Automation interrupted_ How autonomous vehicle ...

Political Geography 65 (2018) 57–66

Contents lists available at ScienceDirect

Political Geography journal homepage: www.elsevier.com/locate/polgeo

Automation interrupted: How autonomous vehicle accidents transform the material politics of automation

T

David Bissell School of Geography, University of Melbourne, 221 Bouverie St, Carlton, Victoria 3053, Australia

A R T I C LE I N FO

A B S T R A C T

Keywords: Automation Posthumanism Digital technology Mobilities Materiality Labour Politics

This paper develops our geographical understanding of the material politics of automation. Through the empirical site of the autonomous vehicle, the paper argues that dominant understandings of the politics of contemporary automation draw on a restricted understanding of materiality where political agency is concentrated in the hands of powerful individuals or institutions. However, this focus potentially obscures the complex material agencies of the systems of automation themselves. In response, this paper develops the conceptual potentials of the accident to bring these overlooked interruptive material agencies to the fore. This provides us with an opportunity to appreciate how the sites of power in systems of contemporary digital automation are more multiple and dispersed than is often assumed. In making this argument, this paper seeks to contribute to political geographical research that has turned to questions of ontology to pluralise the sites of politics and diversify the agents of political change.

Introduction In March 2017, online media was abuzz with reports of yet another incident involving an autonomous vehicle. In this case, an Uber driverless car was involved in a high-speed crash in Tempe, Arizona. It was reported that the accident happened when the driver of a second vehicle failed to yield to the Uber car when making a turn. The two vehicles collided causing the autonomous car to roll onto its side. This accident happened in the wake of a series of other accidents involving autonomous vehicles being tested by other major technology companies, including Google, Tesla and NuTonomy. Journalistic commentary at the time became preoccupied with debating the safety of autonomous vehicles, intensifying the already heated debates amongst lawyers and insurance companies about the legal dimensions of evaluating who is responsible when autonomous cars crash. That this was an Uber vehicle also raised pressing concerns about the labour implications of the development of autonomous vehicles. Autonomous vehicles are a disruptive technology. They have the potential to disrupt a system of mobility that has become hardwired into the economic, social and cultural landscape since the early part of the twentieth century (Dennis & Urry, 2009). They have the potential to disrupt the spatial morphology of cities; the discipline and control of vehicle occupants; the generation of public revenues through vehicle taxation; the livelihoods of currently employed drivers; the power geometries of access; and the viability of other modes of transport. Over the next few decades, autonomous vehicles are expected to become a

major part of everyday life (Lipson & Kurman, 2016). Yet relatively little attention within geography has been devoted to evaluating the politics of automation, leaving many questions unanswered about the social transformations that automation is giving rise to. This paper is motivated by profound unease about technologically determinist accounts that imply that disruptive automation is both inevitable and incontestable. Countering popular narratives about the likely impacts of intensified automation that stress automation's efficacy, this paper takes the autonomous vehicle accident that happened in Tempe as an opportunity to enhance our geographical understanding of the sites and operation of power in automated systems. By tracing some of the material agencies of the accident, this paper explores how we might become more attuned to the multiplicity of sites and operations of power in complex automated systems. It considers how the conceptual potentials of the accident might not only be a privileged epistemological moment that helps us to locate power, but also an ontological event that transforms the contours of power in both explicit and subtle ways. Drawing our attention to the inventive material agencies that are implicated in processes of technological change, this paper invites us to think anew the spatialities of automation. Conceptually, this paper is situated within the turn in political geography to questions of ontology which are concerned with exploring the nature of how things are. Research within this turn has sought to pluralise the sites of politics and diversify the agents of political change beyond the state, thereby introducing new and intriguing objects of analysis for political geography. One important strand of this work has

E-mail address: [email protected]. https://doi.org/10.1016/j.polgeo.2018.05.003 Received 1 November 2017; Received in revised form 24 March 2018; Accepted 14 May 2018 Available online 19 May 2018 0962-6298/ © 2018 Elsevier Ltd. All rights reserved.


D. Bissell

been accompanied by popular commentary that argues that intensified digital automation will invariably give rise to a chilling dystopia (Ford, 2015). The power of these ‘strong’ proclamations has contributed to a genre of thinking about the role of technology which Urry (2016) terms the ‘new catastrophism’. Though the changes that these reports predict have yet to unfold, such prophecies have influential performative powers (Anderson, 2010). Referring to the current intensification of digital automation, Wajcman (2017) notes how the power of affectively-charged language in these accounts such as ‘scary’, ‘frightening’, and ‘a perfect storm’ are an indication of the torsion of pleasure and pride that we seem to take in the idea that a robotic utopia or dystopia is on its way. A problem with many popular accounts of automation is their technological determinism, where the location and operation of power is typically displaced to the technologies themselves, without due consideration of the possibility of a more complex range of forces at play, including the institutional interests which might be guiding these developments. Through their technological determinism, popular accounts overlook crucial political questions about the sites of power and combinations of forces that might be critical to understanding how contemporary automation is evolving. The advent of new forms of automation, such as the autonomous vehicle, clearly intensifies the need for more nuanced thinking about the politics of contemporary automation. In this section, I set the scene by describing some of the key geographical contributions to these emergent debates. Geographical thought contains key legacies that can help us to situate automation historically. Economic geographers have consistently underscored Marx's observation that the intensified mechanisation of production increases profitability for capitalists as the relative demand for labour falls (Peet, 1975). However, the geographical implications of automation is complex, as Massey's (1984) work on the spatial divisions of labour demonstrated. Her work highlighted the need to consider changing production processes in industry-specific ways which could then help to understand transformations in the specific kinds of labour that were required, their gendered composition, and their geographical recomposition. Writing at around the same time as Massey, Wallace (1985) summarised some of the major changes taking place in the evolution of agri-food systems, highlighting the geographical distinctiveness of these changes whilst also drawing attention to how automation was having impacting on different parts of agri-food systems. Cultural geographical enquiry has explored the impact of automation in terms of the more micro-spaces of production themselves. Work on the transformation of manufacture in Taylorist production during the early decades of the twentieth century emphasised how industrial mechanisation was not just about replacing workers with machines, rather it also involved new ways of managing workers. Cresswell (2006) describes how workers in different occupations were increasingly subject to scientific management to improve their efficiency. From bricklaying to cleaning, key here was the excision of superfluous movements, so that bodies could operate more like well-oiled machines. Even domestic practices in the kitchen became the subject of motion studies that aimed to rationalise tasks. Workers in this regard were being invested with new forms of productive power, through the inculcation of new habits and redesigned spaces that enrolled them into labouring in new ways. Where previous geographical accounts of automation have tended to focus around the spatial and bodily politics of mechanisation, contemporary geographies of automation have foregrounded the pivotal role of digital technologies in changing the politics of automation. The political implications of digital automation have been traced through three new forms of governance. The first of these is algorithmic governance, which spotlights how the algorithm is a new form of political authority used to govern populations. Algorithms here can be understood as ‘both technical process and synecdoche for ever more complex and opaque socio-techinical assemblages’ (Amoore & Raley, 2017, p. 3). Accordingly, algorithmic governance is ‘the manifold ways that

