Abstract
In this article, we reassess Marxist notions of labour and value for our datafied societies, where data is allegedly becoming one of the dominant sources of economic value. Our contention is that the existing accounts of value, which assume that value is produced exclusively by human labour, are unable to fully account for the processes of exploitation that take place in our digital platform dominated economy. We begin addressing these shortcomings by critiquing the anthropocentric notion of agency that informs the Marxist account of labour. This notion of agency locates productive activity exclusively in human intentionality. After offering an overview of anthropocentric concepts of labour that still dominate (post-)Marxist theories today, we draw on the work of Gilles Deleuze and Félix Guattari to develop a post-anthropocentric account of agency that we term machinic agency. Machinic agency sees activity as a matter of connectivity between different human and nonhuman actors (technologies, organisms, minerals etc.), which productively combine and amplify their capacities to act. These affective connections precede and shape, but often also completely bypass, human consciousness. We make a case for the concept of machinic agency by comparing it with Actor-Network Theory (ANT), an established theory that conceptualises agency as arising from compositions of both human and nonhuman elements. Our contention is that, unlike ANT, machinic agency is able to collapse both, the distinction between human and nonhuman, and that between mechanism and vitalism. We conclude by suggesting that machinic agency allows us to demonstrate that data capitalism exploits and appropriates not only the surplus value produced by conscious human effort, but also the co-production of affective, technological, and ecological aspects of our existence.
Keywords
Introduction
In recent years it has become commonplace to speak of data as a new source of economic value. With the massification of mobile phones, apps, platforms, online services, data capturing technologies and data processing capabilities, data seems to promise a new age of economic, political and even epistemic transformations. Both advocates and critics alike often deploy extractivist metaphors (‘data as the new oil’; ‘data harvesting’; ‘data mining’; ‘data as raw material’, etc.) to emphasise the key role that data appears to play in contemporary capitalism. Srnicek (2017: 6) contends that the platform ‘has emerged as a new business model, capable of extracting and controlling immense amounts of data’. Similarly, Zuboff (2019: 14) argues that ‘surveillance capitalism unilaterally claims human experience as free raw material for translation into behavioural data’. Gago and Mezzadra (2017: 586) even call for an ‘expanded concept of extractivism’ in order to ‘delineate some of the fundamental features of the logic that characterises the processes of valorisation and accumulation in contemporary capitalism’.
In spite of the key role that data is gaining in both enthusiastic and critical accounts of capitalism, it is important to bear in mind two observations. First, the recent turn to data must be understood within a broader history of capitalism in which knowledge and information become integral elements of the valorisation process. Notions such as ‘the knowledge economy’ (Drucker, 1969), ‘post-industrial society’ (Bell, 1976), ‘valorising information’ (Alquati, 1962), ‘the informatization of production’ (Hardt and Negri, 2000), and ‘immaterial labour’ (Lazzarato, 1996) have all aimed at highlighting the passage from an industrial mode of production grounded on manual labour and heavy machinery (Fordism) towards a postindustrial mode of production grounded on information technologies, knowledge and cognitive labour (post-Fordism). In this sense, the theoretical work of authors such as Alquati (1962), Lazzarato (1996), Morris-Suzuki (1997), Hardt and Negri (2000), Terranova (2000), Virno (2004), and Vercellone (2007) can function as a significant plateau from where to develop a critique of today’s ‘data revolution’.
The second and more pressing observation is that despite the growing awareness of the dominant role that data is playing in contemporary capitalism (and in spite of the long tradition of critical accounts of knowledge, information and cognitive labour from a Marxist perspective), there is little agreement regarding the concrete valorisation mechanisms behind this process. As Fumagalli et al. (2018: 2) put it, from a Marxian perspective there is still no proper agreement on ‘what the origin of value in the platform economy is’. One critical strand claims that the turn to data is radically changing the valorisation process of capital (Beller, 2016a, 2016b; Fumagalli et al., 2018; Gago and Mezzadra, 2017; Pasquinelli, 2017). As a consequence, this strand suggests that the passage from industrial to postindustrial capitalism has turned Marx’s own labour theory of value (LTV) obsolete (Negri, 1996; Virno, 2004; Vercellone, 2007).
A different strand contends that despite all its technical innovations, capitalism is still grounded on the exploitation of human labour. From this perspective, a critique of the data economy must focus not on how data produces value, but on how it intensifies the exploitation of labour as a social relation (Caffentzis, 1997; Crawford, 2021; Fuchs, 2019). Finally, a third strand claims that in light of the recent technical transformations of the mode of production, we would need a new definition of value (Massumi, 2018) or even a new, post-capitalist understanding of contemporary society (Wark, 2019). Despite the novelty of this third strand, most critical discourses on the data economy are still dominated by the first two. While these two first strands may disagree on the validity/obsolescence of Marx’s LTV, we suggest that they both nevertheless inherit a key aspect of Marx’s critique of political economy: its anthropocentrism. This hinders a more radical examination of the role that nonhuman agents play in the production of value under current processes of datafication.
