ABODE OF AUTOMATION

As one of Silicon Valley’s largest employers, Google hosts a substantial segment of its workforce across three million square feet of office space at the company’s global headquarters. Commonly referred to as a “knowledge campus,” the Googleplex comprises an opulent recreational and technological infrastructure that epitomizes digital capitalism’s knowledge-based mode of production and a postindustrial work ethic centered on personal fulfillment, cooperation, flexibility, and amusement. According to some critics, the company couches extraction in the language of play through a horizontal organizational structure and a consent-based management regime rooted in its “don’t-be-evil” ideology and fulfilled materially by its leisurely amenities (Levy Reference Levy2011, 121–66; Vaidhyanathan Reference Vaidhyanathan2012, 1–12). On this account, Google appears to confirm the progressive, postindustrial trend in management discourse characteristic of capitalism’s “new spirit” (Boltanski and Chiapello [1999] Reference Boltanski, Chiapello and Elliott2005). And although such renderings of Google as a quintessential expression of postindustrial regimes of control do well to describe its means to organize and discipline skilled knowledge workers (Fuchs Reference Fuchs2014, 223–30), they tend to miss the crucial fact that, throughout the early 2000s, the company deployed management practices characteristic of the factory model to preside over workers who “flipped pages in time to a rhythm-regulated soundtrack” in a hidden data factory in the outskirts of the Googleplex (Irani Reference Irani2015b).

In 2007, Andrew Norman Wilson (Reference Norman Wilson2016), then a part-time employee at Google, recalls this very cohort of data workers—predominantly people of color—leaving a secluded building as though “a factory bell had just gone off.” Identifiable by distinctive yellow badges, most of these subcontracted workers—known by the codename “ScanOps”—were hired to scan printed matter for Google Books, page by page, from 4:00 a.m. to 2:00 p.m. Google’s color-coordinated badges—green, white, red, and yellow—are a visual representation of the company’s employment hierarchy; they determine the places, services, and facilities employees may and may not access across the Googleplex. Unlike yellow-badge workers, “Googlers” wearing red, white, and green badges make up the company’s knowledge-based workforce and have access to most spaces and amenities across the headquarters. “Purposefully kept separate,” yellow-badge workers were by contrast “not allowed [to] set foot anywhere else on campus, except for the building they worked in” (Norman Wilson Reference Norman Wilson2011; Ptak and Norman Wilson Reference Ptak and Wilson2013, 126). Management enforced the secrecy and confinement of this workforce through a rigid apparatus of discipline and surveillance that included punitive measures to restrain data workers from disclosing any detail about their work and workplace (Norman Wilson Reference Norman Wilson2011). Their yellow badges signified “getting paid $10 an hour, going to the bathroom only when a bell indicated it was permissible to do so, and being subject to a behavioral point system that could lead to immediate termination” (Norman Wilson Reference Norman Wilson2016). Not only was the regime of discipline and supervision that Google deployed toward its data workers decidedly at odds with its organization of skilled labor at the knowledge campus, but the working conditions of yellow-badge workers were distinctly precarious: high turnover rates, long shifts, low wages, lack of security and benefits, and a taxing—mostly manual—fragmented work routine. Despite Google’s relentless efforts to stylize digital production as horizontal, skilled, creative, and edifying, its data factory disclosed instead a labor process strictly organized by protocols of discipline, supervision, and fragmentation. Inert and quiet at their scanning stations in a room without windows, data workers obeyed three directives that structured their 10-hour shifts: “press button, turn page, repeat” (Ptak and Norman Wilson Reference Ptak and Wilson2013, 129).