been to rethink geopolitics through the complex and diverse materialities of force relations rather than just social relations (Meehan, Shaw, & Marston, 2013; Shaw & Meehan, 2013; Fregonese, 2017), opening up political geography to a broader ecology of distributed, interacting agencies and inhuman material forces (Woodward, 2014). This turn has therefore helped to problematize previous geographical understandings about the nature of power's operation, which is less premised on the determinate identities of things, people or institutions, and more an effect of the modulation of indeterminate capacities (Anderson, 2014). Inspired by this turn to ontology, this paper develops two themes of political geographical enquiry. The first theme is the politics of new digital technologies. This is a nascent field concerned with exploring the emergent power dynamics of a diverse range of new technological forms. Influenced by posthumanist theories, this body of work has sought to replace determinist accounts of technology's power with more messy and fragmented theorisations of the agencies of the technological nonhuman and its role in the emergence and transformation of urban life (Ash, 2015, 2018; Richardson, 2017, 2018; Rose, 2017). The second and related theme is the politics of future mobilities. This field has consistently emphasised the significance of relations of movement and stillness for our geographical understandings of space. Research in this field has been concerned with evaluating the diverse enablements and constraints experienced by the bodies caught up in these movements, and speculating on the effects that new technological developments may have on these power relations (Adey, 2017; Sheller, 2016; Urry, 2008). My overall argument is that we need to develop a more expansive understanding of power that acknowledges how the material agencies of the systems of automation themselves might not necessarily be reducible to the intentions of powerful individuals or institutions. Automation's accidents bring these overlooked material agencies to the fore, providing us with an opportunity to better appreciate how the sites of power in systems of contemporary automation are more multiple and dispersed than is often assumed. My argument also seeks to provide a corrective to accounts of technological change that stress the efficacy of such technologies. Immanent to the concept of the accident is a negative that is often overlooked when tracing the operation of power and evaluating political agency (Philo, 2016). As Rose reminds geographers concerned with the technologies and capacities of state power, we have a tendency to ‘become over attentive to what power does and blind to that which power fails to do’ (2014, p. 217). My argument unfolds in four sections. The first section sets the scene by describing how popular technologically determinist discourses about automation have been countered by geographical research that has sought to trace the more complex power relations at play in automated systems. The second section moves forward by suggesting that there is an overlooked ambivalence concerning the materiality of automation that requires new methodological and conceptual tools. The third section introduces the accident as one such conceptual tool for thinking about automation differently, by bringing the thought of Bernard Stiegler and Catherine Malabou into dialogue to foreground its inventive capacities. The fourth section speculates on what the Tempe autonomous vehicle accident produced along four different lines, drawing attention to the multiple sites of material politics. The final section evaluates the role of ontological thought in opening up possibilities for progressive and democratic intervention concerning the ongoing evolution of systems of automation. Political geographies of automation Automation has been prominent in recent popular debate, influenced by a series of recent high-profile reports that prophesise how a high proportion of jobs will be lost to new forms of technological automation (Brynjolfsson & McAfee, 2014). Frey and Osborne (2015), for instance, predict that almost 50% of existing jobs in the UK are at risk as a result of automation over the next two decades. These reports have 58


D. Bissell

provided important correctives to technologically determinist accounts of technological change. This work has therefore begun the important work of pluralising the sites of politics, by tracing some of the complex operational logics built into these technologies. In the next section, I explain how turning to the materiality of automation itself as an object of analysis introduces conceptual and methodological challenges that furthers the task of pluralising the politics of automation.

algorithms and code/space enable practices of governance by conducting and mediating behavior’ (Crampton, 2016, p. 141). Automation is central to algorithmic governance, although its functionality has shifted from pattern recognition algorithms to non-rule-based learning algorithms which effectively surface things or people for attention of analysts. Key political questions that emerge from this concern the prejudices built into these algorithms, and the site of responsibility for the authority of decision-making algorithms which are increasingly deterritorialised from conventional territorial jurisdictions in cloudbased data banks (Amoore & Raley, 2017). Shaw (2017) pushes these themes further in his work on drone warfare, where algorithms are fused with sensors and motors to form robotic bodies, raising vital questions about their political and ethical capacities. The second area where automation has been explored is through work on logistical governance. Where logistics is concerned with the efficiency of linkages (Thrift, 2004, p. 589), digital automation intensifies this efficiency through enhanced control and securitisation of supply chains. Much of this work explores the implications of intensified automated control and security on labourers (see also Bissell & Del Casino, 2017). Kanngieser's (2013) work is exemplary in this regard. Her work explores the way automated systems intervene in the everyday governance of logistics workers through intensified securitisation and monitoring practices. Rossiter's (2014) exploration of contemporary logistics emphasises how the digital automation that is central to the smooth operation of supply-chain capitalism devalues the worker, because ‘labour and workers’ knowledge is increasingly transferred to the algorithmic agency of machines and code’ (2014, p. 57). The third area of focus is the automation of urban governance. This builds on work on algorithmic governance by exploring the implications of automated management and coordination used in ‘smart city’ applications and practices (Leszczynski, 2016). At the scale of the city, Leszczynski's (2016) work has explored how urban algorithmic governance is a speculative practice that anticipates particular kinds of future city. This theme is developed through work by Kitchin, Coletta, and McArdle (2017) who explore how city dashboards and urban control rooms are implicated in new systems of automated urban governance. They note how automated systems are implicated in a general shift in urban governance from discipline to control, in other words, from the moulding and restricting of specific practices to the modulation of affects, desires, opinions and actions. McNeill (2016) extends this work further by analysing the individuals and institutions involved in facilitating ‘smart city’ developments, as well as shedding light on the way that some technology companies providing services that rely on digital automation exert powerful self-serving influences on urban regimes. In contrast to the technological determinism that characterises popular discourses, these related bodies of geographical work demonstrate how the political implications of automation are complex and contingent. As Kitchin et al. (2017) emphasise, there is no one form of governmentality being enacted by automated smart city technologies. They have mutable logics and they are translated and operationalised in diverse and context-dependent ways. Relatedly, Dodge and Kitchin's (2007) work on the automated management of drivers and driving spaces which involve automated surveillance systems and capture systems to reshape activity highlights the multiple logics of power at play, rather than one single dominating logic. Some logics are benign, some are empowering, whereas others enhance the power of the state and corporations. Furthermore, Rose's (2017) evaluation of new automated technologies reminds us that the way that they are experienced by city dwellers is contingent and circumstantial. To summarise, in this section I have explained that there is currently significant concern about the social impacts of new digital automation technologies. Popular discourses about contemporary automation tend to be characterised by either utopian ‘boosterism’ or dystopian angst; or ‘oscillating breathlessly between technophilia and technophobia’, as Lamarre (2012, p. 44) puts it. However, in response, geographers have

