WORK/ Algorithmic Co-workers

Kitchin states that there is a ‘pressing need to focus critical and empirical attention on algorithms and the work that they do in the world.’ (2017, 14); the reality that they do work in the world is implied.

Nevertheless, I find it necessary to spend some time on this implication, rather than tucking it away and addressing the work being done from the offset. What does it mean to do ‘work’ in the present day? Understood primarily as a human occupation, the concept of an algorithmic worker can be disputed for its inherent contradiction. Definitions aside though, I turn to Hannah Arendt’s (1998) The Human Condition to begin to tackle exactly this issue; gaining a nuanced understanding of ‘work’, but at the same time opening an argument that was originally intended to describe the ‘human’ (condition) to the nonhuman. Likewise then, I could be criticised for working within the framework of this same inherent contradiction.

With this essay I explore notions of work and begin to formulate ideas for where the calculations of algorithms and artificial intelligence can slot into the equation. Rethinking Arendt’s influential discussion of labour and work in the context of so-called immaterial labour which shapes daily activities in the world today, I consider some of the ways in which nonhuman agents like algorithms can be seen as co-workers or simply as labourers/workers. Bodies, the space between them, communication and self-disclosure, all caught up and embedded in a web of interdependence; an image that blends together some of the topics that I touch on as I work my way through. Understood as machines built to assist or take on intellectual labour, the workplace and/or sites of exchanges of its products are some of the spaces in which people may interface with algorithms. I consider these sites, as well as the process of ‘interfacing’ formative of public discourses on what algorithms like neural networks are, or could be.

Machines as neither tools nor workers

Viewed as a subservient tool, the exclusion of a nonhuman agent, like an algorithm, from doing work can be justified in a manner that is not much worth disputing. Arendt allows however, that not all nonhuman things created by humans are merely self-serving tools. Interestingly, she makes a clear distinction between the tool and machine; ‘even the most refined tool remains a servant, unable to guide or to replace the hand. Even the most primitive machine guides the body’s labour and eventually replaces it altogether.’ (1998: 147) Machines then, can take jobs. From this perspective, the issue with allowing for machines to do the job, is in the notion that the machine replaces the labour, rather than replacing the originally labouring human. This can be read as making the job obsolete; dissolving the labour-ness of the activity, rather than taking over in doing the labour.

Based on Arendt’s tool/machine distinction, the human is reduced to a labouring body. The materiality of the body involved in the labour seems to be the primary qualifier for the nature of the activity. When the body’s labour is replaced, the activity ceases to be the same as before, even if it continues to achieve the same ends through different means. A labourer, in other words, can be seen as such by the materiality of their body.

Working today

Arendt laid out three types of activity that humans engage in, and which therefore condition human life in the world: labour, work and action. While my interest was initially drawn to the domain of work, algorithms can be viewed as participants across all three categories. Work, here, is concerned with the making of durable objects; artifices that can outlast the humans who made them. Labour yields products that are consumed, it is the necessary condition for staying alive. Action is that through which humans distinguish themselves in the world. In Arendt’s terms, labour and work are primarily and directly concerned with the production and creation of material objects. It is only with action that human activities take an immaterial form when, for example, humans speak about past conversations and the material is lost in both the means and content of the exchange.

Since the late 1950s when Arendt’s text was first published, our understanding of modes of work and labour have undergone some noteworthy changes. Whether through (scholarly/dialectic) inquiry or transformations taking place IRL, at the workplace, the concept of ‘immaterial labour’ has shaped what it means to do work. Addressing what would breach both categories – work and labour – without the nuanced distinction posed by Arendt, Lazzarato (2014) explains that the immateriality arose from manual labour becoming more intellectual, and at the same time, a changing relationship between production and consumption. Work, in Lazzarato’s terms is ‘the capacity to activate and manage productive cooperation’ (2014: 79). In this context, communication has become both a form and content of work. With a large part of the world engaging in this so-called immaterial labour on a daily basis, the term is colloquially implied in most work, or, somewhat superfluous. What the discussion around and about immaterial labour does highlight though, is most explicitly, a focus on the intangible, and at the same time the blurring between the means and ends of labour/work.

The collapse of the clearcut distinction between labour and work can be anticipated through Arendt’s text; work and labour converge as the activity of consumption subjects more things in the world. Discussing things created through work, Arendt explains that durability gives things relative independence from ‘men who create and use them’ (1998: 137). With an inflation of the things that fall under our understanding of and availability as consumer products, an object’s independence once granted through durability, is for a large part threatened by its disposability (arguably a form of consumption, but that’s an argument for another discussion). A pair of flimsy shoes, that Arendt argues are independent as they can ‘survive even for a considerable time the changing moods of their owner’ (1998: 138), are today, I would say, readily thrown out, replaced or exchanged by that same owner. Their flimsy durability, in other words, is no guarantee of enduring in a fixed, unencumbered state – of lasting. The shoes’ relative independence from their owner is perhaps better viewed as a mutual interdependence of the agents involved.

Inter-

In an immaterial labour economy, production and consumption is extended to intangible things. As I mentioned above, The Human Condition can be read to reserve these for the domain of action. Today, I would argue, action permeates labour/work, often to the extent of constituting it.

