Narratives of labour as infrastructure and the automative imagination

Abstract

This article argues that the ways automation is imagined illustrate a wider problematisation of labour. The concept of an ‘automative imagination’ is proposed to articulate these different habits of considering and discussing automation. In these forms of imagination, I argue labour is discursively reconfigured as a logistical infrastructure. The concrete value of labour, the labouring body and the place of work as such are abstracted into an opaque logistical infrastructure in the narratives of an automative imagination. The impetus for this analysis comes from press releases and reports concerning automation and COVID-19 focused on the UK economy, creating a vanguard of abstracting ‘labour’ into infrastructure. The work of automation can usefully be understood as a relation, both in its implementation and in its imagining—a relation that geographers can, and should, interrogate. The automative imagination powerfully articulates the normative force of the performative abstraction and devaluation of work.

Introduction

The urgent need to maintain essential economic activity during the lockdowns of the global coronavirus disease 2019 (COVID-19) pandemic gave fresh impetus to longstanding discussions of automation and the associated perceived risks to employment (Lund et al., 2021; Wallace-Stephens and Morgante, 2020). Automating tasks away from workers was perceived by some as essential to public health and maintaining economies. Debates quickly turned to speculating about permanent effects on employment especially in sectors concerned with infrastructure—not least distribution and supply chains (Lin, 2022; Lund et al., 2021). However, these are not new claims. As highlighted elsewhere (for example: Bagrit, 1965; Cooley, 1987; Wajcman, 2017), automation has been variously figured as the solution to crises in capitalism for a very long time. It is a discourse with many previous iterations and bifurcations, recently garnering significant traction around developments in machine learning such as the popular Large Language Model ‘ChatGPT’ for automated ‘chat’ and Generative Adversarial Network ‘Stable Diffusion’ for the text-to-image generation of photo-realistic images. Prompted by UK national newspaper headlines that, in turn, draw on management consultancy reports to claim the pandemic affected the speed and scope of automation, I argue that the imaginings of automation warrant continued critical scrutiny, not least because these arguments are picked up in reports by the UK parliament Business, Energy and Industrial Strategy Committee and explanatory notes by the UK Parliamentary Office for Science & Technology. The imaginings of automation, as expressed across the public sphere (Barnett, 2019), influence the ways in which expectations and understandings of automation function throughout society.

‘Automation’ can function as a shibboleth for many different forms of inference concerning technological change. Indeed, as Jesse Ramirez (2017: 627) argues: ‘The term automation often functions as an abstract universal that condenses and obscures a host of social relations’. Automation can signify a bewildering, perhaps opaque, range of systems and things, a review of which is outside the scope of this work. For the purposes of this discussion, rather than assert a definition from a dictionary, specific scholarly tradition (such as the wealth of Marxian social science literature on ‘automation’), or other authoritative source, I derive a broad meaning from the discourses in question. Thus, adopting an ‘ordinary’ language, ‘automation’ means enhancement or replacement of human labour, both ‘manual’ and ‘knowledge’ work, with (apparently independent) technology. Automation is figured by some as a boon for economic productivity and by others as a threat to livelihoods and even, in hyperbolic terms regarding ‘AI’, the very existence of humanity. I am not concerned with arguing for a ‘correct’ meaning of ‘automation’. Instead, I interrogate what kinds of work the idea of ‘automation’ (however vague) does in how we collectively negotiate perceived technological change and the place of labour. Importantly, ‘automation’ is employed in its ‘ordinary’ language meaning precisely because I am interested in the ways that it is used as a catchall term in a variety of contexts, from newspapers to policy documents, which collectively convene that ordinary meaning. Thus, ‘automation’ can be helpfully understood as a ‘problematisation’ (Barnett, 2015; pace Foucault, 1984a, 1984b)—what ‘automation’ means is a relational quality of the situation to which it is offered as a ‘solution’. In this case, a broadly conceived ‘automation’ is offered as the technological solution to reduced productivity and the necessities of health controls relating to the global COVID-19 pandemic. Automation is best explored, in the scope of this discussion, as a form of relation through meaning, power and space (see ‘An automation relation’).

The concept of an ‘automative imagination’ is proposed to articulate the different habits of considering and discussing automation. In these discussions, I argue that labour is reimagined as an infrastructure. Jobs are atomised into tasks to be rationalised and automated. The inference is that the automated process is the most important element in the production of value, workers are merely a means of bridging automated tasks. I argue this mode of imagining is becoming normative—in two senses. First, it is normative because it implies who or what is ‘normal’, and therefore worthy of consideration. Second, and following from the first sense, it is normative because what is imagined becomes the criteria by which we judge one another about whether anything we say about ‘automation’, ‘AI’ or ‘robots’, for example, is appropriate, or not, to the context of discussion. We are invited to normatively consider automation personified in the shape of ‘robots’ and not consider the personhood of the abstract worker—who, in actuality, will be variously classed and racialized. The politics of automative imaginations is significantly bound up with normative judgements about who and what matters through how they are imagined.

The urgency of rethinking the status quo in the face of unprecedented challenges such as a global pandemic opens out a discursive space for proponents of (what they see as) ‘solutions’ to gain traction with decision-makers in governments and industry. Evangelists and critics alike (in academia and beyond) fall into the same rhetorical (determinist, pace Wyatt, 2000) trap of presenting the technology as monolithic and unstoppably powerful. Technology is positioned as the ‘other’ of the ‘human’, regardless of the underlying ethical judgment. From the Luddite ‘machine breakers’ (Hobsbawm, 1952) to the (always yet-to-come) eschatological ‘Singularity’ (Kurzweil, 2005, 2023), speculation about and fantasies of the replacement of labour with machines have been the source of both anxiety and celebration. In broad terms, the proposition of automation as both ‘salvation’ and ‘threat’ is performed across a wide range of speech and writing actions. Conceptualising an ‘automative imagination’ advances the critique begun by geographers (notably Bissell, 2020; Lin et al., 2022) that troubles the simplistic binary of celebration or fear by identifying ‘a wide spectrum of leanings, inclinations, and tendencies’ (Lin et al., 2022: 66) in relation to automation. An automative imagination frames the complex ‘dispositions’ (Lin et al., 2022) in relation to automation as contextually situated ‘cultural – and not just technical – form(ul)ations… deeply intertwined with specific societal expectations, ambitions, and strategies unfolding and generated across space and time’. ‘Dispositions’ towards automation are fluid and always already relational and situated. The automative imagination describes our cultural interrelations of expectation and speculation revealed in the problematisation of automation and labour, specifically addressed in here in the wake of COVID-19.