Diagramming automation The geographical contributions summarised in the previous section highlight the importance of developing more nuanced ways of understanding the politics of complex automated systems. What this work demonstrates is that there is an ambivalence to automation (see also Richardson, 2018) that becomes obscured through popular discourses about the location of and operation of power in automation systems. The remainder of this paper seeks to extend our geographical understanding about the politics of automation by foregrounding the materiality of automation, in other words, the nature of what automation is and how specific events and processes implicated in automation occur. This can help us to think differently about the location and operation of power in automation systems. This section outlines three methodological challenges associated with the study of complex automation systems. First, there are many objects and entities that processes of automation are contingent on that are often purposefully backgrounded. Stacey and Suchman (2012) situate this point historically in their exploration of the rise of automata. They describe how part of the magic of automata requires suspending the human labour that is often doing the automating. These might be the mundane ongoing bodily labours that automation requires, which might include practices of maintenance and repair (Carr & Gibson, 2016). The concealment of specific parts of automated systems is strikingly clear in Irani's (2015) work on gig workers who undertake low value work through platforms such as Amazon Mechanical Turk. This company extols the virtues of creativity, innovation and agency undertaken by its software engineers, whilst at the same time covering up the unglamorous, tedious labour undertaken by workers. Indeed, as Stacey and Suchman argue, ‘restoring the visibility of labour and care might help to recover the lived body-work that gives animation and automation their life’ (2012, p. 31). Second, and relatedly, complex automated systems have an ontological complexity that can make them challenging to trace and delimit. Through their concept of the automatic production of space, Thrift and French (2002) argue that automation is part of the city's technological unconscious. In other words, rather than something that presents itself to conscious attention, automation is the ubiquitous background combination of hardware and software that beckons the city into existence. Their examples are mundane automatic technologies such as automated traffic lights that sense a car's presence, but which in the normal onflow of everyday life remain below our threshold of consciousness. This presents a methodological challenge since, as Wise (1998) points out, what was formerly foregrounded becomes woven into the background of our habitual ways of life and can therefore be difficult to make out. Third, given the complexity of systems of automation, it is often difficult to have access to a comprehensive knowledge of what is actually taking place. As geographers have argued, there are techniques that complex automated systems employ that purposefully make gaining a more extended perception of that system very difficult. In their discussion of automated logistical systems, Gregson, Crang, and Antonopoulos (2017), for instance, talk about how in such systems there are many designed-in frictions and blind spots which give rise to a multitude of partial, situated ‘oligoptic’ knowledges that workers must negotiate. This serves to restrict a more comprehensive ‘panoptic’ gaze. Furthermore, the situated work of analysing technological systems is a practice that always enacts its object differently. What these three methodological points indicate is that the objects 59


D. Bissell

decision-making skills. Such discourses indicate how autonomous vehicles are typically evaluated from a utilitarian perspective according to their functionality (Verbeek, 2008). Testing regimes are about improving the functionality of automated technologies. From this dominant utilitarian perspective, accidents are viewed as an inevitable part of ongoing technological development. However, in response, in this section I propose an alternative way of thinking about the accident in relation to automation. My aim here is to consider the conceptual potentials of the accident, and how the accident can potentially enrich our understanding of the politics of automation, rather than just be interpreted as a temporary set-back or technological flaw. Heeding the methdological and conceptual ambivalences to automation outlined in the previous section, here I consider automation ontologically with assistance from the work of Bernard Stiegler. By way of context, Stiegler's work is useful here because he helps to collapse some of the divisions between humans and machines that popular accounts of automation have tended to revert to. His work has consistently engaged with how our being in the world is always already technical (Stiegler, 2010). Indeed, the guiding thread in his work insists that there is no human that stands separate from technology, broadly construed. On the concept of automation, Stiegler (2017) reminds us that far from just a concern related to understanding evolving modes of capitalist production, we need to understand how automation is at the very core of life itself. Drawing on diverse examples from biology and psychology, he invites us to reflect on how many living processes happen automatically. For instance, cells replicate automatically. So too do our drives, reactions and reflexes. Therefore, rather than imagining that to be autonomous or to be automated are oppositional, his point is that to be autonomous requires the integration of many automatisms. Such an argument has parallels in the geographical literature on habit which highlights how so much of what we do takes place on autopilot (Bissell, 2015). For Stiegler, the problem is not the general concept of automation, because so much of who we are is formed of automatisms. Rather, it is that certain forms of automation are inducing potentially harmful effects. It is the intensification of automated digital calculation which Stiegler finds especially problematic. Evidencing his Derridian roots, Stiegler sees automation as pharmakological, that is, as both having potential for poison and cure. On digital automation, he says ‘it is inevitable that this Pharmakon will have toxic effects, as long as new therapies or therapeutics are not prescribed—that is, as long as we do not take up our responsibilities’ (2017, p. 40). However, he is particularly concerned that through the interconnected infrastructures that are encapsulated in the idea of the smart city, such as the autonomous vehicle, we are evolving into a society of hyper-control. To paraphrase Stiegler's argument using his own distinctive vocabulary, previously long-circuits of psychic and collective individuation characterised by negentropic indeterminacy are increasingly short-circuited by ‘digital processes of transindividuation, founded on automation in real time’ (2017, p. 41). His point is that digital automation, such as those involved in autonomous vehicles, gives rise to the possibilities for controlling other automatisms. Stiegler shows how automated digital systems work by analysing our backward-tracing retentions in order to automatically create specific forward-leaning protentions, or expectations. Stiegler's central issue is that, because of their hyper-standardisation, these processes of digital automation lead to entropy, or determinism. This is probability without bifurcation, where new forms do not emerge. Stiegler's mission, then, is to speculate on tactics for catalysing ‘negentropic’ ways of life, ways of living that generate creative indeterminacies which run counter to digital automation's tendency towards entropy and standardisation. Stiegler's writing shares similarities with the political geographical work on digital automation introduced earlier through the way that it highlights the ambiguities of automation. However, his ontological account sharpens our attention to the importance of locating sites where indeterminacies might be introduced. Here, I take up Stiegler's

and spaces that make up systems of automation are complex and ambivalent. In this context, a seemingly coherent automated object such as the autonomous vehicle becomes much less fixed. Berlant helps us to appreciate this complexity by pointing out that the coherence of any object is often just an analytical effect. She argues that ‘objects are always looser than they appear. Objectness is only a semblance, a seeming, a projection effect of interest in a thing we are trying to stabilize’ (2016, p. 394). Acknowledging these qualities means that it becomes more difficult to make stronger claims about the politics of contemporary automation which have the potential to objectify the world in a way that belies the multiplicity of events and processes that might be taking place. So how might geographers develop new ways of exploring the material politics of automation? Writing about technological aesthetics and the problem of ‘unrepresentable’ problems, Parikka (2013) suggests our challenge is to create different ‘diagrams’ of the phenomena in question. He says that we need techniques that ‘elaborate the not immediately perceptible’ (2013, p. 114), and recognise that these diagrams then are a new mode of perceiving the phenomena in question. There are many possible ways of doing this diagramming. One approach would be to explore the material specificity of automation technologies. Ash's (2018) geographical research is exemplary in this regard. His post-phenomenological enquiry into the politics of various ‘smart objects’, including the autonomous vehicle, diagrams the precise capacities for acting and sensing that different components of automated systems involve. His posthumanist focus on the in-situ technical capacities of autonomous vehicles allows him to go beyond broad-brush diagnoses of the location and operation of power towards a more precise investigation about how different technical components act together to create specific forms of space and time. A second geographical approach draws from STS to trace the extensiveness of an object outwards from a specific site to attend to the spatial complexities of automation (see Johnson, 1988). Investigation here might seek to trace the multiple locations where automation happens, and subsequently, diagram the precise nature of the different versions of automation that are constructed and performed at different sites. For instance, for autonomous vehicles, a different performance of automation might take place in airport bookstands; in media articles; in different institutions; and in participant interviews. In each of these sites, automation might be differently detectable and differently expressed. This form of diagramming is therefore a way of attending to the spatialities of automation in terms of its complex topology. In the remainder of the paper, I offer an alternative way of diagramming automation to explore the material politics of automation. My approach develops a speculative way of thinking differently about automation through the event of the accident. I do this not to make generalised claims about automation, but as a way of thinking about its politics anew. This way of acknowledging automation's ambivalence responds to Ash et al.’s invitation to ‘critically reflect upon the wider dispositif or assemblage of the digital’ (2018, p. 36-7). Crucially, taking this more material approach is to recognise the insufficiency of thinking about the operation of automation's power in a simplistic top-down or bottom-up manner (Luque-Ayala & Marvin, 2016, p. 2113). Following Forlano and Halpern, in doing this, my aim is to show how technologically determinist narratives about automation overlook the ‘complex, non-linear, and iterative relationships’ (2015, p. 33) at play. Automation's accident In the days and months that followed the Uber autonomous vehicle accident that opened this paper, the dominant responses in the popular media orbited around the question of whether autonomous vehicles are safe. Referring to a previous accident involving an autonomous vehicle, one information scientist at RAND corporation suggested, ‘when it comes to safety, autonomous cars are still ‘teen drivers’’ (Hsu, 2017), implying that much needs to be done still to improve their driving and 60