The aim of this paper is to explore how post-anthropocentric accounts of agency can contribute to a novel understanding of the question of value in the current context of data capitalism. Key to this endeavour is Deleuze and Guattari’s (1983, 1987) conception of agency as a machinic process. By redefining the concept of machine and disentangling it from a mechanistic philosophy, Deleuze and Guattari allow for a novel approach to the relation between humans and technology that can offer significant insight into the question of value in contemporary capitalism. On the one hand, Deleuze and Guattari’s concept of machinic agency provides us with an alternative model for understanding the relation between human labour and technology in a world in which the boundary between these two spheres becomes increasingly difficult to sustain. On the other hand, their framework makes it possible to overcome anthropocentric normative frameworks that are still persistent (in the Marxist distinction between living labour and machines, for example) while at the same time safeguarding the possibility of conducting a critical analysis of contemporary power relations under capitalism. To properly develop these arguments, however, it is necessary to first reconstruct the intrinsic relation between human agency and value both in Marx’s philosophy as well as in recent applications of Marxian thought.
The anthropocentrism of labour
We contend that the current process of datafication calls for a post-anthropocentric understanding of value and labour, that is, for a definition of value and labour not grounded on the exceptionalism of the human animal as opposed to nature and machines. This seems to clash with most of the recent applications of Marxist theory to the critique of platform capitalism and data extractivism (Beller, 2018; Fuchs, 2019; Fumagalli et al., 2018; Gago and Mezzadra, 2017; Srnicek, 2017) which reproduce (advertently or inadvertently) Marx’s anthropocentric account of labour and value. The limitations of Marx’s anthropocentrism have been noted by Haraway (2008: 47) in the following terms: “Of all philosophers, Marx understood relational sensuousness, and he thought deeply about the metabolism between human beings and the rest of the world enacted in living labour. As I read him, however, he was finally unable to escape from the humanist teleology of labour—the making of man himself ”.
1
In “a process between man and nature, a process by which man, through his own actions, mediates, regulates and controls the metabolism between himself and nature. He confronts the materials of nature as a force of nature. He sets in motion the natural forces which belong to his own body, his arms, legs, head and hands, in order to appropriate the materials of nature in a form adapted to his own needs.” (1976: 283)
According to Marx, this process is unique to humans and does not include animal activity (Marx, 1976: 283–4). This distinction rests on the belief that animal activity is the result of merely instinctual forces whereas human activity is purposeful: “A spider conducts operations which resemble those of the weaver, and a bee would put many a human architect to shame by the construction of its honeycomb cells. But what distinguishes the worst architect from the best of bees is that the architect builds the cell in his mind before he constructs it in wax. At the end of every labour process, a result emerges which had already been conceived by the worker at the beginning, hence already existed ideally.” (Marx, 1976: 284)
Marx’s definition of labour is hence grounded on an anthropocentric understanding of human exceptionalism (as radically different from nature). As such, it entails a definition of humans that conceives consciousness, conceptual thought and agency as key features that distinguish them from other living beings. These features appear as the ‘directing mechanisms’ behind labour, that is, behind the process through which humans transform nature (Braverman, 1974: 32). In his reading of these passages of
To fully grasp this distinction between human and nonhuman agency, it is important to remember that for Marx (1976: 284) the labour process is composed of three elements: first, ‘purposeful activity, that is, work itself’; second ‘the object on which that work is performed’; and third ‘the instruments of that work’. Machines and tools fall under the third category: they do not labour; they are instruments of labour. The differentiation between these three elements allows Marx (1976: 307–8) to speak of the ‘agency of labour’, further emphasising the difference between human activity and nonhuman objects: “The spinner adds labour-time by spinning, the weaver by weaving, the smith by forging. But although these operations add labour as such, and therefore new values, it is only through the
This means that the distinction between human and nonhuman agency applies not only to animals but also to the artificial products of human labour: tools and machines. Tools and machines, like animals, possess only a nonhuman agency that is neither conscious nor purposeful. Put differently, the anthropocentric definition of labour in
The anthropocentric definition of labour translates in Marx into an anthropocentric definition of value: if only humans labour, then only humans produce value. This establishes an intrinsic connection between human agency and value. Yet, unlike Locke and Hegel, Marx (1976: 149) insists on the fact that value is a ‘social relation’ and not simply the externalisation of purposeful activity. Nevertheless, the fact that only humans enter into social relations just pushes the anthropocentric account of value further back. This applies to both use-value and exchange-value. Marx is quite explicit about this. In relation to use-value, Marx (1976: 289) argues that a machine that is not activated in the labour process is useless: “Living labour must seize on these things, awaken them from the dead, change them from merely possible into real and effective use-values. Bathed in the fire of labour, appropriated as part of its organism, and infused with vital energy for the performance of the functions appropriate to their concept and to their vocation in the process, they are indeed consumed, but to some purpose.”