The ‘automative imagination’ is a heuristic for studying the ways labour is imaginatively abstracted from ‘jobs’ into a more diffuse ‘infrastructure’. The struggle over work and what constitutes a ‘job’ is part of longer term economic-political relationships between labour power and value and political–economic relationships between the right (or power) to work and control (power) over work (see, for example Harvey, 2014; Hudson, 2000; Massey, 1984; Storper, 1997), the breadth of which is outside the scope of this article. These articulations of relationships between work–power–value must be seen in terms of their spatial character and especially in terms of geographical debates concerning work and infrastructures (see Furlong, 2020, 2021; Strauss, 2020). Simply put, for the purposes of this argument, infrastructures are the stuff of movement of people, things and value. Further refined, ‘infrastructure’ is a lens through which we can, variously, problematise and ‘defamiliarise’ the ‘taken for granted’ (Appel et al., 2018: 4) fabric of contemporary society. Pushing this further still, with Bowker (2018), I argue that ‘knowledge’ is as integral to infrastructures, and our problematisation of them, as material objects, such as pipes and rails. Imagination as a form of relation, and its expression as dispositions (pace Lin et al., 2022), can be instrumental to infrastructures as the boundary object that brings together disparate propositions of automation as the ‘future of work’. In contrast to Stokes and de Coss-Corzo (2023: 439), who worry about romanticising ‘heroic’ infrastructural labour, I am concerned with problematising the elision of labour as such and its rendering as a subordinate infrastructure itself. Indeed it is the ‘promise’ of infrastructure, following Appel et al. (2018), that is most potent for our understanding, for infrastructures have long been ‘technologies that modern states use not only to demonstrate development, progress, and modernity but also to differentiate populations’ (Appel et al., 2018: 5) and the elision of the worker as such is the replacement of labour, as understood till now, with (what has been understood until now as) ‘the means of production’, i.e. ‘machines’, as the sine qua non of the flow of capital and value.

In the years immediately preceding the COVID-19 pandemic, a wave of predictions about an impending intensification of automation and the resulting effects on society swept through the catalogues of large publishing houses and the pages of the popular press. Reflecting on the momentum of the discourse, Benanav (2019b) identifies ‘self-described futurists’ and senior civil servants that all variously propagate the trend: Brynjolfsson and McAfee (2014) in their ‘Second Machine Age’, Ford’s (2015) ‘Rise of the Robots’, as well as Robert Reich and Lawrence Summers, former Labour Secretary and Treasury Secretary, respectively, to President Clinton. For Benanav (2019b: 11–12), drawing on Leontief (1983), these are the latest iteration of a ‘spontaneous discourse of capitalist societies, which, for a mixture of structural and contingent reasons, reappears in those societies time and again as a way of thinking through their limits’. Whereas Benanav (2019a, 2019b, 2020) and others (e.g. Autor, 2015) seek to interrogate the veracity of the predictive claims about automation, my concern here is to examine the work done by such imaginings: I argue that political work is done in imagining automation. The automative imagination of labour as an infrastructure abstracts out the social and concrete, constructing norms of problematising labour and where it takes place. The articulations of automation and the forms of labour relation espoused within an automative imagination serve to intensify what Elson (2015: 125; citing Marx, 1976a: 175) describes as ‘the domination of abstract labour’, which ‘signifies “a social formation in which the process of production has mastery over man, instead of the opposite”’.

Imaginings of automation are economically and politically potent. Statements made by management consultancies and governments alike connote that the COVID-19 pandemic functions, somewhat, like a classical economic ‘crisis’—a large external shock against which are set macroeconomic strategies to rectify our economic system. To be clear, my analysis is agnostic in terms of theories of ‘crisis’. Crisis, in the social sciences, is often figured as a political-economic idea. The legacy of Marx’s (lack of) theorisation of crisis is an orthodoxy that crises are necessary and central to the contradictory function of capitalism (pace Mandel, 1975), while Keynesian scholars understand crisis as a symptom of inadequate institutional structures and/or policies (Clarke, 1993). I am not concerned with arguing for a particular flavour of ‘crisis theory’; rather, there is a risk of falling into what O’Connor (2009) argues is the fatalism in such endeavours. In some respects, the specificity of the pandemic per se is less meaningful than the opportunity that a significant event—a ‘crisis’—presented for those seeking to make claims about the apparent necessity of change via automation. Positioned as ‘solutions’, regardless of the supposed ‘problem’, is a powerful rhetorical play. Indeed, crises can be simultaneously figured as both case and effect:

When a triggering factor is also an outcome, we find ourselves within a spiral. This can be very fruitful and worthwhile, or it can enclose us – in the absence of new criteria – in a vicious circle that we can then describe as a “downward spiral”. (Stiegler, 2015: 28–29)

What is analysed as an automative imagination here is an absence of criteria for evaluating what can and should constitute work in the contemporary era and, arguably, invites the politically fraught ‘spiral’ feared by Stiegler. I argue that automative imaginations speculatively reimagine labour as a form of logistical infrastructure, a means to bridge the gaps between already automated tasks in complex multi-scalar processes of production and service.

This article unpacks the reimagining of labour in relation to automation in four sections. In the next, second, section, I discuss the methodological approach taken to interrogate the problematisation of automation. The third section , An automation relation, unpicks the discursive steps through which particular stories concerning the apparent threats to jobs caused by automation play out, focusing on materials produced by management consultancies and ‘think tanks’. The focus of this analysis is the ways in which jobs are rhetorically atomised into tasks and become reimagined as the logistical glue that fixes together already automated tasks. In the fourth section, Managing expectations, the storytelling work of management consultancies is examined as a key component in the production and performance of the automative imagination, as the production, validation and dissemination of the promises and threats of automation for business and for wider society. Finally, I argue that the automative imagination is a heuristic for revealing the normative power of the abstraction and devaluation of labour-as-infrastructure.