D. Bissell

event where new and unexpected forms emerge through discontinuities—provides us with an opportunity to consider the potential bifurcations in contemporary forms of digital automation, and, in doing so, a richer appreciation of the material forces in play.

speculative challenge by turning to the concept of the accident as a way of thinking differently about the politics of automation. My argument here is that within processes of automation are counter-tendencies that take automation elsewhere, subverting the tendency that Stiegler describes towards standardisation. I explore the concept of the accident by turning to Catherine Malabou's writing on the topic because she provides a powerful ontological argument for thinking about the transformative potentials of accidents. The Ontology of the Accident (2012a) continues Malabou's fascination with the idea of plasticity as a vital characteristic of matter and its malleability. From the neuronal brain to the organisation of capitalism, her previous writing shows such systems are creatively adaptive, and demonstrate the capacity of matter to both give and receive form in ongoing processes of metamorphosis that are never defined in advance. Where much critical thought has been devoted to the idea of ‘constructive plasticity’, a form of plasticity that sculpts and models connections, the accident gives us pause to consider the forms that emerge from discontinuities, breaks and ruptures, or ‘creation through the destruction of form’, as Malabou (2012b, p. 17) puts it. Malabou's empirical sites for exploring the accident are a series of events, often unfortunate, that can befall subjects. She focuses on a range of explicit traumas, such as brain damage and excessive alcohol consumption, and more subtle traumas, such as old age. Each of her examples clearly critiques the idea of a stable, unchanging subject. However, what sets these diverse events of transformation apart from the incremental developments of constructive plasticity is that a new subject emerges from its own destruction. In each of these accidents, ‘the path splits and a new, unprecedented persona comes to live with the former person, and eventually takes up all the room’ (2012a, p. 1). Malabou's interest in theorising the accident as a form of destructive plasticity responds to what she sees as too strong an emphasis on continuities rather than discontinuities within various schools of thought, especially psychoanalysis, where through the persistence of traces and memories, a subject is always presumed to be conditioned by and therefore beholden to their past. Malabou therefore considers the idea of the accident in a rather different way to other social scientific approaches. Arguably the dominant approach within geography has been to emphasise the revelatory potentials of accidents, through their capacity to reveal the usually-backgrounded ‘normal politics’ of circulation. As Graham explains, ‘infrastructure disruptions bring fleeting visibility to the complex practices and technologies, stretched across geographic space, that continually bring the processes of urban life into being’ (2010, p. xi). Studies that emphasise the revelatory potentials of accidents draw inspiration from STS approaches where the focus is on tracing the complex webs of causality that can help to understand why an accident took place. Take Law and Mol's (2002) exploration of a fatal train crash in the UK, for instance. Their backward-tracing evaluation of the accident explores the contours of the accident investigation process. This is a task which is principally about assigning responsibilities (and, ultimately, blame) for the accident. Through detailing some of the complex performativities at play in the accident investigation, Law and Mol highlight the multiple and distributed forces at play, clouding any simple ‘billiard-ball’ notions of causality. In contrast, rather than trying to figure out why an accident took place by reconstructing possible lines of causality, Malabou encourages us to focus on the bifurcations and splits that accidents create, and the new forms that might emerge. What is significant about this is that the accident hints at a kind of interruptive agency that is not reducible to singular sites of political control. Taking her cue, in the next section of this paper, I develop Malabou's ideas about destructive plasticity to speculate on the novel forms that might emerge through accidents involving digital automation. The breakdown of automated systems provides us with a privileged analytical moment where new ways of perceiving and sensing the complexity of automation might emerge. My argument is that the accident—understood in Malabou's terms as an

The interruptive politics of the accident In response to Stiegler's anxiety that digital automation produces hyper-standardisation, and developing Malabou's ideas about the generative potentials of accidents, my aim in this section is to speculate on how interruptions involving digital automation might produce new and unexpected forms. I speculate on the transformative potentials of the accident through zooming in on the accident involving an Uber autonomous vehicle that was introduced at the start of this paper. Thinking through one accident provides an opportunity to draw out the complexity of what might happen when complex forms of automation fail. It also enables us to draw out the multiple spatialities that are involved in this one accident. To do this, this section explores four sites that the accident intervened in. Each exploration draws out a different spatial form, and each turns the spotlight onto a different set of materialities, thereby enabling us to more comprehensively evaluate the material politics of this accident. I was based in southern Arizona at the time of this accident, and so my decision to focus on the Tempe accident is circumstantial; its affectivity for me intensified through its proximity. I experienced the story of the accident as it unfolded on local news and social media; and through conversations with colleagues and Uber drivers. Methodologically, the four sites selected draw from my inhabitation of this event and the different lines that were palpable for me at the time. Conceptually, the four sites explored here help us to appreciate the multiple relations of force at play in the accident. Site 1: Intersection of McClintock Drive and Don Carlos Avenue, Tempe The first transformation that happened is that the accident brought vehicles to a standstill in Tempe. The materialities here involve the algorithm-driven sensing devices at the heart of the automated technology itself, and the onflow of life happening on this suburban street. The site of the accident in this regard is the in-situ vehicle crash. Zooming in on this site invites us to draw out the specificity of the accident and the material politics at play. The accident happened at an intersection in Tempe where an Uber autonomous Volvo was hit by a Honda CR-V. News reports after the accident indicated that the driver of the Honda was attempting to turn left across three lanes of traffic. The driver was able to cross the first two lanes that were backed up with cars, and so she thought that she was clear to cross the third. In the police report, the driver commented that since the third lane had no one approaching it, she was clear to make her turn. However, as she got to the middle lane, about to cross the third, she saw a car flying through the intersection, but couldn't brake fast enough to avoid a collision. The autonomous Uber was driving through a yellow light just below the speed limit at 38-mph when it was hit by the Honda, causing it to turn on its side. News reports indicated that the autonomous Uber did not break road rules since it had the right of way approaching the yellow light. However, bystanders contested this saying that the Volvo was at fault for trying to ‘beat the light and hitting the gas so hard’. I suggest that this accident created in-situ transformations. The two Uber ‘safety drivers’ and the driver of the Honda were hurt, and three vehicles sustained extensive damage. Through the bystander testimony in the police reports, we can discern how the accident created different local distributions of feeling at the site: empathy with the driver of the Honda vehicle, and antipathy towards the autonomous vehicle. However, in doing so, the accident intensified an appreciation of the complexity of urban materiality and the shortcomings of autonomous vehicles to respond to this complexity. Ash (2018) provides a lucid overview of how a suite of different sensors in autonomous vehicles 61


D. Bissell

Site 3: ABC 15 News

gather data on surrounding objects, categorise them through algorithms, and instruct the car to move or stop accordingly. However, this accident shifts our appreciation of the capacities of autonomous vehicles to anticipate and respond to the subtle cues, intuitions and instincts that human drivers learn through years of driving. Such autonomous vehicle incapacities have been drawn out by artistic interventions such as Bridle's (2017) performance piece Autonomous Car Trap 001, where a circle of salt placed around an autonomous car is sensed by the car as an unbroken white line that cannot be crossed, thereby pacifying it.