In relation to exchange-value, in turn, Marx argues in the “only to the extent that it enables the worker to work a larger part of his time for capital, to relate to a larger part of his time as time which does not belong to him, to work longer for another. Through this process, the amount of labour necessary for the production of a given object is indeed reduced to a minimum, but only in order to realise a maximum of labour in the maximum number of such objects” (Marx, 1973: p. 701).
According to Braverman (1974: 35), it is only from the perspective of capital (which conceives labour-power as any other commodity), that animals, humans and machines appear as equal sources of value. From the perspective of capital, both human and nonhuman agencies appear as mere ‘factors of production’ (Braverman, 1974: 35). Hence, the anthropocentric singularity of human labour becomes. “the starting point for [Marx’s] labour theory of value, which bourgeois economists feel they may safely disregard because they are concerned not with social relations but with price relations, not with labour but with production, and not with the human point of view but with the bourgeois point of view” (Braverman, 1974: 35).
Braverman’s point raises important questions: is Marx’s LTV concerned with the ontological distinction between humans, animals and machines? Or is it meant mainly as a political strategy against the discourse of classic political economy (“the bourgeois point of view”) that wants to reduce these three domains to mere ‘factors of production’? And most significantly, how does the transformation of the mode of production alter the balance between the ontological and the political perspectives?
Value, computers and data
During the second half of the 19th century, Marx’s claim that machines create no value (and that they can only function by losing some of the value already contained in them) found a strong ally on ‘the laws of conservation of energy and entropy’ (Caffentzis, 1997: 47). During the 20th century, however, the invention of computers reactivated the debate regarding the agency of machines in the production of value. Given that this new type of machines could automate not only manual labour but also cognitive processes, the classic opposition between human conscious labour and the blind activity carried out by machines became blurry. In recent decades, the rapid progress of artificial intelligence and machine learning has revived this issue (Dyer-Witheford et al., 2019). Some argue that machines will soon achieve human-like capabilities and hence could be considered potential sources of value. Others, on the contrary, insist on the ontological gap between human and nonhuman agency as the key framework through which to examine the technological transformations of contemporary capitalism.
Correspondingly, this section will examine the relation between computers and value from the perspective of two authors: Caffentzis (1997) and Virno (2004). These authors represent two strands of contemporary Marxism and hence two perspectives from where to approach the question of value in the current context of datafication. Whereas Caffentzis assumes the perspective of Marx’s LTV) to argue that computers can create no value, Virno’s treatment of the concept of ‘general intellect’ addresses the valorisation process of capital beyond the LTV. At the same time, however, both Caffentzis and Virno inherit the anthropocentric definition of value and labour examined in the previous section. This, we argue, hinders us from advancing towards a post-anthropocentric critique of datafication.
Not even intelligent machines can create value
According to Caffentzis (1997: 31), economists and analysts fail to explain the relation between automation and capitalism mainly because they ‘assume that technology [has] had a qualitatively new role to play in contemporary capitalism and that machines can create value, hence surplus value and profits’. This mistake, Caffentzis (1997: 31) adds, is not only shared by neoclassical economists but also by all those Marxist authors that assume that the technological transformations of the 20th century have entailed a qualitative transformation of capitalism. 3
In
As mentioned, the development of information technologies has, nevertheless, revived the debate regarding the production of value by machines. Computers have shown ‘that mathematics was no longer the dividing line between mental activity and manual labour’ since they can “replicate the behaviour of any human ‘worker’ who is following (consciously or not) any fixed, finite decision procedure” (Caffentzis, 1997: 51). Just like the steam engine could replace any form of manual work that had already been simplified by the division of labour (Marx, 1955: 112), the computer can potentially replace any form of human activity that can be symbolised in mathematical terms. This brings back once again the question regarding the singularity of (human) ‘living labour’ as opposed to machine ‘work’.
According to Caffentzis, given that computers can in theory simulate every aspect of human labour, the singularity of living labour (and the reason why machines create no value) cannot be found on a positive feature of labour itself. If the source of value were a specific feature of labour (eg cognitive ability, creativity, goal-oriented action, etc.), then the question of value-producing machines would be a matter of technical development (ie of having a machine complex enough to perform that specific feature). Hence, Caffentzis contends that the singularity of living (human) labour cannot lie on a positive feature of labour: ‘if there was, therefore, a positive aspect of labour that created value […] then one concludes that machines also, at least theoretically, can produce value’ (Caffentzis, 1997: 54). This means that labour’s capability to create value has to be found in its negative feature, its capacity to ‘refuse to be labour’ (Caffentzis, 1997: 54). Labour is the ‘outside’ of capital (Marx, 1973: 274): labour is not a commodity and has no value, it is the source of value and the source of all commodity production. Hence, the key aspect that gives human activity its value is not its ‘nonmechanizability’ but rather its ‘self-negating capacity’: ‘as long as it can be refused, as long as the transformation of labour-power into labour is self-reflexively nondeterministic, then it can create value in its actualization’ (Caffentzis, 1997: 55).