Methods and problematisation

To contribute to discursive, political and spatial interrogations of claims about automation, specifically in relation to the COVID-19 pandemic, I draw on analyses of a selection of press release and reports from management consultancies, (inter)government agencies and third sector (‘think tank’) organisations. I argue that ‘automation’, specifically in relation to the COVID-19 pandemic, is a form of ‘problematisation’ (Barnett, 2015; Foucault 1984a; Rabinow, 2003). Following Barnett (2015), ‘problematisation’ is often, mistakenly, figured (invoking Foucault 1984a; cf. Rabinow, 2003) as having an axiomatic double meaning—a critical ‘method’ on the one hand and an object of study on the other. Both (mis)understandings—of a Foucauldian approach (pace Foucault, 1984a, 1984b)—according to Barnett (2015), make ‘an assumption that social life is ordered through processes of naturalization or universalization’ which is ‘a projection of favoured theoretical paradigms … that place a premium on debunking the appearance of ontological fixity upon which the reproduction of social relations is assumed to depend’. Retaining the ‘problematic’ of problematisation, by contrast and as Barnett (2015) goes on to argue, necessitates thinking:

contingency as a relational quality rather than a fact to be demonstrated, so that the task of analysis becomes one of clarifying the conditions and situations to which problems are a response.

Furthermore, in interrogating such forms of problematisation, clarifying the conditions to which they respond, ‘the pressing analytical task is no longer viewed as one of critical disruption, but rather one of rearranging what is already known, of seeking to “make visible what is visible”’ (Barnett, 2015; citing Orford, 2012: 618). Automation as a problematisation invites us to think of the idea and ideal of ‘automation’ as a relational quality of the situation to which it is offered as a ‘solution’. Rather than test the, broadly Marxian, theory of (yet another) crisis of capital constituting an opening for the solutionism (Morozov, 2013) of automation, we can interrogate automation in terms of the intensification of an always already difficult situation (Barnett, 2015) as a regular way of organising uncertain situations (Boltanski and Thévenot, 2006), such as a global pandemic.

Reflecting on the claims made throughout and immediately following the worst of the pandemic, I was struck by the speed with which claims about the ‘future of work’ and the importance of ‘automation’ were made and repeated in the popular press. The argument takes its impetus from newspaper articles (in print and online); however, it is their reliance on a selection of contemporaneous management consultancy and third sector reports concerning automation and the COVID-19 pandemic focused on the UK economy that serves as the principal focus of scrutiny. This is supplemented by academic and governmental analyses that the reports in question often rely on to make claims about automation and the risk to particular job roles. The choice of newspaper articles and the reports on which they draw reflect the situation of the problematisation and how this has been revealed both in the intensification of a particular situation, the pandemic, and to me as a researcher. While it is certainly the case that the newspaper articles and reports have been widely shared and discussed across social media, those media are not analysed here for pragmatic and methodological reasons. First, access to social media data has been significantly limited and placed behind paywalls (Walker et al., 2019). Second, I seek to take forward existing work that has focussed on affective registers of ‘dispositions’ towards automation (Lin et al., 2022), thickening them by further interrogating the underlying problematisations from which they emerge.

Discourses of automation and labour are the empirical focus of this research and content analysis is the methodological basis for the interrogation of claims made about automation amidst and in the wake of the COVID-19 pandemic. For this article, I focus on widely cited management consultancy and ‘think tank’ reports that position automation as, simultaneously, an emerging crisis and a significant political-economic opportunity. The scope of the materials I draw on for this article is English-language reports produced by multinational management consultancies, the more prominent of which, in terms of citation in newspaper articles, are: Accenture (Daugherty et al., 2020), Ernst & Young (Kroukos, 2020), Deloitte and McKinsey Global Institute (Lund et al., 2021); governmental and non-governmental agencies: the Asian Development Bank Institute (ADBI) (Petropoulos, 2021), the International Monetary Fund (Acemoğlu, 2021), the Organisation for Economic Cooperation and Development (OECD) (Arntz et al., 2016), the Royal Society of Arts (RSA) (Wallace-Stephens and Morgante, 2020), and the World Bank (Beylis et al., 2020). These do not emerge from a systematic review, but rather emerge from a review (via Lexis Nexis) of United Kingdom national newspaper stories concerning ‘automation’ or ‘robots’ in relation to ‘coronavirus’ or ‘COVID’. Of the 3173 initial search results, 558 explicitly addressed a ‘post-pandemic’ future and of those approximately 152 make some reference to statistics or ideas that may come from these reports, via other sources, and 91 explicitly cited one or more of the above reports. Furthermore, the influence of both the newspaper articles and the consultancies’ reports is evident in reports, minutes and responses published by the UK House of Commons Business, Energy and Industrial Strategy (BEIS) Committee (2019, 2020; 2023).

The central context is the COVID-19 pandemic, thus the temporal range of the materials analysed is between January 2020 and December 2022. However, this is complicated by the national context of the UK withdrawal from the European Union, which becomes entwined in discussions of the future of work as the pandemic subsides in 2022. This is especially the case in relation to reported skills shortages in health and social care, discussed in particular in House of Commons committee reports (House of Commons—BEIS Committee 2019, 2020, 2023) and in the report compiled by PwC commissioned by BEIS relating to potential job losses due to AI (2021). A broadly ‘abductive’ (after Latour, 2004) method of analysis of these, and other materials, was foundational in the development of a theory of ‘automative imagination’, which is ongoing. As such, these analyses form one part of a wider project to elucidate and interrogate the ‘automative imagination’, through analysing longer-term, recurring, narratives about automation across a range of literature and media, which are beyond the scope of this article. Automation, specifically the automative imagination, as addressed in this article, is a form of relation, and crucially it carries with it particular scalar relations. An automation relation is a means of configuring various people, places and things—and it is the nature(s) of that relation to which we turn in the next section.