The third transformation that happened is that the accident shifted public dispositions. The materialities here involve the images of the accident and their capacity to affect people. The site of the accident here in this regard is even more dispersed, as these images circulated rapidly through global news media, and myriad social media channels. Tracing this line of the accident invites us to consider how the accident potentially participates in the shift in public dispositions towards autonomous technologies, as well as the reputation of Uber. News of the Uber accident in Tempe was broken by two photos taken by Mark Beach who uploaded them to Fresco News, another tech startup company that pays users for their visual content. The dramatism of the images was intensified by their close-up nature, the Uber Volvo turned on its side, headlights still shining, leaning into a silver Ford SUV with a crumpled rear side, giving a sense of the powerful forces involved in the collision. As news of the accident in Tempe unfolded, images of the crash began circulating, first on social media, then on late afternoon news channels showing video footage with police officers pacing the scene bathed in peachy sunlight, and then on online, accompanied by written reports. As news reporters noted, and as social media feeds verified, these images have performative potentials to affect people's dispositions towards autonomous cars and their perception of safety. However, more than this, the capacity of the accident images to affect is contingent upon a longer series of events within which it is situated. As the reports of the accident demonstrated, the images circulated in the wake of circumstances that were already highly charged. Two months earlier, the company was seen to be profiting from Trump's first ‘travel ban’ executive order by contravening an airport taxi-drivers’ strike at JFK airport and lifting surge-pricing to drive business. This gave rise to the #deleteUber campaign encouraging users to delete the app in protest. This antagonism was intensified the following month by a plethora of gender discrimination and sexual harassment stories, including the departure of a female engineer who accused the management of fostering a misogynistic culture, which led to prominent investors chastising Uber's leadership. Five days later, a video was leaked showing Uber CEO Travis Kalanick thrashing obscenities at a driver who claimed that, because of changed pricing conditions, he had lost almost a hundred thousand dollars through an Uber vehicle lease program. The transformative capacities of this accident are therefore not just spatially dispersed through the circulation of images, but that these images resonate within an experiential field that is conditioned by the charge of a series of previous events. The image circulated within charged circumstances that were created by a series of events that happened in quick succession that were progressively eroding the integrity of Uber's brand. Such transformations in public dispositions towards Uber can be discerned through the way the company began a brand overhaul shortly after the accident, replacing senior figures in a bid to change the culture of the organisation.

Site 2: Tempe, Pittsburgh, San Francisco The second transformation that happened is that the accident forced Uber to suspend their autonomous vehicle testing programme. The materialities here involve institutional decisions and communications between corporations and city governments. The site of the accident here in this regard extends out from the in-situ site of the accident, affecting other jurisdictions, and shifting the reputational landscape of the technology companies involved in developing autonomous vehicles. Tracing these lines of the accident indicates how the material politics of the accident therefore comes to affect technology companies themselves and their relationship with regulators. Reports of the accident stated that in the days following the accident, an internal investigation took place within Uber, and the company offered to share details of the accident with City of Pittsburgh, the other jurisdiction where Uber was testing autonomous vehicles, once the accident investigation was complete. What this indicates is how the accident forced Uber to re-establish trust with regulators who had previously treated them favourably. Tempe and Pittsburgh were testing sites principally on the basis of their accommodating regulatory environment. Three months prior to the accident, Uber moved its vehicle testing from San Francisco to Tempe owing to the more lenient regulatory environment in Arizona, following a ‘war of words’ between Uber and the California state government where Uber were infringing state regulations by not applying for a $150 permit to test autonomous vehicles in San Francisco. In Arizona, Uber only needed to hold a minimum liability insurance policy to operate autonomous cars, and they are not required to track accidents or report information to the state. In this regard, the accident intensified questions about the clash of public and private interests in terms of the strength and nature of regulatory oversight required to ensure the safety of autonomous vehicle testing, where different levels of government have to balance the benefits and seductions of facilitating ‘innovation’, with protecting the safety and needs of citizens. The accident also shifted the corporate landscape between rival technology companies involved in the race to develop autonomous vehicles. News reports took the opportunity to connect the accident in Tempe to the different approaches to development taken by rival technology companies, flagging how Google's Waymo programme is much longer established, having clocked up over two million miles of testing on public roads over an eight-year period, relative to Uber's much more recent entry into the field in mid-2016. These reports implied that Uber demonstrated a more cavalier approach to its development programme, and had led to an overblown estimation of its capacities. Furthermore, the accident also prompted news reports to restate accusations by Waymo that Uber had stolen their proprietary Lidar sensor technology through poaching computer scientists and engineers, and with it, valuable corporate knowledge. So, although the accident itself was not necessarily evidence of Uber's subordinate position in relation to Waymo from the point of view of its technical capabilities, the accident opened up a discursive space in which Uber's technical expertise and attitude to regulatory environments could be debated.

Site 4: N Campbell Avenue, Tucson The fourth transformation that happened is that the accident intensified capacities for some Uber drivers to sense the company's future plans. The materialities here involve the bodies of the drivers that Uber is currently dependent on, and their relationships with significant others. The site of the accident here involves a more intimate and personal rupture to the sense of security felt by Uber's drivers in terms their future plans and aspirations. Tracing this line of the accident therefore invites us to consider how the accident potentially participates in shifting the emotional constitution of a workforce who sense the threat of automation. The evening after the accident, I hailed an Uber to help carry my shopping home in Tucson, just over a hundred miles south-east of 62


D. Bissell

Tempe. My driver was at first upbeat and chatty, as the conversation segued from her busy day on the roads to how much she enjoyed driving. I learned a few snippets about her family who used to live on the Tohono O'odham Nation Reservation to the south of the city but had moved to Tempe for her to work, which included another job and driving for Uber to top up her low wage. Unprovoked, she asked me whether I had heard about the crash in Tempe on the news the previous evening. I said that I had. She said that she didn't know that Uber was testing driverless cars. ‘I have no idea,’ she said, pausing. ‘I have no idea what it means for us.’ I paused, not wanting to respond immediately, as the significance of her statement seemed to ripple out in different directions, suggesting uncertainty about both her ongoing position as a driver, but also a sense of uncertainty about the impact that this might have on her young family. In this brief encounter, I sensed how the accident had given rise to other, more intimate transformations. For my driver, I sensed in her words and the tone of her expression a hint of betrayal by a company that is already widely accused of putting its drivers at arms-length from an employment perspective, their status as independent contractors removing the obligations of Uber to provide a proper range of employment benefits. For her, the accident materialised the company's plans to transition to autonomous operation that she was previously unware of. The accident had therefore shifted her sense of security, not from the perspective of vehicle safety, but rather from the perspective of how long she might be able to continue to rely on driving for Uber to provide the necessary financial support to her family. The accident in this regard seemed to intensify a sense of precarity about her livelihood, introducing uncertainty about what she might do in the absence of this work; as well as threaten the sensory pleasures of driving that she had mentioned earlier. In this encounter, the accident materialised the spectre of technological unemployment, whilst at the same time reminding us how such transformations intersect with entrenched lines of social difference.