If this is correct, then neither the invention of the computer or the development of machine learning algorithms affect the fact that only human labour produces value. As long as these technologies are deployed within the capitalist mode of production, they only ‘give more concrete support for Marx’s claim that the use-value of labour […] is not crucial for analysing the value-creation aspect of human labour’ (Caffentzis, 1997: 52). Hence, ‘instead of forcing a major revision of Marxist theory of value’, information technologies seem to ‘confirm and generalise its essentials’ just like thermodynamic theory did in the 19th century (Caffentzis, 1997: 53).
From the perspective of Caffentzis, datafication can be considered a technical process that transfers knowledge from the sphere of ‘variable capital’ (human labour) to the sphere of ‘constant capital’. From this perspective, data produces no value. Put differently, within a capitalist mode of production, the datafication of society entails no radical transformation of the relation between machines (‘constant capital’) and human labour (‘variable capital’), but only modifies the ratio between them (‘the organic composition of capital’). In this sense, platforms and data are functional to capitalism not because they appear as new sources of value, but because they can intensify the exploitation of human labour by other means. In this sense, Caffentzis would agree with Srnicek’s (2017: 75–88) analysis of ‘lean platforms’: platforms that use data in order to reduce the ratio between variable capital and constant capital to a minimum. Key examples of this are Uber and delivery apps, which use data to invert the flow of the production process in a way that living labour becomes actualised only once the commodity has been purchased and the extraction of surplus value has been secured. And even if ‘lean platforms’ constitute only a ‘minor player’ within the overall process of datafication, they represent a ‘long-term trend’ of the organic composition of capital in post-industrial capitalism (Srnicek, 2017: 87–88).
The cognitive labour of the multitude
Contrary to Caffentzis, Paolo Virno (2004) claims that the technical passage from an industrial to a postindustrial mode of production has entailed a qualitative transformation of capitalism. One of the key aspects of this transformation is that labour time ceases to function as the sole source of value (Virno, 2004: 102). Virno thus displaces the focus of the critique of contemporary capitalism from the exploitation of abstract labour time towards the exploitation of what Marx called the ‘general intellect’, that is, ‘the complex of knowledge which makes up the epicentre of social production and at the same time prearranges its vital confines’ (Virno, 2004: 100). In the “on the relationship between dead labour—that is, labour objectified in machinery and technology—and living labour, creative human activity identified with the collective potentiality of working bodies. He suggests that the
According to Virno’s interpretation, however, the passage from industrial to postindustrial capitalism has entailed that the ‘general intellect’ no longer coincides with the productive forces objectified in machinery, ‘but manifests itself principally as a linguistic reiteration of living labour’ (Virno, 2004: 106). This means that the general intellect that is now exploited by capitalism includes ‘formal and informal knowledge, imagination, ethical propensities, mindsets, and ‘linguistic games’’ (Virno, 2004: 106). Rather than exploiting manual labour in the form of abstract labour time, contemporary capitalism turns ‘thoughts and discourses’ into ‘productive machines’ (Virno, 2004: 106). In this context, the general intellect ‘becomes an attribute of living labour [since] the activity of the latter consists increasingly of linguistic services’ (Virno, 2004: 106). For Virno, this defines the key transformation of labour in cognitive, post-Fordist capitalism. Value is produced by capturing the ‘general intellect’, by appropriating ‘the most generic aptitudes of the mind: the faculty of language, the inclination to learn, memory, the ability to abstract and to correlate, the inclination toward self-reflection’ (Virno, 2004: 108).
Information technologies play a key role for this new form of valorisation. Since the computer is able to automate cognitive operations, it has allowed capital to standardise and capture some of the new attributes demanded from living labour. Already in the 1960s, Alquati (1962) examined how the introduction of information technologies in the productive sphere made it possible for capital to capture the ‘valorising information’ contained in the living labour of the worker. Updating Alquati’s study, Pasquinelli (2015) has argued that the rise of big data and algorithmic technologies is giving passage to a capitalism of ‘metadata’ grounded on the exploitation of a surplus of information. In the current context of platform capitalism and data extractivism, Virno’s thesis allows for an alternative interpretation to that of Caffentzis. Rather than simply intensifying the exploitation of abstract labour time, the extraction and processing of data can be understood as a tool for the capturing of the ‘general intellect’. Data extractivism is the process through which this capturing is achieved while the platform offers the socio-technical infrastructure for data to be captured, stored and processed.