Actions are that through which humans distinguish themselves in the plurality of choice that they each face. According to Arendt (1998), it is through actions that humans ‘appear’, where ‘appearing’ is made explicitly distinct from the fact of a ‘mere bodily existence’; it is rather the outcome of an initiative. Without action, often channelled through speech, a human would be dead to the world.

Communication can be understood as a mechanism of what Arendt refers to as action. She explains that self-disclosure of the agent, implicit in action, happens in exchange with other agents (rather than in their isolation). This agential disclosure is one of the key factors that distinguishes action from for instance, the production of an object – the latter being a means to an end. The content of such exchanges makes up the most literal meaning of the agents’ inter-est; that which is in-between and can relate the interlocutors. Relating here spans from the two or more agents involved in a single exchange, to the broader web of (human) relations in which any exchange is implicated and with which it needs to interface (Arendt, 1998). I write ‘human’ in brackets as this is the way Arendt qualifies this web of relations – as a ‘“web” of human relations’ – but it is also an exclusion I slowly want to move away from. ‘Interfacing’ is the exact skill that Lazzarato (2014) highlights as crucial for the new layer of workers demanded in large restructured companies. Workers are required to ‘interface’ the relationship between labour and control; their labour consists of handling information.

— A screen clipping from a Google Scholar search for ‘employee recruitment with neural networks’. AI researchers are working to find methods to optimise or potentially outsource tasks like recruitment to suitable algorithms. Algorithms are being employed in the management of labour more generally, at a range of levels; management layers are a prime example of what was first labelled as ‘immaterial labour’. —

Understanding interfacing as a form of communication, and communication as a mechanism of action, we can take another step towards understanding action as a contemporary form of labour/work, rather than being a strictly distinct type of activity. Lazzarato succinctly captures this in stating that ‘[t]he process of social communication (and its principal content, the production of subjectivity) becomes here directly productive because in a certain way it “produces” production’ (2014: 86). This proposition once again points to the blurring of means and ends, as communication is constituent of both to a present-day labourer/worker. Inherent to subjectivity, self-disclosure can be seen as a ‘productive’ activity. In other words, action achieves labour/work. This can result from its means, be its ends, or both; put more concretely, in some cases production is a byproduct of action (data being collected online while you browse), in others it may be the primary goal behind action (working as an influencer, getting and maintaining a revenue-generating base of followers).

— A screenshot of a webpage about ‘Carin’, a customer support chatbot. Chatbots like Carin can handle 50-90% of customers’ questions posed to support centres according to Helpy, the customer support software company offering the support bot Carin to organisations; an efficient form of communication. —

Bodyless work(ers)

Whether or not they are geared towards production, action and speech are our means of inserting ourselves into the world; repeating the statement made above, without action a human would be dead to the world. This process of insertion is one in which the (human) body gets left behind. Arendt explains that:

‘In acting and speaking, men show who they are, reveal actively their unique personal identities and thus make their appearance in the human world, while their physical identities appear without any activity of their own in the unique shape of the body and sound of the voice.’ (1998: 179)

If such action is seen as a form of labour or work, the materiality of the body that qualified a labourer to be viewed as one in face of a potential machinic replacement, suddenly seems like an empty qualifier. Referring back to the brief discussion of the distinction between tool and machine that I opened with, it is worth returning to this issue. What is the difference between a labouring human body and a labouring machine, or even body-less algorithmic agent, when there is no bodily labour involved in a job that needs to get done to begin with? If what Arendt termed action is a form of labour today, an agent’s self-disclosure, entirely independent from their body, becomes a tool for doing labour. Removing the condition of materiality in the means and ends from qualifying labour/work as such, allows for other than human material bodies to participate in these activities. Shifting focus away from the human body allows for a machine to do work. Same goes for the algorithm.

— A screenshot of a webpage about ‘Slutbot’, a ‘free virtual coach’. Chatbots can also take on the work of personal trainers. Slutbot, for example, is especially designed to train customers’ sexting skills. You don’t need a body to sext. —

Embedded Bodies

With some grounds for understanding work, I would like to move on to speaking about the work of working algorithms more concretely. Code-constituted, algorithmic agents challenge our ability to – quite literally – put a finger on them. Although, as I have argued, intangibility is perfectly compatible with different dimensions of what makes up work today, the issue becomes more nuanced with hybrid agents like algorithms. Kitchin and Dodge (2011) elaborated the idea of code doing work in the world in their book entitled code/space. Here they make the necessary discursive leap to say that it is not code itself that does work in the world, but rather the code embedded in infrastructures, processes and assemblages. Indeed the ‘bodies’ that do work alongside the human ones today have remarkably distributed forms.

Challenging the discrete unitary image of a worker, or employee, algorithms have been characterised as being embedded in wider socio-technical assemblages. Their implications are far and deep reaching, and near impossible to isolate; it is feels underwhelming to mention that they are for instance viewed as formative of cultural forms (Striphas, 2010 in Gillespie, 2014), crucial to our information ecosystem (Anderson, 2010 in Gillespie, 2014), but also our primary media of expression (Gillespie, 2014). These cited points read as though implicit in the notion of an algorithm in today’s terms, with little need for their explicit qualification. Indeed if we read certain narratives on algorithms, their assemblages reach as far as everywhere – also phrased as: everyware (Kitchin & Dodge, 2011).