An automation relation

Automation, as characterised here, falls within the interstices of a range of theorisations of labour/work as a relation. Perhaps the most prominent forms of the labour/work relation within automation are: between effort and value, between power over action and power to act, and, perhaps, between human and machine—or worker and the means of production. It is outside of the scope of this article to attempt a comprehensive synthesis of these forms of relation—I have more modest aims. In this section, I surface the articulations of value and power that I argue are present within imaginings of automation. Without wishing to over-generalise or elide contributions, geographers have made a substantial contribution to the analysis and theorisation of labour and work through variously studying labour and work (Castree, 2007; Herod, 2001; Massey, 1984; McDowell, 2009). For the purposes of this analysis, I focus in particular on: (i) economic–political relationships between labour power and value; and (ii) political–economic relationships between the right (or power) to work and control (power) over work. Both articulations of relationships between work–power and work–value must be seen in terms of their spatial distribution and, importantly, spatial division. Crucially, building from the work of Cox (1998), Massey (1984, 1991) and Strauss (2020), I argue that these relations are scalar relations. The ‘economic–political’ and ‘political–economic’ articulations of the labour/work relation are a useful heuristic for understanding the ways automation is imagined and understood in contemporary context of a post-pandemic world. To be clear, in this section, I am principally addressing the ways these relations are rationalised and storied, rather than attempting to synthesise neo-classical economic and politico-geographical debates. I begin by discussing the ‘economic–political’ relation of labour power and value—most famously articulated by Marx—and the ways this appears in an automative imagination. Second, I consider the power relation/s of ‘work’ potently analysed by geographers and how this plays out through an automative imagination. Finally, I argue that an automation relation can be usefully understood, at least in part, as an instance of the (ongoing) spatial division of labour (Massey, 1984). The aim of this section is to explore the ways that automation as a relation abstracts out human labour from stories about production and work.

First, automation as a relation disrupts and reconfigures how the labour power–value relation is imagined, rationalised and thus narrated. If ‘[v]alue is an immaterial but objective social relation which arises out of the monetization of acts of exchange’ (Harvey, 2020: 108) then significant changes in either the acts of exchange or the ‘objective’ social reality of labour challenges the settled status of value. Extrapolating from pre-pandemic data and reporting of task-work, such as ride-sharing, food delivery and information classification tasks, forms the basis of claims in a number of the analysed reports—in particular, Accenture, McKinsey Global Institute and the evidence collected in the UK House of Commons—about significant shifts in acts of exchange. With the pandemic destabilising social reality, these reports argue that this disruption drives a turn towards infrastructures of task-work, displacing traditional, contractual, waged work. The reimagining of the role of labour/work in relation to the economic production process challenges a more-or-less stable social reality in which one or more classes of worker sell their labour power. This, in turn, determines exchange value, and, furthermore, exchange value may be seen as the determinant of labour-time. When the socially necessary labour power for a given commodity fluctuates, or is imagined to be reduced to the point of a facilitating infrastructure for a wider production process, then it necessarily cannot dictate price. The forms of automative imagination discussed here significantly rethink and abstract out the importance of labour. For example, reports by Accenture (Daugherty et al., 2020) and McKinsey Global Institute (Lund et al., 2021) place emphasis on the automation of customer-facing work and a broad shift to remote working. ‘Work’ in these reports is abstract and less concerned with the role of the worker than it is, as Richardson (2024) observes in other contexts, with the places and mechanisms of the tasks. The economic elements of the automation relation invoked in contemporary discourses of automation, labelled ‘futurological fodder’ by Samers (2021), skew what Elson (2015) sees as the two ‘one-sided abstractions’ of the labour–value relation of production: abstract–concrete and social–private. A ‘value theory of labour’ (Elson, 2015) describes labour as always having abstract and concrete, social and private aspects. When applied to the automation relation, the heuristic of the ‘value theory of labour’ asks important questions about how labour is figured in an automative imagination.

In the imagining of an automation relation, the articulation of labour-as-infrastructure intensifies what Marx (1976b: 1012) identifies as a simplified ‘value–creating substance, as social labour in general which is in the act of objectifying itself, and whose sole feature of interest is its quantity’. The alienation of the worker from their own labour has been a central feature of theories of the capitalist process of production, from Smith, Ricardo and Marx to the present day. However, when labour-power is abstracted to the point of no longer signifying a predominantly human activity then the surplus value of the commodity puts the alienation relation between social labour and value into question. Whereas many theorisations of value follow, explicitly or implicitly, an assumption that ‘labour is the source of all value’ (Armstrong et al., 2016: 21; see also Steedman, 1981), the discursive force of the imagining of automation as contingent upon labour-as-infrastructure poses again longstanding questions around the exploitation of workers (Massey, 1984; McDowell, 2009; Peck, 2017). With an automative imagination, the automation relation confronts us with the possibility that labour itself, as currently understood, is abstracted away from a human-social basis.

Second, imaginings of automation articulate some of the latest phases of power relations that course through the contemporary condition of work. There is, of course, a wealth of politics and political activity in the debates and practices of work; I want to focus on one particular ‘power repertoire’ (Piven and Cloward, 2000) here: the embodied dispositions of individual workers. A relation of labour control, power over work, has the necessary counterbalance of the relation of labour rights, the power to work. Control (following Deleuze, 1992; Foucault 2010), whether in the disciplinary relationships of workers regulated within factories or in the biopolitical self-surveillance of the ‘anywhere worker’ (or perhaps in this case: the worker-from-home) concerns the regulation of bodies. As Bissell (2020) carefully elucidates, it is precisely through the performances of storytelling that understandings of automation both gain a form of power and ‘catalyse thinking and feeling’ (Bissell, 2020: 369), which are inseparable from the bodily dimension of experience. The stories of work told by workers bind the contemporary experience of the so-called ‘Fourth Industrial Revolution’ to working bodies (Schwab, 2014) through their embodied knowledges. Yet the automative imagination appears to abstract away the body. The automation relation of the automative imagination is, arguably, a move to untether ‘work’ from the body of the ‘worker’. Where ‘effort’ and ‘power’ are, for many, inextricably bound to bodily capacity, the automative imagination dissociates work into an abstract potential.