The encounter between Virno’s interpretation of the general intellect and current forms of datafication has been used by Fumagalli et al. (2018) to develop a critique of “the process of creating value is no longer limited to the single working day, but extends to incorporate the whole of human existence, that is, the lifetime that is necessary to generate, again, physical strength, but also affections, relationships, social relations and imaginaries, and therefore social knowledge” (Fumagalli et al., 2018: 13).
At the same time, however, despite all the developments in information technologies, living labour can never become fully subsumed by capitalism (Virno, 1996: 194). According to Tola (2017: 240), this entails a different reading of the general intellect: whereas Marx ‘identified the general intellect with the abstract knowledge subsumed by the machines’, Virno contends that the general intellect ‘cannot ever be fully integrated within machines’ because it is ‘inseparable from the interaction of a plurality of living subjects’. In the context of datafication, this would mean that despite radical technologies being put at the service of the capturing and processing of the cognitive and linguistic capabilities of living labour, there is always a remnant of living labour that cannot be fully captured nor automated since it depends on the existence of what Virno (1996, 2004) calls the ‘multitude’.
It is precisely here that Tola (2017: 240) points out the anthropocentrism of Virno’s interpretation of Marx, that is, the idea ‘that contemporary capitalism produces value by harnessing the ‘biological invariant’ common to human individuals: the potentiality of speech and relationality’ (Tola, 2017: 240). Virno’s concept of the multitude reproduces Marx’s idea according to which only humans ‘act upon the world’, whereas other living beings merely exist (Tola, 2017: 243). As Tola (2017: 244) puts it, Virno’s ‘insistence on the coincidence between human language and labour’ reproduces the anthropocentric definition of labour and value present in Marx’s LTV.
From Caffentzis’ perspective, the current process of datafication can be read as producing no value; it only intensifies capital’s exploitation of human labour. On Virno’s account, on the other hand, datafication appears as a mechanism through which capital captures the value produced by the ‘general intellect’. In spite of their divergent views, both perspectives reproduce Marx’s anthropocentrism, that is, a form of human exceptionalism according to which only human labour can create value. Furthermore, both perspectives establish an intrinsic relation between value and human agency. As we have shown, Caffentzis sees agency as the capacity of human labour-power to refuse to labour, while Virno defines it in relation to the capabilities of cognition and language proper of the living labour of the human ‘species being’. It was mentioned that Marx distinguished between labour, the object of labour, and the means of labour, limiting agency only to the first domain. But what if these accounts of value, labour and agency are insufficient for a critique of the current processes of datafication? What if the relation between capital and data is moving beyond the domain of human agency and hence an adequate critique of datafication calls for a post-anthropocentric notion of agency?
Nonhuman agency
Several theoretical approaches across the humanities and the social sciences have sought to conceptualise agency in terms that do not limit it solely to the domain of the human. Despite significant differences between them, concepts such as ‘agential realism’ (Barad, 2007), ‘material agency’ (Bennet, 2009), ‘deviant agency’ (Alaimo, 2010), ‘distributed agency’ (Rammert, 2008), and ‘environmental agency’ (Hansen, 2009) all challenge anthropocentric accounts of agency that reproduce a belief in human exceptionalism. As Rammert (2008: 63) puts it, these different accounts of agency attempt to undermine a difference in kind between human agency, defined by categories such as intention, reflexivity and reason, the activity of animals, ‘characterised as instinct driven and only tool-using’, and the functioning of machines, which is seen as ‘a repetitive and pre-programed activity’. By destabilising such distinctions, these post-anthropocentric accounts have sought to extend agency to nonhuman and non-animate entities.
A definite map of such accounts of agency is beyond the scope of this paper. However, in order to gloss the advantages of the conception of agency developed by Deleuze and Guattari, we begin with a brief engagement with another key theory that accounts for nonhuman forms of agency: Actor-Network Theory (ANT). We argue that Deleuze and Guattari’s account of machinic agency has two main advantages over this other post-anthropocentric account of nonhuman agency. Firstly, Deleuze and Guattari do not only conceptualise agency as the capacity to act of both human and nonhuman entities, but they also challenge the dualism between mechanism and vitalism. Secondly, Deleuze and Guattari’s account allows us to analyse the relations between human and nonhuman actors without forgetting that these relations are always embedded in social arrangements characterised by contingent asymmetries of power. In doing so, they avoid the risk of losing a critical (and political) perspective (a common critique to ANT).
Actor-Network Theory, undoubtedly one of the most prominent accounts of agency as distributed between human and nonhuman actors, was developed by Bruno Latour, John Law, and others. Actor-Network Theory is a widespread and sometimes contradictory set of approaches that refuse to grant any privilege to the agency of humans, and instead understands action as emerging from a constantly shifting network of relationships. From the perspective of ANT, Law claims that ‘everything in the social and natural worlds [is seen] as a continuously generated effect of the webs of relations within which they are located’ (2009: 141). A node in these webs of relations is what is referred to as an actant or an actor. Latour defines an actor as ‘something that acts or to which activity is granted by others’, and adds that it ‘implies no special motivation of human individual actors, nor of humans in general’ (1996: 373). These networks of actors are, thus, constituted from heterogeneous, human and nonhuman entities (such as animals, machines, ideas, infrastructure, geographical arrangements, etc.), which ANT considers as entities of the same kind. Such webs of actors are formed as a result of their elements being able to form productive relations, and mutually shape and reshape each other. Correspondingly, the aim of ANT is to explain how these networks of actors come together to function as a whole and produce actions.