Google’s data factory reflects broader industrial patterns in digital production. It followed an earlier digitization initiative, the Million Book Project, which consisted of a massive human workforce and a vast logistical infrastructure, from scanning centers in India and China rendering thirty-four million pages a day to transcontinental container ships ferrying hard-drives and printed materials halfway across the world, which laid bare—if nothing else—the indispensable industrial fabric of the global knowledge economy (St. Clair Reference St. Clair2005, 1196–7). Ironically, this digitization process is exceedingly more labor-intensive, fragmented, and manual than the production of physical books (Roth Reference Roth2001). While industrial printing has been at the forefront of automation since the nineteenth century (Fyfe Reference Fyfe2012, 55–64), the current process of digitizing books relies on an assembly-line production method long since abandoned by commercial publishers, thus marking the emergence of “digitization on an industrial scale” (Milne Reference Milne, Earnshaw and Vince2008, 5). In principle, then, Google Books encapsulates digital culture, immaterial labor, and the knowledge economy at their most advanced state; in practice, it illustrates the extent to which the factory model—with all its manual drudgery, discipline, surveillance, alienation, and precarity—remains a fundamental element of digital capitalism. Indeed, Google’s application of the factory model in its covert data center registers a pervasive trend in data-driven sectors of the global economy, from “gaming factories” that combined employ over four hundred thousand gamers worldwide, predominantly in China (Dibbell Reference Dibbell2016; Heeks Reference Heeks2010), to the field of supply chain management where companies like Walmart and Amazon have adapted the factory model to the organization of logistical work. Like Google’s data center, Amazon warehouses are highly technological workplaces where scanners are not only ubiquitous productivity tools that augment the capacity of workers to identify items across cavernous warehouses, but also instruments of control through which supervisors direct, evaluate, track, and discipline workers, tethering them to the factory’s protocols as the omnipresent “eyes” of management (Lecavalier Reference Lecavalier2016, 156–62). And like Google’s ScanOps workers, packers at Amazon are required to stand still at automated workstations and engage in highly repetitive tasks in routine shifts of 10 hours (Struna and Reese Reference Struna, Reese, Alimahomed-Wilson and Reese2020, 91–6). Within the sprawling network of global supply chains, each distribution center employee becomes a data point that must be “managed, hoarded, and deployed” as—in the words of one Amazon manager—“a robot, but in human form” (Lecavalier Reference Lecavalier2016, 156, 158). These aspects of the factory model’s digital immersion are just as prevalent in the “virtual” world of the internet.

While one of the most anticipated developments of automation since the 1940s has been its prospect to end the scourge of human drudgery on the assembly line, the postindustrial society slated to fulfill this promise never arrived (Frey and Osborne Reference Frey and Osborne2017). To compensate for the shortcomings of automation, technological companies masquerade a vast supply of human workers as a new “technology” called upon to perform repetitive digital tasks that require little technical skills and training (Irani Reference Irani2015a, 225). Insofar as it allows companies to hire someone for a few minutes, “pay them a tiny amount of money, and then get rid of them” (Prassl Reference Prassl2018, 4), MTurk’s algorithmic infrastructure has lent form to the most efficient configuration of the factory model to date, one that would be unthinkable in physical space. Indeed, the networked systems of algorithmic management deployed by MTurk to track productivity, monitor and assess workers, measure work rates, standardize production, fragment tasks, and steer workflow mark not only the factory model’s adaptation to the online workplace, but its unprecedented amplification and intensification in terms of the precision of work metrics, the scope of surveillance, and the consolidation of capital’s control over the labor process (Altenried Reference Altenried2020, 149–50; Reference Altenried2022, 93–106, 120). MTurk’s infrastructure organizes production through a software architecture that connects microworkers (“Turkers”) to employers (“Requesters”) who are given the technical means to evaluate output, monitor performance, and adjudicate how much—if anything—to pay Turkers based on fulfillment rates, accuracy standards, delivery targets, and output quality (Altenried Reference Altenried2022, 104; Gray and Suri Reference Gray and Suri2019, 13, 70–81, 124–30; Irani Reference Irani2013, 725–7). MTurk’s infrastructure illustrates the factory model’s adaptability to multiple formats regardless of the technology it employs and the commodities it produces. Although a distributed human–computer network powers MTurk’s infrastructure, the specific ways in which Amazon disciplines, supervises, and fragments the labor process is a social, political, and juridical organization of labor irreducible to this technology. By combining the legal categories of independent subcontractors and the piece-rate wage form with its algorithmic infrastructure, MTurk is able to create both a precarious workforce and the technical means to control it (Altenried Reference Altenried2022, 107, 120).