A spatial dissociation of workers from recognisable places of work, widely predicted across the newspaper articles and consultancy reports (especially by Lund et al., 2021), positions labour as a flexible resource that supports the delivery of services. For McKinsey, this not only reshapes the urban fabric, with higher office vacancy rates (Lund et al., 2020). The significance of the automative imagination to what might be otherwise considered broader trends of the abstraction and devaluation of labour by capital is the way it articulates scale. The forms of abstraction represented by the automative imagination are at both hyperbolic and ill-defined. For example, an Accenture report (Daugherty et al., 2020) identifies a ‘tech-clash, ‘a collision between old models that are incongruous with people’s current expectations’ (Daugherty et al., 2020: 9), in which businesses create ‘human-centred’ experiences that are ‘accountable’ and built on the capacities of AI and ‘generative design’. The role of workers is only inferred, briefly, as a vaguely defined ‘curator’, connoting a management class, or as ‘human–AI collaboration’, in which the ‘human’ is more-or-less an assistant or error checker. The scale of the embodied worker is broadly denied. The inference of an infrastructure of tasks is at the scale of institutions—companies, governments and the territories of nation states, or supranational trading blocks—not of workers.

The discursive power of the automative imagination evident in the literature explored here lies in a forceful and imposing abstraction. Rather than, as Richardson (2018) espouses, attending to contextual specificities of particular sectors and workplaces, the automative imagination, of the hyperbolic boosters and pessimists alike, makes general pronouncements, eliding and ignoring the entanglements already evident in actually existing automation. Whereas geographers, not least Bissell (2020) and Sumartojo and Lugli (2021), attend to the embodied sensibilities and liveliness of automation as a worldly relation, often in specific workplaces, labour is increasingly reimagined not in terms of personified workers but instead as a unit of human–AI collaboration, with employees working from home (for example in Daugherty et al., 2020). Both reports by McKinsey and the RSA (Lund et al., 2021; Wallace-Stephens and Morgante, 2020), for example, narrate the worker as one of many elements of a diffuse infrastructure of production in which labour is, arguably, no longer the most prominent part. Furthermore, the differences of and within working bodies are elided in several of the reports. Differences of ability, gender, race and other forms of understanding workers in bodily ways are abstracted away (for further discussion, see Bissell, 2020; Bledsoe and Wright, 2018; Elwood and Leszczynski, 2018; Richardson, 2018). While it is outside the scope of this work to attend to the specific ways these abstractions may function, as powerful work within geographical scholarship attests (Bledsoe and Wright, 2018; Bledsoe and Wright 2019; McDowell, 2009; McKittrick and Wood, 2007), differences are constructed and experienced with different intensities and they have a political importance to which I am unable to do justice here. However, I note the striking contrast between the abstraction of working body and the deep individualisation of the consuming body. For example, in an Accenture report, Daugherty et al. (2020) heavily focus on the ways automation ‘personalises’ the consumption of goods and services while simultaneously abstracting out the labour that may involve.

Third, an automation relation can be usefully understood as an instance of the (ongoing) spatial division of labour (Massey, 1984). I want to, albeit briefly, address the broad patterns of spatial delineation that undergird the narratives of the automative imagination as I have explored it here. However, I must note that it is beyond the scope of this work to map out the specific institutional forms of the spatial divisions of workers in what have become, to a lesser or greater extent, highly regulated translocal and inter-dependent operations of production. Following Peck (2017), we can see how the narrative logic of the automative imagination continues a significant history of complex movements of labour and workers across national borders and between regulatory regimes. ‘Offshoring’ and ‘outsourcing’ are the early 21st century rationales for a continuing globalisation of the spatial division of labour. The narratives of ‘hollowing out’ of ‘high-waged’ economies in North America and Western Europe, whereby low-skilled jobs were relocated to ‘low-waged’ economies, exemplify the ‘lean’ rationales of corporate restructuring (pace Peck, 2017). The ‘outsourcing’ narratives of the late 20th and early 21st centuries further demonstrate a longer history of the abstraction and elision of particular forms of work that we see undergirds the automative imagination. ‘Skilled’ work resides, as always (pace Harvey, 2014; Massey, 1984), within the management classes, albeit with a significant exception in healthcare (especially for Lund et al., 2021; Wallace-Stephens and Morgante, 2020). Where once the ‘risks’ were outsourced to alternative labour ‘elsewhere’, the narratives of a contemporary automative imagination abstract away the risks associated with labour as such. Indeed the ‘risks’ of a workforce are blurred between the financial risks and the urgent (at the time) health risks of the pandemic personified by customer-facing service workers, in retail and hospitality especially (in particular, Lund et al., 2021).

Where once apparently unskilled labour was spatially disaggregated and moved elsewhere, in the worlds of the automative imagination, it moves to no-where, ‘no-shoring’ (following Peck, 2017), it is placeless, subsumed into the logistical operations of automated production and services. What is elided in these rhetorical sleights of hand is significant—hidden are the maintenance works that keep the machines running, so too are the host of support roles and new forms of work created by the novel forms of automation. The spatial divisions of labour in the automation relation are an abstract problematisation of, and by the same token, a form of spatial deception. An abstraction of work, from roles performed in specific places to tasks performed ‘anywhere’ is illustrated in all of the reports analysed, with references to ‘agile’, ‘on-demand’ work that takes place ‘in the cloud’ and ‘from home’. The elision of the forecasting of work by abstracting it is a spatial deception because it operates under the pretence that labour can somehow float free of place. This is not necessarily an intentional subterfuge. Rather, by adopting the frame of reference and the lack of situational grounding in imaginings of automation, we are collectively deceiving ourselves.

Ultimately, the automation relation is principally a relation of abstraction. It is the abstraction out of the social aspects of the labour–value relation, the abstraction out of the embodied dispositions of individual workers and the abstraction out of the place of work. The automation relation figured through the automative imagination of labour-as-infrastructure is abstraction of humanity from what is commonly held to be socially necessary work. In this section, I have demonstrated how the abstraction of the automative imagination re-intensifies the political economic ‘regimes of truth’ (Foucault, 1977) of the 21st century. To explore how such narratives function in practice, the next section draws out common discursive tropes and interrogates examples to illustrate them.