According to Law (2009: 144), ANT, as a theory, cannot be considered separately from its practical applications, that is, the case studies informed by it. These case studies seek to explain the agency of different constellations, ranging from 15th century Portuguese maritime networks (Law, 1986) to scallop farming in Brittany (Callon, 1986), as a relational effect emerging from a web of actors. Perhaps, the most instructive case study for our purposes is Latour’s (1988) investigation of the career of the highly influential French scientist, Louis Pasteur. Latour sets out to demonstrate that the success of Pasteur, who is credited with a variety of significant inventions, depended upon a whole network of heterogeneous factors. Instead of seeing these inventions and their purchase as a work of his genius, Latour maps relational webs in which formation of domesticated labs and laboratories, professionalisation of vets and medics, bacilli generation, public hygiene movement, and colonial interests intermesh. In this way, Latour is able to show that, from the perspective of ANT, human actions are not guided by their unrestrained intentions and decisions, but are, like everything else, a mere relational effect of a heterogeneous network of actors.
Actor-Network Theory inspired several different accounts of hybrid agency. These conceptions of agency have been frequently motivated by the need to account for the increasingly pervasive agency of advanced technologies. An example of such a concept of hybrid agency is provided by the German sociologist Rammert (2008), who seeks to offer a more nuanced concept of agency. In place of an instrumental account of agency, which posits an interaction between an active human agent and an inert technical object, Rammert (2008: 65) develops ‘a gradual concept of distributed agency’. The dual concept of self-determined human action and mechanistic operation of the machine is in this way substituted with an agential spectrum of human and nonhuman entities, which is developed through different steps.
According to Rammert, the distributed account of agency is necessary as our technological environments, constituted by artificial intelligence, algorithmic operations and smart technologies, exhibit more and more self-activity. While these technologies remain man-made artefacts, he suggests, ‘they lose their passive, blind, and dumb character and gain the capacities to be pro-active, context-sensitive and co-operative’, which is why ‘it is justifiable to define them as agents’ (2008: 67). These agential technologies are, for Rammert, characterised by the ability to register changes in its environment and to communicate this to its heterogeneous components. As a consequence, the activity of these technologies is not pre-fixed, but is capable of adapting itself to the changes in its environment. In opposition to simpler technologies such as a hammer, a record-player or a car, which can be still subsumed under the banner of
Rammert develops his gradual concept of distributed agency in three steps. Firstly, he suggests that the interactive agency of humans and machines does not have a single source, but is instead dispersed among different loci. The action of flying an aeroplane is thus distributed between a pilot, co-pilot, radio coordinator, flight-controller, etc. At the same time, the aeroplane is not a single unit, but a system of communicating parts, all of which are actively involved in the action of flying. Rammert suggests that the second step should involve abolishing the clear distinction between human action and technological operation, which loses its relevance with advanced technologies. Distributed agency for him refers to ‘hybrid constellations made of heterogeneous units of agency’ (2008: 85). The action of a flying plane is thus a result of constant interactivity of a closely knit network between the pilot, the nautical pilot program, automatic navigation system, etc. Thirdly, for Rammert, smart technologies are distributed and integrated in an
Yet, while Rammert shares ANT’s extended approach to agency as being a matter of both human and nonhuman actors, he argues against seeing agency as an undifferentiated or, as he puts it, ‘flattened’ category (2008: 75). Instead, Rammert insists on a gradual, three-level model of agency, which can be understood in terms of causality, contingency, and intentionality (2008: 75). The lowest level of agency, that of causality, concerns any action that produces effects or exerts influence; the contingent agency concerns the ability to act in different ways and to choose between these options; and the intentional agency concerns any reflective goal-oriented action capable of responding to other agents. Rammert suggests that all three levels are applicable to human as well as nonhuman actions, but acknowledges a distinction when it comes to intentional agency. For him, coordinating and communicating actions of technologies can be understood through the lens of the ‘intentional vocabulary […] with similar semantics’ to that of humans (Rammert, 2008: 76). Still, Rammert concedes that the technological action here is only a simulation of human action as it lacks the level of self-reflective understanding, but also the efficacy of bodily co-operation. From a certain angle, hence, it could be argued that despite his extended concept of agency, Rammert’s account still contains traces of human exceptionalism that characterise the anthropocentric perspective. It is important to mention, however, that Rammert is aware of this potential point of criticism, and emphasises that he is assuming the point of view of ‘pragmatism’ (2008: 76). From this perspective, he argues, ‘the vocabulary of intentionality is used for the control or interpretation of activities of people as well as of technical objects’ (2008: 76). Regardless of whether Rammer’s account of agency still contains traces of anthropocentrism or not, our contention is that Deleuze and Guattari’s account offers other analytical advantages. As noted, their concept of agency collapses not only the distinction between human and nonhuman, but also that between mechanism and vitalism. Furthermore, unlike ANT-based approaches, it gives us conceptual tools to examine the asymmetries of (capitalist) power relations, and thus account for the political aspect of socio-technical networks. These two advantages are developed in the following section.
