A review of: En attendant les robots – by Antonio Casilli
Since Antonio Casilli published his book En attendant les robots in 2019 [Waiting for the Robots – soon translated to English at U. Of Chicago Press], the yearly global corporate investment in Artificial Intelligence doubled—it had nearly tripled between 2019 and 2021—according to the AI Index that Standford University publishes every year. Despite a decrease between 2021 and 2022, the first of the decade, yearly AI-related investment have increased a total thirteenfold since 2013 (Maslej et al., 2023, p.184). In the midst of what deserves to be called a gold rush, Casilli’s critical focus on the production-side of the digital economy is warmly welcome. The analysis of digital labour that his book develops participates to the effort of shifting focus to the numerous and often precarious people who are being put into activity to produce the value that irrigates those profitable industries without reaping any—or hardly any—of this value. This is an important and accessible book, also—I think—beyond an academic readership. It offers a wide-ranging perspective on the political economy of digital technology by taking labour as the common thread that connects activities as diverse as that of producer-consumers of social media, app-based gig workers, microworkers across the Global South, and others.
In what follows, I will first present what ‘digital labour’ refers to in Casilli’s book and note that the concept itself is a bit too loosely defined to live up to its analytical ambitions. Notably, the discussion of the “digital labour” of unpaid social media users strikes me as too quick. Nonetheless, the book’s quality does not crucially suffer from these conceptual loose ends. I then move to discuss the critique of the political economy of digital platforms, which is at the core of Casilli’s Marxist approach. In my view, this critique is fundamentally a critique of the exploitation to which value-producing digital users are subjected, but a significant part of Casilli’s argument is also dedicated to uncovering mainstream discourses on digital technology as ideological, in the sense that they obstruct our capacity to see the exploitation at play between platforms and their users. I conclude with a brief discussion of Casilli’s propositions for an emancipatory digital economy.
As a word of notice: my focus, as a political philosopher, is primarily on the conceptual arguments of the book. While he clearly wrote with a large audience in mind, Casilli works as a sociologist. For specialised opinions on the sociological quality of the book, I would advise to search for other reviews[1].
Three types of digital labour
There is a lot to what Casilli groups under the heading of “digital labour”. Before discussing the concept directly, it seems useful to briefly walk through that variety of activities. About a third of the book is dedicated to developing a classification of digital labour between three distinct types: on-demand labour, microwork, and social networked labour.[2]
Casilli characterises on-demand labour (chapter 3) mainly in reference to the figure of Uber drivers, but it concerns a variety of services that provide logistic, transportation, and care services—such as Lyft, Airbnb, Taskrabbit, Glovo and Deliveroo, to name a few. Of all three types of digital labour, on-demand labour is that which appears most ostensibly as work. This is due to its components of strenuousness; because it isn’t hard to recognise an employment-like dependency between labourers and the platforms that pay them; and because some of the tasks that compose it fit with a traditional idea of what work is—e.g. delivering a meal, getting paid for it. Nonetheless, Casilli argues that on-demand labour should be understood as “mainly a work of data production” (2019, p.18). He shows how the economic model of platforms that rely upon on-demand labour is structured around not just the provision of the services, but furthermore around a multitude of digital actions realised by the users—actions that are typically unpaid. For example: in order to produce the data which are vital to its services, Uber systematically lets some drivers drive around without clients. In this way, it organises and valorises a crucial activity of data production at the expenses of its drivers.
The second type of digital labour, microwork (chapter 4), consists of small repetitive tasks outsourced to crowds of independent providers. These tasks are strictly connected to the practice of human-based computation, which consists in dispatching workers to carry out operations that the machines are incapable of accomplishing on their own. Microworking tasks can consist of annotating videos, sorting out tweets, transcribing scanned documents, correcting values in a database, etc. Microwork is important to ‘automate’ several functions, from accounting, to sales, to manufacturing, and the training of Artificial Intelligence. It is typically paid per task, for a few cents, and its laborious dimension tends to be obfuscated by mechanisms of gamification, but unpaid forms of microwork are also common. This workforce is principally concentrated in the Global South, but not only.
Finally, social networked labour (chapter 5) is the type of digital labour in which the worked dimension of the activity is less ostensible. It designates the activities that are indispensable to run social media and generic user-based platforms, such as Google, Facebook, Instagram, or YouTube. Beyond the specialised work of e.g. computer scientists, a considerable amount of the activity of these companies is carried on by unpaid users, as well as by precarious microworkers, who generate content and data. To underline the ‘laborious’ dimension of using social media, Casilli points that the very same activity carried on for free by a social media user—e.g. sharing and annotating content—might be paid for to a microworker who would be, for instance, charged with producing ‘likes’ on a specific content to boost its visibility.