Managing expectations

“Seventy-two of us are out of jobs,” said Bud. He slumped even lower in the couch. “[Our] job classification has been eliminated. Poof.”—Player Piano, Vonnegut (1952: 75)

The business of managing the expectations—the desires and fears—of and for automation is lucrative. Knowledge about a (supposed) future and expertise around how to get there is marketed and sold by multinational management consultancies. We can follow the arguments put forward, the claimed statistical bases for them and the deft word play used at various stages of the proposition and promulgation of near future of automation. Whereas some of the statistical analyses of labour trends in reports by ADBI and World Bank, for example, may appear dry, both statistics and references to prior reports are used to valorise speculative assertions by all of the reports and press releases examined about the intensification of automation. I want to unpick some specific elements of that process through examples taken from responses to the crisis of COVID-19 concerning the (further) adoption of and advancement of automation. I begin by examining stories communicated in the popular press and the ways in which their storying make assumptions and translate possibilities into near certainties.

First, urgency is identified, and a timeframe is constructed. Predictions about automation and futures of work fall within a more-or-less proximate future. Generally, they lie within the lifetimes of many of those already of working age and if not then certainly within the scope of their children. This is perhaps reasonable—we want to know about our own future, not an abstract one. This also lends a certain urgency to what is anticipated, and with that can come some pushback. As Michael (2000: 25) argues:

A near future can warrant swift action, but it can also attract the accusation that it is no more than opportunism on the part of the actor who gains from some sort of “scare” or other.

Even so, ‘scares’ about job losses to automation periodically gain traction. In so doing, there is a revitalisation of the determinism concerning automation. The common sense that ‘the robots are coming’, that many of us ‘will all be out of jobs’, took hold in relation to the uncertainties brought by the pandemic. Here we see the temporal horizon brought firmly within the working lifetime of many in work today. What is more: it is possible to trace how this takes place. As coronavirus took hold in Northern Europe, newspapers reported an apparent ‘speeding up’ of automation as lockdowns kept workers at home. These stories pitch an enforced absence of workers as an impetus for managers to increase or quicken automation processes. For example, the Press Association (2020) published a story, carried by The Guardian, on 30 March 2020, based upon analysis by EY that suggested:

41% of respondents in a survey by the auditing company EY said they were investing in accelerating automation as business prepared for a post-crisis world.

It is interesting to note that the source of that number, an EY report, is less certain. In response to a question concerning decisions on a number of factors, a number of ‘executives’ were given three options: ‘We are taking steps to change’, ‘We need to re-evaluate’ and ‘No change’ (Kroukos, 2020). Regarding decisions on ‘speed of automation’ in response to COVID-19, 41% of survey respondents stated that they would ‘need to re-evaluate’ (Kroukos, 2020). Here we see a key feature in the propagation of narratives of the ‘risks’ of automation, a shift from possibility or probability to certainty.

Second, forms of valorisation are adopted through the use of official and/or reputable empirical sources. Governmental labour market statistics are used as a ground truth in many claims made about future job losses due to automation. Many claims made by management consultancy and third sector organisation reports analyse government statistics in order to make their claims. Chief amongst the statistics used are UK Office for National Statistics (ONS), UK HM Revenue & Customs and US Bureau of Labour (USBL) statistics on employment. These are supplemented with data from the OECD and other sources. In a sense, redundancies in the USA due to lockdowns and the furlough scheme in the UK are treated as a natural experiment on what happens when roles or tasks need to be rapidly adapted or ‘replaced’ with automation. For example, a report published by the RSA (Wallace-Stephens and Morgante, 2020) infers a direct relation between take up of the UK government Coronavirus Job Retention Scheme, furloughing, and risks of automation, based upon ONS data. From a different tack, McKinsey (Lund et al., 2021) draw together their prior analysis (Manyika et al., 2017) of World Bank and USBL O*NET data, the same data as Frey and Osborne (2013), and data from a survey of ‘executives’ concerning their thoughts and plans in a post-COVID economy. Governmental statistics function as a form of rhetorical regulation that have normative power, they afford authority to those citing them and circumscribe what it is acceptable to say.

Third, the risks of unemployment, as statistical probabilities, are (sometimes) translated into rhetorical certainties. Analyses of risks of unemployment due to automation are, in the first instance, figured as probabilities of risk. Probabilities are, in some cases, translated into a degree of certainty in the rhetoric of press releases and executive summaries. The reports use ONS and USBL data on employment and applying a national-level relative ‘risk register’ for job losses, by employment type, due to automation they have produced. There is a transformation: from ‘may’, a probability, to ‘will’, a certainty. Inferences of similarity between localised or short-term anecdotal evidence and the statistical risks are applied. When framing the discussion of automation in response to COVID-19, RSA report writers (Wallace-Stephens and Morgante, 2020) offer the example of cashierless retail such as the Amazon Go stores and Sainsbury’s ‘SmartShop’ system and cite former Sainsbury’s CEO Mike Coupe suggesting what ‘might have taken three or four years to get to… [has] happened in the space of less than six weeks’ (WARC, 2020). Actually existing forms of automation, especially those used to ‘replace’ jobs, are cited as evidence of the proposed risks that may be brought by future automation. McKinsey (Lund et al., 2021) raise anecdotes of ‘[h]ospitals and hotels around the world’ using ‘robots to sanitize rooms, while the use of AI-powered chatbots has surged in […] retail and healthcare’ (61), citing a Tech Republic Journalist (Dallon Adams, 2020), as supporting evidence of their finding that ‘executives say that COVID-19 is accelerating adoption of automation’ (60). There is a rhetorical sleight of hand, where inference is implied in the use of a non sequitur. Causality is implied, but not asserted, when an anecdote about a contemporary example of automation is given and an apparently related claim is made about a near-future automation.

Finally, through these three steps, we can observe how authority is established, and forms of evidence are translated, from academics to management consultants and NGOs. In many, but not all, of the work making claims about futures and risks of automation, such as reports by EY, PwC and the RSA, there are frequent references to other sources upon which their own analyses are based. Claims are made either by asserting the authority of the cited work itself, by virtue of notoriety for example. Most claims made in reports concerning the risks of automation related job losses draw principally on two papers: Arntz, et al. (2016) and Frey and Osborne (2013). Frey and Osborne published a piece (2013) that modelled the susceptibility of different kinds of jobs to automation that has become extremely influential. Arntz et al. (2016) performed analysis for the OECD, testing Frey and Osborne’s findings and offering their own concerning the risks of automation. What is most striking is the influence of this work and in particular the methodology used by Frey and Osborne (2013). A total of 702 occupations as defined by a classification system taken from USBL statistics are analysed by Frey and Osborne (2013) according to nine variables across three categories of ‘perception and manipulation’, ‘creative intelligence’ and ‘social intelligence’ for whether they represent ‘bottlenecks’ for computerisation. This forms a basis for many subsequent analyses, including Arntz et al. (2016).