MTurk’s digital embodiment of the factory model is nowhere more evident than in its patent, which describes the platform’s infrastructure as a “hybrid machine/human computing arrangement” composed of “a central coordinating server and a number of human operated nodes” (Harinarayan, Rajaraman, and Ranganathan Reference Harinarayan, Rajaraman and Ranganathan2007, 2; emphasis mine). This description evokes the conventional definition of factories as units of production in which human labor is regulated by “a central, typically inanimate source of power” (Landes Reference Landes1998, 186; emphasis mine). If in the nineteenth century, the common shorthand for the factory was a self-acting “automaton” (Marx [1867] Reference Marx and Fowkes1990, 502; Ure Reference Ure1835, 13) and, in the early twentieth century, industrial engineers designated the factory as “the master machine” (Biggs Reference Biggs1996, 49, 53), MTurk’s inventors refer to their architecture of production, synoptically, as “a computer” or “computer system” that “decomposes a task […] into subtasks for human performance” (Harinarayan, Rajaraman, and Ranganathan Reference Harinarayan, Rajaraman and Ranganathan2007, 2). In describing MTurk’s computing arrangement as a technology that assembles “humans to assist a computer system to solve particular tasks” thereby “allowing the computer system to solve the tasks more efficiently,” the inventors categorically place the agency of production on the computer while understating its human component. This is, again, a digital redeployment of the conceit used by industrial engineers to depict factory workers as attendants to the machine or, in the words of an efficiency engineer in 1912, to “ignore the human element entirely” and “describe the people handling the operations as […] animate machines” (Biggs Reference Biggs1996, 51). At the level of our screens, then, microwork platforms are virtual workspaces powered by algorithms and automated technologies; beyond our screens, however, they are central nodes of data production within a prodigious, global network of physical infrastructures—data centers, server farms, gaming factories, and distributions warehouses—whose predominantly contingent and precarious workforce is tasked with fulfilling the internet’s incessant logistical and productive demands (Ensmenger Reference Ensmenger, Mullaney, Peters, Hicks and Philip2021; Irani Reference Irani2013).

FACTORY DOMINATION: ALIENATION AND PRECARITY

If factories are “sensuous embodiments” of capital’s productive power (Rofel Reference Rofel1992, 103), then the differences in design and applications of technology at Google and MTurk illustrate the ways in which the factory’s embodiment of productive power can take shape through a plurality of forms. Historically, the particular spatial configurations of a factory have promoted correspondingly specific practices of control, exploitation, and commodification. Whereas in nineteenth-century British factories, work began the moment workers entered the factory and continued until they left, in German factories of the same period, work began not when workers entered the factory but only once they were engaged in production (Biernacki Reference Biernacki1996, 93–144). On the surface, these are two disparate modes of delimiting work. In the first template, the boundary of employment is marked by the position of labor in space, while in the second, work is determined by the use of labor in production (125–6, 141–2). But a more meaningful difference is at stake here: each approach to ascertaining where or when work begins and ends also betrays a complementary approach to domination based on capital’s control of space and time. While in nineteenth-century British factories subordination turned on the “momentary regulation of workers’ bodies as workers stepped over the factory threshold,” German employers “used workers’ bodies as a marker of the continuous alienation of the labor power lodged in the person of the worker” (127). This distinction does well to illustrate the different means of control and exploitation at Google and MTurk.