out to affect multiple material forms shows how accidents are sites of politics in and of themselves. As I have described, the accident at Tempe opened up a space for technical negotiation about how autonomous vehicles could interface with the complex materialities and rhythms of the urban environment; it opened up a space for policy negotiation between different jurisdictions; it opened up a space for bodily negotiation in terms of dispositions towards autonomous cars and towards Uber; and it opened up a space to negotiate futures for the already precarious life of an Uber driver. Accidents are not only revelatory but they are sites where things change. A focus on how the accident in Tempe took place through multiple materialities and how it resonated across diverse sites compels us to appreciate how the evolution of automated systems is complex. This complexity is partly an effect of the many different material dimensions involved. An appreciation of this material complexity is particularly helpful for developing a more nuanced understanding of power at play in complex automated systems. Where more technologically determinist understandings about the inevitable march of automation locate power in the technology itself, drawing out the multiple sites of the accident demonstrates how power in the evolution of automated systems is much more distributed. From bodily dispositions to policy regulations, there are multiple sites of negotiation involving different materialities. The intersecting materialities involved in automation's transformation demonstrated through the four sites of the accident explored in this paper indicates the need to think about automation ecologically. The activist thought of Félix Guattari is particularly useful in this regard. Where many geographical accounts of automation focus on excising a specific dimension of a system for analysis, Guattari's writing reminds us that the way that we dwell within technical systems is hugely complex and diffuse, caught up within specific evolutionary histories and the envisioning of specific futures. Whilst Guattari certainly warns us about the downsides to new technologies, through ‘machinic isolations’, ‘addictions’ and ‘infantalisations’, like Stiegler, he concedes that automation technologies are always contingent and circumstantial, and so are not intrinsically good or bad. As Genosko summarising Guattari writes, ‘everything depends on its articulation within collective assemblages of enunciation’ (Genosko, 2015, p. 8). Guattari's writing on machinic assemblages reminds us that systems always have multiple components that are economic, political, material and semiotic. So, to speak of automated systems is always to invoke each of these components. Furthermore, life takes place through overlapping registers of mental, social and environmental ecologies. Finally, these machinic assemblages are always evolving. As Guattari says, ‘a machine is something that situates itself at the limit of a series of anterior machines and which throws out the evolutionary phylum for machines to come’ (Guattari, 1996, p. 126). Through its focus on these ecological contingencies, what Guattari's ontological thinking offers, then, is a way of appreciating the ‘wiggle-room’ that exists in any situation, by virtue of its material complexity. Countering more oppressive analyses of technological automation that can suggest a lack of alternatives, Guattari emphasises how there are always potentials for minor ruptures that can open up alternatives to dominant realities. Owing to the material complexity of systems, power cannot be understood as centralised or top-down. Writing on technological evolutions of complex urban ecologies in a manner that targets ‘rise of the robot’ discourses, Guattari notes that ‘at the limit, there is just one machine on the horizon … Not as the science-fiction of yesteryear imagined, in the form of a tyrannical mega-machine, but as a powdery molecular machinic multiplicity’ (2013, p. 74). This observation echoes previous analyses of power by complexity theorist De Landa who was similarly distrustful of centralised command imperatives. Highlighting the self-organisation of emergent complex systems, De Landa argued that ‘from the point of view of the machinic phylum, we are simply a very complex dynamical system. And like any other physical ensemble of fluxes, we can reach critical points (singularities, bifurcations) where

Material ecologies of automation Writing about the relationship between new technologies and accidents, Virilio argues that dominant, modernist technophilic discourses typically overlook how new technologies simultaneously invent new kinds of accidents. Offering a powerful corrective to positivist interpretations of technological progress, his writings draw our attention to the often obscured negative in the development of new technologies. Thus, he argues that ‘to innovate the vessel was already to innovate the shipwreck, to invent the steam engine, the locomotive, was again to invent the derailment, the rail catastrophe’ (1989, p. 81, original emphasis). Virilio's ‘anti-futurology’ (Crogan, 1999, p. 173) challenges the utilitarian notion that accidents are merely temporary set-backs on an inevitable march of technological progress. His writings invite us to recognise the irreducible contingency of accidents to technological change. Paralleling Malabou's (2012a) argument about the accident as a form of destructive plasticity, Virilio encourages us to apprehend the accident itself as inventive. If we follow this logic, we can suggest that the development of automated technologies therefore introduces a qualitatively different kind of accident. A normative, technocratic response to the accident in Tempe involving an autonomous vehicle would suggest that the accident is merely a temporary setback along a path of technological development. However, this is just one dimension of the accident which is much more than an exercise in establishing causality and assigning responsibility. In tracing the different lines of the accident across different sites, as the previous section has done, what I have attempted to show is that accidents can be understood as transformative events as they reconfigure material relations in both explicit and more subtle ways. Where others have highlighted the capacity of accidents to reveal the politics of ‘normal circulation’ of systems whose politics is often occluded through their complexity (Graham, 2010), highlighting how the accident ripples 63


D. Bissell

highlights a material politics whose complexity means that we need to be cautious about any singular evaluation about the locus and operation of power in the emerging autonomous vehicle system. Guattari's writing in particular encourages us to question whether there might be a much more dispersed ecology in terms of the evolution of autonomous vehicles without transcendent control where, because of this complexity, the potential for accidents is endemic. The multiple and dispersed sites of the accident in Tempe alert us to a knotty complex of material powers. The accident opens up political questions about the judgements and biases made by autonomous vehicle sensors; the judgement and reporting of such incidents by police and bystanders; the sharing of information between different jurisdictions involved in autonomous vehicle testing; the support by different jurisdictions to embrace autonomous vehicle developments; the reputation of competing autonomous vehicle developers; the nature of public dispositions regarding autonomous vehicle development; and the dispositions of drivers who currently provide rideshare services. Crucially, the accident subtly reshaped the terrain of the politics of autonomous vehicles through diverse materialities. This took place through images; through the words and tenor of eyewitness testimony and newscasters; through the police reports; through the communications between city councils; and through encounters with drivers.