Automation both connotes and denotes debates, ideas and practices that perform various kinds of work for different interested parties. Automation constitutes ways of positioning ‘work’, and the identification of who or what counts as ‘workers’ and work-places freights political beliefs and values. I argue an automation relation is a significant means of attempting to configure various people, places and things—and it is that relation that forms the basis of an ‘automative imagination’. In the next section, I conclude by opening out a space for critique in which the articulation of the concept of an ‘automative imagination’ may contribute to understandings of the geographies of automation more broadly.

The automative imagination

The imaginings of automation reveal more about the desires in some quarters to undercut and undermine the political capacities of labour in the present than shining visions of an automated future. As both the apparent benchmark and harbinger of change, narratives about automation produce a short circuit in the supposed progress from ‘cause’ to ‘effect’. Speculative ‘effects’ rhetorically float free in the fever dreams of evangelists and pessimists alike, while ‘causes’ remain opaque or yet-to-be determined. ‘Crises’, understood in ordinary language—such as those in the global COVID-19 pandemic or in global financial systems, present opportunities to advance and propagate an automative imagination, as evidenced in the rhetorical tropes of the literature discussed in the second and third sections. This is part of longer term (i) economic–political relationships between labour power and value; and (ii) political–economic relationships between the right (or power) to work and control (power) over work. Both articulations of relationships between work–power and work–value must be seen in terms of their spatial distribution and, importantly, spatial division. The automative imagination of labour abstracted into an infrastructure demonstrates another phase of what Harvey (1985) identifies as ‘jumping scales’—strategies for mitigating the contradictions between mobility and immobility in capital. Harvey’s is a spatial politics constituted in the tension between spatially fixed capital—in factories, worker skills, and social and physical infrastructures—and mobile capital—as finance and money. As Cox (1998: 4) observes ‘capital in its mobile form, through its constant search for higher profits, continually threatens to devalue capital in these fixed forms’. If spatial fixity is the enemy of profit, then labour is both necessary and a problem for capital and it is an abstract problem that I argue has been addressed through performances of further abstraction—in the guise of imaginatively abstracting labour out of its spatial fixes.

The ‘automative imagination’, in the context of this article, is a heuristic that enables two things: (i) a critique of the spatial politics scale pregnant within debates around the opportunities and threats brought by automation; (ii) an interrogation of the normative character of different habits of considering and discussing automation. I am not using the neologism ‘automative’ to assert any kind of authority but rather as a pragmatic tool. Other words do not quite fit—to speak of an ‘automated’ or ‘automatic’ imagination does not describe the characteristics of automation but suggests the imagining is itself automated, which is not the argument I am seeking to make. My aim is not to validate/invalidate particular narratives of automation. A ‘technologically determinist’ reading of the automative imagination might see what is imagined as goals towards which companies are specifically working, largely driven by economic imperative (Galbraith, 1974). Predictions of intensified automation can thus become institutionalised, rendering them regulatory (Foucault, 2007: 29–54), insofar as they can be figured as programmes—which are ‘sets of calculated, reasoned prescriptions in terms of which institutions are meant to be reorganised, spaces rearranged, behaviours regulated’ (Foucault, 1991: 80; see also Rabinow, 2003: 39–40). The rationale here would be the calculation of the future as ‘present future’. Alternatively, if we understand anticipation as the performative representation of a ‘future present’, then power is less clear-cut. The ‘mechanisms of power’ (Foucault, 2007: 1–2) emerge from, and are an intrinsic part of, the relations in which the performative artefacts of anticipation—government policy documents, presentations, press releases, reports—are situated.

Automation articulates a relation of value, power and, significantly, of spatial division. Indeed, the separation out of specific workplaces in the rhetorics of ‘post-pandemic’ work is notable, with manufacturing, retail, travel infrastructure singled out as the workplaces that may be transformed or most notably ‘disappear’ (e.g. Wallace-Stephens and Morgante, 2020). Even when the work is predicted to remain, a ‘hybridisation’ of work—working from home, working remotely and so on—is posited as the new ‘normal’ (see Daugherty et al., 2020; Lund et al., 2020, 2021). When automation is used as a device to rhetorically disaggregate tasks from job roles or ‘occupations’, the nature of the relation between labour power and value is challenged. The class relation between employer and worker is skewed. Whereas a ‘worker’ with an occupation (at least notionally) has their humanity respected in the class relation with an employer, a task worker is, more easily, positioned as a logistical resource. The scale of the worker and their workplace is elided in the imagining of more abstract networks of labour-as-infrastructure. Work itself, in this re-imagining, lacks any specificity of place. Furthermore, whereas the temporality of the commodity form of labour is tied to the working day for an occupation, task labour has no such temporal restriction. The agency of workers, and the use-value of labour power, is deprecated.

The ‘power repertoires’ (Piven and Cloward, 2000) of an automation relation (discussed in An Automation Relation) put into question the distinction between ‘incremental’–‘non-incremental’ and ‘workplace’–‘non-workplace’ strategies. Even-so, resistance strategies (at least those protected by law in countries where they form part of the social contract) remain a significant factor in the power relations of automation, as unionisation and strike action by task workers such as Deliveroo riders demonstrates (Briziarelli, 2018). Perhaps most significantly, automation demonstrates the uneven spatial divisions of labour (pace Massey, 1984), such that, as Giddens (1990: 19) noted: ‘the visible form of the locale’, of the apparent work place—such as the distribution centre, the shop or the office—continues to ‘[conceal] the distantiated relations which determine its nature’. Thus, the ‘spaces of engagement’ (Cox, 1998) constituted in these global imaginaries of automation (which are significantly skewed to the global North and West) overcode and overpower the (more immediately local) ‘spaces of dependence’ (Cox, 1998) of the actually existing workplaces negotiating with implementations of automation. Imagining automation, similarly, does spatial (imaginary) work. The automative imagination figures workers not within ‘jobs’ but as an abstract infrastructure, it thereby deterritorialises automation at a local scale. Scale is thus a key element to the political work the ‘automative imagination’ does. The automative imagination is, before anything else, a form of geographic imagination.