While physical space delineates the threshold of employment and exploitation at Google’s data center, MTurk’s digital factory extends the frontier of work and subordination beyond fixed space. In other words, the limits of domination at Google are bound by the data center as a circumscribed and supervised workplace; Google begins to discipline, supervise, and regulate workers as soon as they step into the factory’s premises. By contrast, the boundaries of the factory at MTurk are digital and thus ubiquitous with regard to physical space; the platform’s threshold of subordination clings to the labor time of workers wherever they may be. MTurk’s approach to delimiting domination is therefore to do away with the physical space of the factory altogether; it effectively retains practices of disciplining labor at a distance by blurring the divide between the spatial-temporal spheres of work and nonwork. In releasing supervision and discipline from the limitations imposed by physical space, MTurk’s extraction of value from labor unfolds through a remote regime of discipline, surveillance, and fragmentation; the “space” upon which capital appropriates value from labor is thus vested in the worker’s body-at-work. The contrasting layouts and uses of technology in these data factories are a testament not only to the myriad forms through which factories can embody productive power, but just as much to the assorted techniques of control, exploitation, and commodification to which these unique designs and technologies can lend form.

Above all, the factory’s technical makeup, its social organization of labor, and the protocols that call its production process to action transmit and enforce a script of alienated work to workers who in turn live out that text in their capacity as producers. Indeed, the technical conditions of the factory’s architecture, be it a physical building or an online platform, are reifications of a script that yields analogous results in terms of alienation; buildings and technology are merely the perceptible shapes of a work process that ultimately produces labor as a commodity (Biernacki Reference Biernacki1996, 141). As factories, Google and MTurk are concrete sites of alienation whose designs, technologies, and protocols are means to the factory’s social function and most basic end of capital accumulation. Like Arkwright’s mill and Ford’s auto plant, Google and Amazon perform this function by controlling the labor process. Yet, what is distinctive about MTurk’s platform is that it hones the exercise of control to an unrivaled pitch of perfection. With unprecedented expediency, accuracy, and frugality, MTurk brings workers to work, fragments their labor, assigns them their work, rewards or punishes them for their performance, and disciplines them through an online rating system that lifts surveillance and control from its spatial limitations. The governing conceit behind Jeremy Bentham’s inspection principle—of “seeing without being seen” (Reference Bentham1791, 21)—is nowhere as close to his ideal than in MTurk’s “algocratic” panopticon where software code becomes “the key to governing globally dispersed labor” (Aneesh Reference Aneesh2009, 347).

Another key distinction between both data factories is that, at MTurk, alienation is not only a consequence of how the platform controls production, but also of the uniquely abstract commodity it produces. In a brutally literal expression of the commodification of labor, the metadata of data work—the information about how workers work, how they behave, how frequently they log in, and how fast and accurately they complete tasks—is itself packaged as a commodity. The immense value of this metadata for Amazon Web Services inheres in the company’s applicability of this information as a means of tightening control, surveillance, and discipline over labor not only at MTurk but across all branches of the business (Jones Reference Jones2021, 74–5). One result of this tautological work of producing data about data production is a twofold erasure of the worker. On the one hand, the human elements of thought and judgment in the worker’s operations are erased by the technical demands of a task so minutely fragmented that its content is exceedingly abstract and void of any intelligible human meaning while its final product is utterly opaque to the worker. This alienation is not merely incidental to this type of work; it is rather shored up by the technical aspects of the platform that reinforce the alienated nature of microwork, including the impersonal quality of domination by anonymous Requesters and algorithms, the abstract nature of meticulously fractured tasks, the imposed isolation of workers, and the engineered opacity of the most basic aspects of production, from how and on what grounds algorithms make the decisions that govern the quantity, speed, and frequency of work, to the obscure purpose of their tasks (Jones Reference Jones2021, 65, 69–70). Insofar as it “deprives the work itself of all content” and “confiscates every atom of freedom, both in bodily and in intellectual activity” (Marx [1867] Reference Marx and Fowkes1990, 548), MTurk carries forth the alienated labor process of mechanization in the industrial factory. On the other hand, microwork performs a second, more literal erasure of the worker, one that ensues from what is, in the long run, the most valuable use of microworkers from the perspective of capital: training the artificial intelligence poised to replace them (Tubaro, Casilli, and Coville Reference Tubaro, Casilli and Coville2020).

