Stepping back to move forward: Centering capital in discussions of technology and the future of work

Abstract

Some researchers have warned that advances in artificial intelligence will increasingly allow employers to substitute human workers with software and robotic systems, heralding an impending wave of technological unemployment. By attending to the particular contexts in which new technologies are developed and implemented, others have revealed that there is nothing inevitable about the future of work, and that there is instead the potential for a diversity of models for organizing the relationship between work and artificial intelligence. Although these social constructivist approaches allow researchers to identify sources of contingency in technological outcomes, they are less useful in explaining how aims and outcomes can converge across diverse settings. In this essay, I make the case that researchers of work and technology should endeavor to link the outcomes of artificial intelligence systems not only to their immediate environments but also to less visible—but nevertheless deeply influential—structural features of societies. I demonstrate the utility of this approach by elaborating on how finance capital structures technology choices in the workplace. I argue that investigating how the structure of ownership influences a firm’s technology choices can open our eyes to alternative models and politics of technological development, improving our understanding of how to make innovation work for everyone instead of allowing the benefits generated by technological change to be hoarded by a select few.

The conversation about work and technology has come a long way over the last decade. In their 2011 book Race Against the Machine and highly influential follow-up, 2014’s The Second Machine Age, Brynjolfsson and McAfee (2011, 2014) argued that advances in machine learning would increasingly allow employers to substitute human workers with artificial intelligence (AI)-enabled software and robotic systems. In a widely cited working paper, Frey and Osborne (2013) estimated that 47% of US employment was at “high risk” of automation over the next 10–20 years. These and other similar reports prompted alarmist headlines, public anxiety, and important debates about whether and how governments should intervene to mitigate the potentially transformative effects of technological change on workers and societies.

Those predicting widespread job loss have drawn important attention to the potential for large-scale transformations of work tasks, jobs, and labor markets. However, others have demonstrated the limitations of projections that assume that new technologies will generate convergent and foreseeable outcomes no matter where they are deployed.

Drawing on the long-standing theoretical tradition of social constructivism among scholars of workplace technologies (e.g., Barley, 1986; Zuboff, 1988), social scientists have countered technological determinism by demonstrating that the implementation and effects of AI, algorithmic systems, big data, and metrics vary along with the social contexts in which they appear. Bernhardt (2019) has identified a variety of organizational characteristics that are likely to influence the rate of adoption of AI systems even within the same industry, including variation in firm size, managerial strategy, and managers’ ability to overcome organizational inertia. (For example, warehouses run by large and well-funded firms like Amazon are likely to implement and upgrade robotics far sooner than small- and medium-sized firms.) Others have demonstrated how the introduction of AI systems can actually generate new work processes aimed at reviewing, modifying, or otherwise completing the output of algorithms (Gillespie, 2018; Gray & Suri, 2019; Irani, 2015; Shestakofsky, 2015, 2017). The ways in which workers use and resist big-data analytic tools are also likely to vary according to the institutional contexts and professional norms in which they are embedded (Christin, 2018). Attributes such as race, gender, nation, region, immigration status, and socio-economic status condition workers’ vulnerability to exploitation and discrimination on digital platforms like Uber, Upwork, LinkedIn, and Care.com (Graham et al., 2017; Ravenelle, 2019; Rosenblat, 2018; Schor & Attwood-Charles, 2017; Sharone, 2017; Ticona & Mateescu, 2018). And because the rate of technological development will diverge across sectors of the economy, so too will both the nature of resistance from workers and the public, and policy responses designed to moderate the consequences of innovation (Autor et al., 2019; Jacobs & Karen, 2019; Thelen, 2018; Vallas, 2019; Viscelli, 2018; Wood et al., 2018).

In short, studies that situate AI systems within concrete social settings have changed the conversation. By attending to the particular contexts in which new technologies are developed and implemented, researchers have revealed that there is nothing inevitable about the future of work, and that there is instead the potential for a diversity of models for organizing the relationship between work and AI to coexist simultaneously.

However, in their zeal to counter narratives positing that the future of work will be shaped by the attributes of technologies alone, scholars who situate AI within its local contexts have largely shifted our focus to accounting for variation—across workers, organizations, industries, and societies. Although social constructivist approaches allow researchers to identify sources of contingency in technological outcomes, they are less useful in explaining how aims and outcomes can converge across diverse settings (Bailey and Leonardi, 2015). In other words, social constructivist perspectives are not wrong, but they are incomplete. The backlash against technological determinism has led many analysts to overlook dimensions of social context that may be shared by a broad range of organizations.