There is no doubt automation has brought real social changes; this is, for example, evident in the ways particular job roles have all-but-disappeared in contemporary employment (see Autor, 2015). I argue that the ways automation is imagined and forewarned has an influence on what forms of change may be deemed acceptable or desirable. Even so, as much as these imaginings foreground worldviews, by the same token, they hide assumptions about who and what matters. Dispositions held by individual workers towards automation (pace Lin et al., 2022) may be nuanced while at the same time the language mobilised by corporations and governments (over) simplifies and elides them. Some actors, such as the consultancies publishing the reports analysed here, have significant interests or investment in their imagining. These are, we can posit, for reasons of financial investment, influence and power. The prominent references to these reports, and others like them, in House of Commons committee minutes and reports, such as those of the BEIS committee (2019, 2020, 2023), are testament to the potential for influencing how national conversations about the ‘future of work’ are engaged with by the UK legislature.

To propose or study the automative imagination is not to ‘debunk’ various settled narratives of automation, it is instead to recognise narratives of automation as forms of problematisation. Imagining automation in the ways discussed in this article is the creation of a problem; it is a problematisation of labour, workplaces and the infrastructures of production as a responsive relationship to cultural/political-economic uncertainty. Interrogating the ways automation is discussed by those with an interest in its advancement and intensification reveals the making-normative of debates about automation and work as a problematisation (pace Barnett, 2015). Central to imaginings of automation is a strategy for identifying a problematic ‘future present’ that, as Sturdy and Morgan (2018) suggest, apparently requires a solution. Evgeny Morozov (2013) calls this ‘solutionism’—technologies are proposed as ‘solutions’ to ‘problems’ that are only considered as such by the developers of those technologies. The ‘problem’, for such a perspective, is arguably always the same—the spatial fixity of value. All of these ‘solutions’ are thus, at heart, a means of mobilising value for capital. Even so, we can be more specific: what I have called the automative imagination is the problematisation of labour and where it is performed. A desire for compelling visions of a future improved by technology brings with it a risk of ceding power over that future, and the spaces in which it may exist, to the logic and terms of the problematisation. This is, in a sense, a supposed ‘thought leadership’, especially by management consultancies (Sturdy and Morgan, 2018) but also by governments and third sector organisations. Consultancies move ‘matters of concern’ from general discussion and debate into the realm of certainty. The ‘operative verb tense in projections’, Winner (2004: 37) argues, ‘is will. These things will happen’ and that certainty imposes its normality. ‘Thought leadership’, not least in relation to automation, is the attempt to impose norms.

Narratives of ‘determinism’ continue to be seductive, even when academic and journalistic commentaries highlight their skewed basis. I contend that it is still broadly true that there is a ‘curious paradox that plagues almost all discussions of technological change’ (Winner, 1977: 46). On the one hand there is the belief that technological development progresses under its own inertia ‘resists any limitation, and has the character of self-propelling, self-sustaining, ineluctable flow’ (Winner, 1977: 46). At the same time, there is a parallel but entwined belief that human beings are masters of their own technological destiny, that humans have a ‘conscious choice in the matter and that they are responsible for choices made at each step in the sequence of change’ (Winner, 1977: 46). Thus, the political work done by the automative imaginary is its becoming normative—in two senses. First, it is normative or prescriptive, in the Wittgensteinian sense, because what is imagined becomes the criteria by which we judge one another over how and what we say about something (e.g. ‘robots’) is appropriate, or not, to the context of discussion. Second, it is normative because it implies who or what is ‘normal’, and worthy of consideration, within the terms of the problematization of labour. We are invited to normatively consider automation personified in the shape of ‘robots’ and not consider the personhood of the abstract worker—who, in actuality, will be variously classed, gendered and racialized. For example, the BEIS Committee (2023) cites the Office for National Statistics in its 2019 study of the probability of automation that suggests women employees account for 70.2% of the jobs most at risk. The ways automative imaginations are performed involve judgements about who and what is worthy of concern.

The automative imagination articulates the performativity of problematisation in relation to the relationships constructed with and through the practices and imaginings of automation. Who and what matters in society, and at what scale, are powerfully framed within persuasive narratives of risk and reward. The ‘who’ and the ‘what’ of society and technology, and of social change, are not simply a cause and effect—people making changes illustrated by things—but rather the change is the relation. Social change is not separate from our analysis, but rather intimately includes, and indeed is affected by, how we narrate it. Often, arguments around change in society related to technology considered as ‘disruption’ accept simplified narratives created by those with particular interests as the premise for debate, making them the ‘common sense’, rather than question them. How we construct and receive ‘truths’ about changes to our society brought about by automation is political. I strongly agree with Lin (2022) in arguing that the changes enabled and prevented by technologies and those who develop, sell and use them is central to the contemporary politics of automation. How automation is problematised, by whom, and the ways employment is (re)imagined in stories about our ‘future of work’ matters. The problematisation of automation and labour that constructs as necessary the abstraction of ‘jobs’ and reimagines workers as labour-as-infrastructure has potent political consequences. I argue that the assumptions of who and what matters in forms of automative imagination both reflect and sediment contemporary inequalities. Who performs the construction and reception of ‘truths’ in relation to automation demands political awareness and, ultimately, political action, not least when those arguments influence government policy—as illustrated by the BEIS reports (House of Commons—BEIS Committee 2023; House of Commons—BEIS Committee 2020; House of Commons—BEIS Committee, 2019). The concept of an automative imagination contributes to understandings of the geographies of automation by providing a heuristic tool for interrogating how the automation relation illustrates a problematisation of labour. Using the automative imagination as a lens for inquiry asks for what kinds of ‘problem’ automation is situated as a ‘solution’, by whom and how they do so; and thereby how they are or may be contested.