new forms of order may spontaneously emerge’ (1991, p. 124). What I draw from De Landa's writings, then, are the interruptive tendencies immanent to automation where, rather than having a centralised locus of control, machines are better apprehended as ‘eddies and vortices nested inside more eddies and vortices’ (De Landa, 1991, p. 8). In this regard, accidents can be understood as adding points of creative bifurcation into the ongoing evolution of automated technologies, making the evolution of such technologies far from certain. The writings of Stiegler, Malabou, Guattari and DeLanda are each concerned with ontology, in other words, problematizing the nature of what exists, albeit in different ways, and through different methods. What is particularly appealing about their diagnoses is the optimism that comes from the distributed politics that their ontologies emphasise. The anti-foundationalist ontological thinking of these writers opens our eyes to a world that is rich with latent potentials for being otherwise; latent potentials that can potentially be rerouted away from dominant capitalist logics. Indeed, for some of these thinkers, this optimism is driven by existing injustices that they witness in the evolution of these complex systems. Guattari in particular was highly cognisant of the potential problems of new technologies. Like Stiegler, he was concerned about automation's tendencies towards standardisation, pointing out that ‘the more such a network is globalized and digitalized, the more it is standardized and rendered uniform’ (Guattari, 2015, p. 101). He was also worried about very real employment issues, writing that ‘revolutions in informatics, robotics, telematics, and biotechnology will drive the exponential growth of every form of production of material and immaterial goods. But this production will occur without the creation of a new volume of employment’ (2015, 108). Whilst their thinking is therefore inspired by political struggles, the relationship between the political and the ontological requires explication. Joronen and Häkli (2017) warn us about the dangers of being seduced by an ‘onto-theological’ lock in, where a new ontology becomes the pre-defined lens through which politics is understood. The danger is that the analytical outcome just restates what this new ontology posits from the outset. In this regard, my claim here is not that the emphasis on emergence and self-organisation in De Landa and Guattari is a fixed ontological scaffold on which political struggles take place, rather that these ontological accounts of technology are responses to new situations that can help us to probe and question the actually existing ‘compositions, events and political rationalities’ (Joronen & Häkli, 2017, p. 568) that are taking place in processes of contemporary automation. In this regard, Malabou's ontological account of the accident does not fix the ontological conditions for understanding the material politics of the autonomous vehicle accident at Tempe. Rather, in drawing our attention to the interruptive agencies of accidents, her account invites us to question the specific compositions, events and political rationalities of this precise situation. A clear implication of this is that the material politics of automation must be understood in terms of the specificity of geographically-situated processes actually taking place, rather than in terms of generalities. Popular accounts of automation and its impacts therefore tend to ride roughshod over a much more intricately-patterned material world, where transformations might be happening, but according to specific and situated compositions, events and political rationalities. Thus, whilst processes of automation are characteristics of several different fields, from road vehicles to manufacturing, the material politics at play is highly circumstantial. Tracing the diverse sites of the autonomous vehicle accident at Tempe attunes us to the multiple political negotiations taking place over autonomous vehicles in this particular site and through different materialities. Furthermore, far from an ontological view from nowhere, the question of politics also refers to ‘our participation in the process of revealing’, as Joronen and Häkli (2017, p. 573) point out. In this regard, my sensing of this accident through these diverse sites forced me to confront my own tangled sensibilities and dispositions regarding the development of autonomous vehicles. The multiplicity of sites explored through the accident in Tempe

Conclusion In closing, I consider the implications for thinking about the accident in relation to ongoing attempts to understand the evolution of liberal governance. In this regard, whilst there are a multiplicity of material agents acting in this accident to shift things across an ecology of interconnected parts without transcendent control, there are distinctive power differentials involved that alter the potency of the kinds of transformations that the accident can enact. Of particular concern here is how contemporary governance involving new automation technologies is being heavily influenced by the political power of Silicon Valley technology companies. As Srnicek (2016) argues, Uber represents a new form of ‘platform capitalism’ where technology companies are gradually becoming owners of the infrastructures of society through the ‘expansion of extraction, positioning as a gatekeeper, convergence of markets, and enclosure of ecosystems’ (2016, 98). In this regard, whilst the accident does generate latent potentialities that exist in complex systems, through the way that it shifts relations, we also need to be aware of how such potentialities might be closed down by platform capitalist companies. For Uber, this might be discerned through actions such as the corporate restructure in the months following the accident that sought to resuscitate the brand in light of the damage that the succession of incidents described in this paper created, to further bolster its monopolistic ambitions in the rideshare sector (Srnicek, 2016). In this regard we can could see this as an example of ‘neoliberalism as a strategic containment of potentially more radical futures’ (Braun, 2015, p. 1, emphasis added), for instance the suppression of the development of platform cooperatives or publicallyowned platforms. From a more critical perspective, rather than evaluating the accident as an interruption to the monopolistic aspirations of a platform capitalist company, the accident could be understood much less radically as an endemic feature of complex adaptive systems. Recall here Perrow's (1984) idea of ‘normal accidents’ which he argued are an endemic and unpreventable feature of evolving tightly-coupled complex systems, where small failures in a system can unpredictably cascade into problems of much larger magnitudes. What is particularly significant about this observation is how the accident can become a dimension of the operation of contemporary liberal governance. As Dillon and Reid (2000) point out, the targets of liberal governance are no longer rationally-calculated, pre-formed subjects that are deeply informed by power relations. Rather, governance is increasingly premised on modulating complex adaptive systems where non-linear dynamics 64


D. Bissell

are prized and embraced, and where disruptions are an operative dimension of power, rather than a symptom of power's failure. It can be challenging to imagine any possibilities for resistance, both in the face of some alarmist accounts about the rise of ever more insidious forms of control and standardisation through digital automation, and once we have conceded that liberal governance increasingly operates through disruption. However, reflecting on automation and control in the logistical city, Rossiter argues that the city is always ‘caught between expediency and contingency. The machine dream of absolute efficiency runs counter to the unruliness of labour and life … It could never entirely eradicate the constituent power of refusal’ (2014, p. 67). In other words, the unruliness of material contingencies do change politics. Considering the specificity of the accident in Tempe, the power of this particular technology company is not necessarily so assured or so dominant. As Srnicek (2016) reminds us, Uber is reliant on vast amounts of venture capital investment coupled with the withholding of basic worker rights to its drivers, without which it would be economically unsustainable. As a company that is still unprofitable, its entire business model has been a long-term bet on the development of autonomous vehicles; its growth thus premised on anticipations of future profits, rather than actual profits. In that regard, and drawing on the complexity dynamics that Perrow enlists, the transformations that the Tempe accident created have the potential to continue resonating in ways that we cannot currently predict.