Theorists are beginning to link the outcomes of AI systems not only to their immediate environments but also to less visible—but nevertheless deeply influential—structural features of societies. Scholars can expand on studies of the design, use, and reception of new technologies by situating their analyses within the broader social forces and institutional imperatives that enable, shape, and constrain these processes. How do the interests, goals, and perspectives of the powerful decision-makers who fund new technologies and set design parameters influence the relationship between work and technology on the shop floor? How do power, ideology, and institutions structure technology outcomes (Bailey & Barley, 2020)? Potential objects of analysis include meso-level social orders such as occupational standards (Bailey & Leonardi, 2015) and engineering and design cultures (Forsythe, 2001; Markoff, 2016), as well as macro-level systems of patriarchy (D’Ignazio & Klein, 2020) and racial domination (Benjamin, 2019).

One promising aspect of this approach is that it invites us to center capital in our analyses of work and technology. Instead of viewing capitalism as the background or context in which technological change plays out, analysts can interrogate the interests and motives of the actors who drive the outcomes that we observe (Noble, 1984). What does capital want? How does it go about accomplishing its goals? And how might these imperatives structure the landscape of technological change in the age of AI?

Studies of financialization and its consequences demonstrate the utility of this perspective. Over the past four decades, shareholders’ escalating influence over corporate governance has contributed to far-reaching transformations in work and organizations. To meet investors’ expectations, corporate managers have implemented tactics like mass layoffs and union-busting, outsourcing, cutting benefits, and demanding more effort and flexibility from employees (Davis, 2009; Lin & Tomaskovic-Devey, 2013; Osterman, 1999; Smith, 1997). Managers have developed and deployed new technologies to support these processes—for example, employers use information and communications technologies to coordinate knowledge work across geographic distance (Aneesh, 2006) and algorithmic systems to allocate retail workers to the hours when demand is predicted to be highest (O’Neil, 2016). Similarly, Amazon continues to develop new technologies to monitor, control, and increase the efficiency of its warehouse workforce (Delfanti & Frey, in press). Although workers in different settings may respond to these strategies and systems in unique ways, it is the interests of capital that are driving the implementation of technologies in corporations, with the ultimate goal of boosting stock prices to reward investors.

Yet as we center capital in our examinations of work and technology, we must also attend to the fact that not all capital operates according to the same logic. There are instead varieties of capital, each of which may advance the particular interests of owners and respond to incentives in unique ways (Lee, 2018). For example, startups backed by venture capital (VC) and publicly traded corporations may adopt different approaches to technological change. VC investors build portfolios of high-risk and potentially high-reward enterprises, expecting that most of the firms they fund will fail, but that a small number of big “hits” will deliver exponential returns via initial public stock offerings (Shapin, 2008). Managerial strategies in tech startups are consequently aimed at rapidly, repeatedly, and precipitously inflating the company’s perceived value. Instead of pursuing steady expansion derived from a profitable product and efficient business practices, VC-backed startups engage in continual experimentation to achieve rapid growth at all costs. Everything—including product features, organizational design, business models, and labor practices—is up for grabs.

We know that gig economy platforms like Uber can alter working conditions—often to the detriment of workers—as quickly as their engineers can update an app (Rosenblat, 2018). However, we lack a systematic understanding of how the relationship between work and technology inside of startups is patterned by a firm’s progression through cycles of VC funding. The imperatives of investors can shift as startups develop, as can a company’s fortunes (Shestakofsky, 2017). The ebb and flow of resources at their disposal has implications for how enterprises allocate funds to computational systems and human labor (Shestakofsky & Kelkar, 2020). New research can investigate how the structural constraints imposed by financiers are filtered through managerial initiatives and technology choices to generate pressures and opportunities for workers.

Attending to the varieties of capital driving change in the tech sector can also open our eyes to alternative models and politics of technological development. Privately owned tech companies face fewer external pressures to maximize profits and may thus be less likely to continually update their systems and implement practices that harm users’ well-being (Lingel, 2020). Platform cooperatives owned by and accountable to the workers who rely on them for gigs may be more likely to innovate in ways that stabilize and increase workers’ income (Scholz & Schneider, 2017). State investment banks can fund ventures aimed at generating public goods like green energy technologies (Mazzucato, 2015). Investigating how the structure of ownership influences a firm’s technology choices can improve our understanding of how to make innovation work for everyone instead of allowing the benefits generated by technological change to be hoarded by a select few.

At a time when the dominant narrative surrounding the future of work assumes that technology itself is the driver of change (Schwab, 2017), centering capital helps to remind us that questions of technological design are in reality often questions of political economy—of who has power over how agendas are set and how resources are allocated, of who will take on risks and who will reap rewards. Even the most talented and well-intentioned technologists are incapable of developing systems that can bypass these social conditions. When we observe the deployment and reception of algorithms and AI systems in the workplace without keeping capital firmly in view, we neglect an important influence on the relationship between work and technology while limiting our imagination of how it might be otherwise.