Machinic agency
The intermeshing of human and nonhuman actors, which are mutually affecting and being affected by each other, is central to the idea of agency conceptualised by Deleuze and Guattari. Their idea of agency is linked to the concept of machine, which they develop by challenging not only the distinction between human and nonhuman entities, but also the distinction between life and matter, the organic and the inorganic. These distinctions underlie the fundamental opposition between philosophies that are grounded in vitalism and those grounded in mechanism. To highlight this distinctive aspect of Deleuze and Guattari’s concept of agency, we begin by unpacking this opposition, and explaining their way of overcoming it.
Beyond the mechanism and vitalism divide
While the opposition between vitalism and mechanism has a long and complex history, Scott Lash suggests that the ‘primary distinction between [them] may be in terms of vitalism’s
‘Everywhere it is machines’ declare Deleuze and Guattari in
By conceptualising entities in terms of their machinic operations, Deleuze and Guattari are able to challenge the opposition between mechanism and vitalism. Colebrook suggests that mechanism, which sees the existence in terms of encounters between passive matter devoid of an inner organising principle, ‘is able to explain how systems, bodies, wholes or structures are able to maintain themselves, but […] cannot explain how such wholes come into being’ (2009: 11). Mechanism is thus, in theory, capable of accounting for the workings of a car, or the functioning of a biological organism, by describing the mechanical interaction between their parts, but is incapable of explaining their constitution and change. Vitalism, conversely, is capable of accounting for their formation by positing an external organising principle which animates matter. In the case of a car, this formative agency is provided by the human beings that built it, while an organ such as an eye or brain are said to evolve teleologically, ‘because life is oriented towards the production of an organism that would be capable of apprehending or giving sense to the world’ (Colebrook, 2009: 17).
Deleuze and Guattari’s concept of machine allows us to understand matter as creative without ascribing to it an organising principle that directs it toward a predetermined aim. The machinic operations are ‘directed
A tree is produced as a relatively stable entity only as a result of its continual operations involving flows of water, soil nutrients, carbon dioxide and light (Bryant, 2009: 38). An eye, conversely, becomes an organ of seeing only when it enters in relation with light (Colebrook, 2009: 12). The invention of a lightbulb, to take up McLuhan’s famous example, ‘creates an environment by its mere presence’ (1994: 8). As it allows for extending the workday into the night and generates spaces that would otherwise be uninhabited (eg an evening sports match), the diffusion of a lightbulb modifies human (and nonhuman) activity in a profound way. Similarly, Jane Bennet demonstrates how the collapse of an electrical power grid reveals the complex variety of machinic connections between electromagnetic fields, water, legislation, lifestyles, economic theory etc., whose operations enable, or are enabled by its functioning (2009: 24). 5
Machinic agency, as the capacity to act and produce, emerges through dynamic encounters of machines. For Deleuze and Guattari, human agency is no exception to this. With regard to the latter, they situate desiring production within the domain of the unconscious, which they see not as a representational theatre, but as a factory (1983: 24). This machinic unconscious has no predetermined aims or objectives (like mommy or daddy), but only seeks its own proliferation. Conscious self, which we experience as the domain of our intentions and decisions, and thus the site of our agency, is generated only as a result of the machinic connections that the productive unconscious forms with its environment. Deleuze and Guattari, therefore, argue that we experience our decisions and intentions as free and autonomous as we are conscious of the effects of these machinic relations, but oblivious to the actual relations that cause these effects. Our illusory sense of freedom is, then, grounded in the fact that we are aware of what we are attracted to or repelled by, but completely oblivious to what caused these inclinations. As such, we mistake the awareness of our desires for their origin. 6
From machinic agency to machinic surplus value
Deleuze and Guatarri’s account of machinic agency grounds their concept of ‘machinic surplus value’, which they develop in
In this way, machinic surplus value is capable of accounting for value in terms that go beyond the anthropocentric definition of value grounded exclusively in human labour and human time. By collapsing the distinction between mechanist technical machines (dead labour) and vitalist machines (living labour), the concept of machinic surplus value opens up the space to consider the production of value in a more comprehensive sense. Unlike the marxist perspective, which reduces natural resources to raw materials, and technical machines to means of production, machinic surplus value focuses on the processes of connectivity and amplification between the domains of humans, technical machinery and nature. The agency of these connective processes are here not seen as a recent development resulting from the complexification of modern technologies, but rather as the agential force that has always already grounded the entanglement between human and nonhuman, living and nonliving elements.