Although the technical demands of most microwork tasks such as transcribing texts or rating search engine results tend to be minimal, many assignments also call on workers to exercise cultural fluencies and emotional sensibilities such as those required to, say, interpret and categorize “the sentiment expressed in a comment” (Irani Reference Irani2015a, 228; Kost, Fieseler, and Wong Reference Kost, Fieseler and Wong2018, 101). MTurk’s patent recognizes this by stipulating a division of labor between algorithms and workers based on the general human aptitude to parse cultural objects (Harinarayan, Rajaraman, and Ranganathan Reference Harinarayan, Rajaraman and Ranganathan2007, 1). As a result, algorithms are assigned the role of fragmenting tasks, delegating them to human workers, and collecting the finished output. This scenario points to a dialectal tension within microwork platforms: in their role as mediators of cultural production they depend on the very sensibilities and aptitudes of human workers that their infrastructures undermine. MTurk’s protocols of production reduce the potential linguistic, technical, intellectual, and aesthetic qualities of digital labor to an atomized, repetitive form of mindless “ghost-work” while allowing the creative substance of the finished commodity to be appropriated by the skilled “innovators” who own the rights of all content produced on the platform (Altenried Reference Altenried2022, 104; Gray and Suri Reference Gray and Suri2019). Microworkers are therefore rendered “menial” laborers by the protocols and infrastructure that structure their relations with employers and the cultural content of the objects that, despite helping to produce, they encounter as abstract and alien things (Irani Reference Irani2013, 728–34). So, even if microworkers produce immaterial goods replete with knowledge and cultural information, MTurk precludes them from accessing the intellectual and aesthetic qualities of the materials they work on, which are torn asunder into meaningless data fragments by a digital assembly line. As a result, the platform’s protocols divest microwork from the potentially transformative types of subjective, affective, and creative investments generally attributed to cultural, artistic, and intellectual labor by theorists of postindustrial capitalism.

Finally, while the data center is immediately recognizable as a traditional factory where production is organized and controlled within a confined space, MTurk’s ploy to keep workers at work outside the factory’s premises is just as conventional. This too was a standard practice in nineteenth-century factories, many of which were composed of various departments within and beyond their physical premises. One of these departments is what Marx ([1867] Reference Marx and Fowkes1990) calls “the modern domestic industry,” which takes the form of an “external department of the factory, the manufacturing work-shop, or the warehouse” where the factory’s division of labor and production process are reproduced and continued “on a very large scale” beyond its walls (590–1, 600). As a result, the factory system commands not only the workers “it concentrates in large masses at one spot,” but also “the outworkers in the domestic industries” (591). Like MTurk, these “outside departments” of the factory are characterized by high levels of precarity for labor and accumulation for capital (608). As one of the branches of large-scale industry, the “extended factory” is not antithetical to the factory but an intrinsic and essential part of its production model (590–1). In short, the factory system has been and continues to be able to exploit a vast contingent of precarious workers by extending its operations beyond its physical premises in order to capture a larger, cheaper, and ever more precarious workforce.Footnote ⁶