Understanding the algorithm as an assemblage once more transforms the body at work. Discussing ways to critically understand algorithms and the work they do in the world, Kitchin (2017) states that our interest is generally in the working of algorithmic systems, rather than in individual algorithms. Similarly, Seaver (2013) notes that with perhaps the sole exception of textbooks, ‘algorithms’ are pretty much always ‘algorithmic systems’. Such systems then, work alongside humans, unchanged individuals. This brings new bodily shapes to the workplace.

Calculating, coding

The interaction between humans and algorithms is characterised in the shape of a recursive loop; a multidirectional entanglement that spirals through ‘calculations’ of the agents involved.’Calculations of the algorithm and “calculations” of the people’ (Gillespie, 2014: 183). This hardly comes as a surprise; co-workers are interdependent. In searching for the words to qualify the relations between these categories of worker, scholars like Gillespie (2014) reveal an underlying understanding of work. Viewing the loop as a shape tangled up in calculations once again confirms the perceived negligibility of a (human) body; this time beyond it being a vessel for calculations. Relying on calculations to qualify their shared activity of work probes the extent of algorithmic systems’ entanglement in it. Work, in this sense, becomes a recursive series of back and forth calculations. A calculating agent is potentially a working one.

At least discursively then, the work of algorithms and humans can become indistinguishable if calculated correctly, where correctness lies somewhere in-between. An algorithm designed for a job may strive to mimic the established human practice, while humans may adopt practices in line with new, algorithmically found optimizations. The issue of workers moulding their own work to fit that of machines or algorithms (Arendt, 1998; Gillespie, 2014) is mirrored in, for example, algorithms optimizing their “calculations” based on user data. Next to the element of mutually adapting the practice of work, is an implicit tendency to narrow the space in-between – to become more alike, or in the aspirations of some, even overlapping.

— Screenshot of one of Christie’s past art auction sales: a print of an image generated by a GAN. Some artists are experimenting with working with algorithmic agents. The practice of print-making transforms significantly when working with a GAN. —

— An extract from Jessica Huhn’s ‘Spot the Bot: Keep Bots From Taking Over On Dating Sites’ blog post on DateAha!, March 2019. Whether in one-off blog posts or in more structurally organised efforts of organisations like AlgorithmWatch, people are working to monitor the work of algorithms and their (in)transparency. —

The same issue of indistinguishability is echoed in the complexity of the code that constitutes algorithmic systems. In what they refer to as ‘codework’, Cox and McLean (2013) deliberately choose to treat code and coding ambiguously on the grounds that the work that code does as well as the labour that goes into coding are difficult to tease apart. Seeing as the labour that goes into coding is often tucked away – ‘outsourced’ – for example into places where labour is cheaper, the element of spatiality is intricately entwined in the complexity of possibly distinguishing. In presenting their notion of algorithms as embedded socio-technical assemblages Kitchin and Dodge (2011) similarly stress the importance of space; being embedded without space would after all be a contradiction.

Managing, governing

As a means of concluding, I would like to return to the motif behind the imperative that I opened with; that there is a pressing need to study the work of algorithms (Kitchin, 2017). Scholars, experts, concerned members of the public alike seem to agree that one of the main claims to importance of attending to the work that algorithms do in the world is that they govern. Or at least, that they have to greater or lesser extents come to constitute governance that concerns the human world; this umbrella term is useful, because it seems to capture the way people understand algorithmic work.

Often one of those trendy terms easily flaunted in public discussions relating to digital technologies, the meaning of ‘algorithmic governance’ can feel somewhat diluted. Scholarly reviews of the term itself attest that it is largely multifaceted, an evaluation closely linked to the notion of ‘everyware’ (Kitchin & Dodge, 2011). Granting that governance itself is a contested term, Katzenbach (2019) argues that algorithmic governance amounts to a broad range of sociotechnical practices. Practices that are constantly changing and contingent. These include things like predictive policing, automated content moderation or management of labour; some, or arguably all, of the examples illustrating working algorithms scattered in the images throughout this text could be seen as examples of these practices. With Arendt’s (1998) domain of action permeating into work/labour it indeed becomes difficult to neatly contain the practice of governance. According to Arendt, action is the condition of all political life, where political life, I would say is closely intertwined with the governance of the concerned lives. Arendt points out the synonymous meaning of “to live” and “to be among men” in the language of the ancient Romans and refers to them as perhaps the most political because of their rhetorical culture. Ancient Romans aside, what I find interesting is that again it is the shared practice of communication that guarantees a form of life among the human world. Nonhuman agents with the capacity to participate and indistinguishably self-disclose in the existing and continuously transforming web(s) of relations are political and governing too. Often subjects of state and corporate secrecy (Seaver, 2013), some algorithms can perhaps be seen to work as present day agents in the dual meaning of the word.