In opposition to the marxist concept of surplus value, which appears as the capturing of human labour, machinic surplus value thus concerns the amplification of transversal productive flows between machines that exist on the same ontological level (Pasquinelli, 2015: 58). To fully understand the consequences of this claim, however, Deleuze and Guattari’s (1983: 232) distinction between a ‘surplus value of code’ and a ‘surplus value of flow’ would need to be further addressed. Nevertheless, we believe that the concept of machinic agency offers a significant starting point for redefining value in the current context of datafication. It allows reading data extractivism and the platform economy from a systemic perspective that pays attention to the relations between different flows (of energy, information, people, money, commodities, etc.) rather than restricting itself to already individuated and fixed elements (human labour on the one side, means of production and raw materials on the other). Platforms do not only intensify the exploitation of human labour, but they also create new ensembles between human, natural and technical flows. These ensembles use information machines to connect and amplify these flows, which can have significant consequences not only in anthropocentric terms (eg the exploitation of human labour) but also in post-anthropocentric, systemic and ecological terms.
Conclusion
We are told that data has become a new source of economic value and the platform its new business model. In the Marxist literature, however, there is little agreement on what the consequences of this shift are. As shown, Marx contends that only human actions possess agency and hence that only humans produce value. From this perspective, it can be said that either information technologies produce no value because they cannot refuse to work (Caffentzis), or that they extract value by capturing the ‘general intellect’ produced by the living labour of the multitude (Virno). In both cases, value remains linked to human agency, establishing only an external relation to the means of production (machines and technology). This paper has attempted to deploy an alternative concept of agency in order to show that in every productive process there is an entanglement between human and nonhuman elements, an entanglement that becomes even more visible in the current context of datafication and platform economy. This calls for a revaluation of the concepts of labour and value in post-anthropocentric terms.
Our contention is that these concepts can be most effectively reconsidered through the lens of Deleuze and Guattari and their account of machinic agency. We attempted to show the advantages of their account by discussing it alongside another post-anthropocentric account of agency, that of ANT. It has to be noted that both Latour and Law, two of the main originators of ANT, see this theory as aligned with Deleuze and Guattari’s account of agency. Law (2009: 145) points out that, for Latour, actor-networks could be referred to as ‘actant-rhizomes’, while he himself maintains that there is little difference between the term actor network and Deleuze and Guattari’s concept of assemblage. 7 Both rhizomes/assemblages, and actor-networks, Law suggests, ‘refer to the provisional assembly of productive, heterogeneous and […] quite limited forms of ordering located in no larger overall order’ (2009: 146). As both approaches emphasise the assembled agency of diverse temporarily co-functioning elements, Law and Latour see them as comparable.
Yet, our article insists that employing Deleuze and Guattari’s account of agency provides two main advantages over other post-anthropocentric accounts of agency, including ANT. We maintain that when reconceptualising the notions of value and labour in our datafied societies, these are in fact of central significance. Firstly, we have argued that Deleuze and Guattari’s machinic agency allows us to go beyond seeing the world in terms of the duality of self-organising biological life and passive blocks of matter (as mechanistic and vitalist approaches would have it). By conceptualising matter as creative (i.e., capable of organising itself non-teleologically), machinic agency allows us to see surplus value not simply as the result of productive activity directed by human consciousness, but rather as productive connections emerging through affective encounters between the bodies of humans, technologies and other biological organisms.
This expanded view of surplus value points to the second advantage of a machinic account of agency, namely, a post-anthropocentric perspective that can nevertheless account for the asymmetric power relations in contemporary capitalism. By defining (machinic) surplus value as a process of amplification by connection, Deleuze and Guattari expand the analysis of capitalism beyond the exploitation of human labour. While we acknowledge that still today a large part of this amplification by connection relates to human labourers, it is undeniable that this amplification involves evermore complex networks of nonhuman actors. The asymmetries identified by Marx’s anthropocentric analysis of surplus value thus not only persist in the new context of datafication but are also intensified, as noted for example in Crawford’s (2021) analysis of labour in an Amazon Warehouse. Yet, expanding the analysis of surplus value from human to machinic agency allows us to identify also other exploitative asymmetries beyond the core conflict between living labour and technical machines. The conceptual lenses of machinic agency and machinic surplus value set the ground for examining the processes of valorisation and accumulation in contemporary capitalism without taking human intentional activity as the only measuring criterion. In addition to accounting for the value produced in interaction between humans and technologies, understanding surplus value as amplification through connectivity sheds light also on the environmental and affective forms of exploitation that concern both human and nonhuman actors.
ORCID iD
Jernej Markelj https://orcid.org/0000-0003-0267-6955