Dillon, M., & Reid, J. (2000). Global governance, liberal peace, and complex emergency. Alternatives, 25(1), 117–143. Dodge, M., & Kitchin, R. (2007). The automatic management of drivers and driving spaces. Geoforum, 38(2), 264–275. Ford, M. (2015). The Rise of the Robots: Technology and the threat of mass unemployment. New York: Basic Books. Forlano, L., & Halpern, M. (2015). Reimagining work: Entanglements and frictions around future of work narratives. Fibreculture, 26, 32–59. Fregonese, S. (2017). Affective atmospheres, urban geopolitics and conflict (de) escalation in Beirut. Political Geography, 61, 1–10. Frey, C. B., & Osborne, M. (2015). Technology at Work: The future of innovation and employment. Citi GPS. Genosko, G. (2015). Megamachines: From Mumford to Guattari. Explorations in Media Ecology, 14(1–2), 7–20. Graham, S. (2010). When infrastructures fail. In S. Graham (Ed.). Disrupted Cities: When infrastructure fails. London: Routledge. Gregson, N., Crang, M., & Antonopoulos, C. N. (2017). Holding together logistical worlds: Friction, seams and circulation in the emerging ‘global warehouse’. Environment and Planning D: Society and Space, 35(3), 381–398. Guattari, F. (1996). In G. Genosko (Ed.). The Guattari Reader. Oxford: Blackwell. Guattari, F. (2013). Schizoanalytic Cartographies. London: Continuum. Guattari, F. (2015). Machinic Eros: Writings on Japan. Minneapolis, MN: University of Minnesota Press. Hsu, J. (2017). ‘When it comes to safety, autonomous cars are still teen drivers. Scientific American. 18 January 2017. Available at https://www.scientificamerican.com/ article/when-it-comes-to-safety-autonomous-cars-are-still-teen-drivers1/, Accessed date: 14 September 2017. Irani, L. (2015). Difference and dependence among digital workers: The case of Amazon Mechanical Turk. South Atlantic Quarterly, 114(1), 225–234. Johnson, J. (1988). Mixing humans and nonhumans together: The sociology of a doorcloser. Social Problems, 35(3), 298–310. Joronen, M., & Häkli, J. (2017). Politicizing ontology. Progress in Human Geography, 41(5), 561–579. Kanngieser, A. (2013). Tracking and tracing: Geographies of logistical governance and labouring bodies. Environment and Planning D: Society and Space, 31(4), 594–610. Kitchin, R., Coletta, C., & McArdle, G. (2017). Urban informatics, governmentality and the logics of urban control. SocArXiv, osf.io/preprints/socarxiv/27hz8. Lamarre, T. (2012). Humans and machines. Inflexions, 5, 30–68. Law, J., & Mol, A. (2002). Local entanglements or utopian moves: An inquiry into train accidents. The Sociological Review, 50(1), 82–105. Leszczynski, A. (2016). Speculative futures: Cities, data, and governance beyond smart urbanism. Environment and Planning A, 48(9), 1691–1708. Lipson, H., & Kurman, M. (2016). Driverless: Intelligent cars and the road ahead. Cambridge, MA: MIT Press. Luque-Ayala, A., & Marvin, S. (2016). Developing a critical understanding of smart city urbanism. Urban Studies, 52(12), 2105–2116. Malabou, C. (2012a). Ontology of the accident. Trans. C. ShreadCambridge: Polity. Malabou, C. (2012b). The New Wounded: From neurosis to brain damage. Trans. S. MillerCambridge: Polity. Massey, D. (1984). Spatial Divisions of Labour. Basingstoke: Macmillan. McNeill, D. (2016). Governing a city of unicorns: Technology capital and the urban politics of san Francisco. Urban Geography, 37(4), 494–513. Meehan, K., Shaw, I. G. R., & Marston, S. A. (2013). Political geographies of the object. Political Geography, 33, 1–10. Parikka, J. (2013). Insects and canaries: Medianatures and aesthetics of the invisible. Angelaki, 18(1), 107–119. Peet, R. (1975). Inequality and poverty: A Marxist-geographic theory. Annals of the Association of American Geographers, 65(4), 564–571. Perrow, C. (1984). Normal Accidents: Living with high-risk technologies. New York: Basic Books. Philo, C. (2016). Less-than-human geographies. Political Geography, 60, 256–258. Richardson, L. (2017). Sharing as a postwork style: Digital work and the co-working office. Cambridge Journal of Regions, Economy and Society, 10(2), 297–310. Richardson, L. (2018). Feminist geographies of digital work. Progress in Human Geography, 42(2), 244–263. Rose, M. (2014). Negative governance: Vulnerability, biopolitics and the origins of government. Transactions of the Institute of British Geographers, 39(2), 209–223. Rose, G. (2017). Posthuman agency in the digitally mediated city: Exteriorization, individuation, reinvention. Annals of the American Association of Geographers, 107(4), 779–793. Rossiter, N. (2014). Logistical worlds. Cultural Studies Review, 20(1), 53–76. Shaw, I. G. R. (2017). Policing the future city: Robotic being-in-the-world. Antipode intervention symposiumhttps://antipodefoundation.org/2017/05/19/algorithmicgovernance/. Shaw, I. G. R., & Meehan, K. (2013). Force-full: Power, politics and object-oriented philosophy. Area, 45(2), 216–222. Sheller, M. (2016). Uneven mobility futures: A Foucauldian approach. Mobilities, 11(1), 15–31. Srnicek, N. (2016). Platform Capitalism. Cambridge: Polity. Stacey, J., & Suchman, L. (2012). Animation and automation: The liveliness and labours of bodies and machines. Body & Society, 18(1), 1–46. Stiegler, B. (2010). For a New Critique of Political Economy. Cambridge: Polity. Stiegler, B. (2017). Automatic Society: The future of work. Cambridge: Polity. Thrift, N. (2004). Movement-space: The changing domain of thinking resulting from the development of new kinds of spatial awareness. Economy and Society, 33(4), 582–604. Thrift, N., & French, S. (2002). The automatic production of space. Transactions of the

Acknowledgements I would like to thank Kevin Grove and the three anonymous reviewers for their close engagement with the paper and for their critical feedback and generous suggestions which greatly sharpened the argument. I would also like to thank Lizzie Richardson and Peter Thomas for reading and commenting on an earlier version. I am grateful for the valuable feedback that I received following presentations of the paper at the School of Geography and Development at the University of Arizona, and the Centre for Mobilities Research at Lancaster University. This research was funded by the Australian Research Council (DP160100979 and FT170100059). References Adey, P. (2017). Mobilities. London: Routledge. Amoore, L., & Raley, R. (2017). Securing with algorithms: Knowledge, decision, sovereignty. Security Dialogue, 48(1), 3–10. Anderson, B. (2010). Preemption, precaution, preparedness: Anticipatory action and future geographies. Progress in Human Geography, 34(6), 777–798. Anderson, B. (2014). Encountering Affect: Capacities, apparatuses, conditions. Aldershot: Ashgate. Ash, J. (2015). The Interface Envelope: Gaming, technology, power. New York: Bloomsbury. Ash, J. (2018). Phase Media. London: Bloomsbury. Ash, J., Kitchin, R., & Leszczynski, A. (2018). Digital turn, digital geographies? Progress in Human Geography, 42(1), 25–43. Berlant, L. (2016). The commons: Infrastructures for troubling times. Environment and Planning D: Society and Space, 34(3), 393–419. Bissell, D. (2015). Virtual infrastructures of habit: The changing intensities of habit through gracefulness, restlessness and clumsiness. Cultural Geographies, 22(1), 127–146. Bissell, D., & Del Casino, V. J. (2017). Whither labor geography and the rise of the robots? Social & Cultural Geography, 18(3), 435–442. Braun, B. (2015). New materialisms and neoliberal natures. Antipode, 47(1), 1–14. Bridle, J. (2017). Autonomous Car Trap 001. http://jamesbridle.com/works/autonomoustrap-001, Accessed date: 15 May 2018. Brynjolfsson, E., & McAfee, A. (2014). The Second Machine Age: Work, progress, and prosperity in a time of brilliant technologies. London: WW Norton & Company. Carr, C., & Gibson, C. (2016). Geographies of making: Rethinking materials and skills for volatile futures. Progress in Human Geography, 40(3), 297–315. Crampton, J. (2016). Assemblage of the vertical: Commercial drones and algorithmic life. Geographica Helvetica, 71, 137–146. Cresswell, T. (2006). On the Move: Mobility in the modern western world. London: Routledge. Crogan, P. (1999). The tendency the accident and the untimely: Paul Virilio's engagement with the future. Theory, Culture & Society, 16(5–6), 161–176. De Landa, M. (1991). War in the Age of Intelligent Machines. New York: Zone Books. Dennis, K., & Urry, J. (2009). After the Car. Cambridge: Polity.

65


D. Bissell

Wajcman, J. (2017). Automation: Is it really different this time? British Journal of Sociology, 68(1), 119–127. Wallace, I. (1985). Towards a geography of agribusiness. Progress in Geography, 9(4), 491–514. Wise, J. M. (1998). Intelligent agency. Cultural Studies, 12(3), 410–428. Woodward, K. (2014). Affect, state theory, and the politics of confusion. Political Geography, 41, 21–31.

Institute of British Geographers, 27(3), 309–335. Urry, J. (2008). Governance, flows, and the end of the car system? Global Environmental Change, 18(3), 343–349. Urry, J. (2016). What is the Future? Cambridge: Polity. Verbeek, P. (2008). Morality in design: Design ethics and the morality of technological artifacts. In P. Vermaas, P. Kroes, A. Light, & S. Moore (Eds.). Philosophy and design: From engineering to architecture. Berlin: Springer. Virilio, P. (1989). The museum of accidents. Public, 2, 81–85.

66