Today, microwork platforms have intensified the magnitude and broadened the scope of this model through an algorithmic infrastructure that allows capital to capture a mass of precarious workers on a global scale and avail itself of a diverse spectrum of distinctions—in terms of race, gender, ethnicity, nationality, citizenship, language, immigration status, and criminal record—that intersect with class to form its ideally precarious workforce. In transgressing barriers imposed by geography, work visas, and time zones, microwork platforms are uniquely apt to amplify the factory’s ongoing production of precarity by simultaneously employing an ample, heterogeneous, and contingent workforce the world over. This deliberate and systematic production of precarity is also a byproduct of the factory model’s protocols. For Marx ([1867] Reference Marx and Fowkes1990), this was plainly clear in the factory’s creation of a relative surplus population by, on the one hand, “placing at the capitalists’ disposal new strata of the working class previously inaccessible to him” and, on the other, “setting free the workers it supplants” (531–2, 792–3, 798). Today, data workers are part of a fragmented, global class structure characterized by flexibility, insecurity, instability, hardship, isolation, and constant displacement (Apostolidis Reference Apostolidis2019, 3–24; Standing Reference Standing2011, 5–13).Footnote ⁷ As a central node in a global network of precarious labor, data processing reproduces these conditions on a prodigious scale, spawning a reserve army of digital workers who have been structurally barred from stable employment, including migrants, refugees, and formerly or currently incarcerated workers—disproportionately people of color and women—predominantly in the Global South, from the slums of Kolkata and Caracas to refugee camps in Kenya and Palestine (Altenried Reference Altenried2022, 94; Jones Reference Jones2021, 11–8; Munn Reference Munn2022, 32, 42, 93–8). These patterns are far from incidental. In the same way that the industrial factory system forged its own precarious workforce, the infrastructure of microwork platforms, based on its expedient model of connecting employers to a massive supply of on-demand labor, is systematically deployed to prey on existing structures of precarity among contingent workers, be it by targeting members of the Uyghur ethnic minority in China’s Xinjiang province to work for cotton producers, or actively recruiting refugees in India and Uganda to work for MTurk (Jones Reference Jones2021, 12; Munn Reference Munn2022, 74–8). These are illustrations of a coherent and methodical regime of precaritization that propagates the socioeconomic circumstances for global capital’s optimal workforce by, as Paul Apostolidis notes, designating “specific groups of people for uncommonly deplorable treatment” and making “work-related experiences isomorphic for populations throughout class, racial, and gender hierarchies” (Apostolidis Reference Apostolidis2019, 3).

Google’s data factory illustrates a similar pattern. While standard accounts of Silicon Valley portray the region as a metonym for postindustrial society (Piscione Reference Piscione2013), throughout most of the twentieth century this bucolic crest of Santa Clara County was distinguished less for its startup offices than for its orchards, groves, and meadows, which belied the toil of migrant workers of color across farm fields and inside canneries (Pellow and Park Reference Pellow and Park2002, 46–58; Pitti Reference Pitti2003, 1–77). Today, representations of the region as a mecca of innovation and wealth obscure the labor of Mexican and Latinx janitors and data workers who commute in and out of tech companies every day from the barrios of East San Jose (Zlolniski Reference Zlolniski2006). Google’s “data janitors,” then, are a material ramification and outgrowth of the racialized hierarchies of labor and spatial segregation that have historically sustained Silicon Valley’s economy through a vast pool of precarious, casual, and informal labor. As was the case with agriculture and microchip manufacturing, the integration of migrant workers of color into the local high-tech sector has been distinguished by precarious working and lived conditions. Not only does the material reality of unskilled data work refute the alleged tendency of the general intellect to abolish “the division of labour into specialized hierarchical tasks” and eradicate the “separation between the workers and their reified work” (Gorz [2003] Reference Gorz and Turner2018, 14), but it also reveals the extent to which digital capitalism depends on hierarchies of knowledge and skill that are intimately enmeshed in racially marked patterns of economic inequality and spatial segregation (McPherson Reference McPherson and Gold2012, 149–51; Mezzadra and Neilson Reference Mezzadra and Neilson2013, 131–66). This disparity, however, is less a factor of unequal access to technology than of unequal access to the linguistic, technical, intellectual, affective, and cultural competencies that this technology can yield—the very competencies that postindustrial thinkers construe as the meaningful and transformative character of immaterial labor. As much in the Global South as in Silicon Valley, then, data work conveys the crucial ways in which digital capitalism is shored up by structural patterns of precarity that attest to the dominant and reinforcing expressions of digital capital’s real subsumption of the labor process today: “the redundancy of labour and its expulsion from the value circuit” (Best Reference Best2024, 46).