The concept of digital labour
With his concept of digital labour, Casilli ambitions to capture a wide variety of phenomena under one banner. But how precisely does he go about bringing together these disparate elements, and furthermore argue that they should be sorted under a theory of labour? This is not an easy task and, in my view, one that Casilli does not fully convince he has achieved. Indeed, there is a problematic broadness to the concept of ‘digital labour’. This has been noted elsewhere. According to Gandini, the term “has acquired some kind of genericity, becoming a sort of umbrella term that is increasingly delinked from its origins as a critical Marxist stance on labour and value”, hence becoming an “empty signifier” (Gandini, 2021, p.370). With his return to these Marxist origins, Casilli overcomes at least one half of that critique and provides important insights into his target phenomena (see next section). However, we may still regret too loose of a conceptualisation for it to live up to the ambitions of such a cutting-edge concept.
So, what is “digital labour”? At the outset of the book, Casilli defines it as a “taskified and datafied work that is used to train automatic systems” (2019, p.15). Throughout the text, he enriches that basic definition with observations, writing for instance that we should understand digital labour as a “continuum between unpaid activities, underpaid activities, and activities paid in a flexible way” (p.59). Chapter 7 is where the central arguments for understanding digital labour as work are fleshed out. There, he argues that the unpaid nature of certain forms of digital labour should not lead us to misrecognise their laborious dimension. He defends that all forms of digital labour are work by arguing that they are “value-producing”, that they take place within a “contractual frame” (Casilli interprets agreements to terms and conditions as near-labour contracts—p.321), that they occur “under surveillance” and within social relations of “subordination” (he interprets the constraints embedded within technical architectures as markers of subordination—p.328). But while that chapter is a rich critical take on the political economy of digital activities, the scope of its conceptual ambition leaves some loose ends and raises questions. Notably, the precise way in which we should understand recreational social media users to be “subordinated” to platforms owners, and whether such subordination really is analogous to employment, as Casilli suggests (p.325), would deserve more discussion. I was left to wonder whether the mere capacity to issue instructions de-correlated from a “social and economic dependency of an employee towards their employer”—which is Casilli’s general notion of subordination in this discussion (p.310)—really was significant enough to justify an expansion of the field of labour relations to the whole of social media. This is a question about the analytical boundaries of the concept of digital labour. Such questions are important, however, because they determine the scope of Casilli’s critique of exploitation, which, as I turn to argue, is the normative core of his book.
Critical perspective
But whether or not Casilli’s analysis extends as far as he proposes, the book remains an insightful take on the social production of value for the digital economy. It also remains that much of his discussion targets things that are uncontroversially types of labour, and so is directly relevant at least for this bulk of phenomena. For the rest of this review, I will continue to use the phrase “digital labour” in Casilli’s sense.
Casilli’s insights are, at their core, informed by the critical outlook of Marxist theory. It is a critique that can be usefully reported along two axes: on the one hand, a critique of common perspectives on digitalisation as a sort of ideological construct; and on the other hand, a critique of the social arrangements that make up digital labour themselves.
Digital labour as exploitation
Let us begin with the latter, the critique of the social arrangements themselves. Casilli denounces the digitalisation of labour as “the transformation of human productive actions into underpaid or unpaid micro-operations to fuel an informational economy primarily based on data extraction and the delegation of constantly devalued productive tasks to human operators, because they are considered too small, too inconspicuous, too pleasurable, or too unrewarding” (p.15). Throughout the book, Casilli reports a long list of abuses that certain individuals or groups across the broad horizon of digital labour are subjected to—abuses pertaining to being surveiled, precarious, and harmed in various ways. However, what strikes me as the normative foundations of Casilli’s critique of digital labour is the view that digital labourers all have in common to be exploited by the owners of platforms. Indeed, the most basic unifying thread that connects various forms of digital labour, following Casilli, is this: from the Uber driver to the microworker to the Instagram scroller, all are deploying their forces within a process of economic valorisation controlled by another party than themselves, and none of them get much of the value that they produce.[3] All the abuses that may or may not occur in the life of a particular digital labourer are underwritten by their situation of being treated as a cog in a value-producing machinery.
The book’s central contribution to that critique of exploitation is the detailed analysis of how exploitation is supported by the fragmentation of value-production across a multitude of clicking fingers. As chapter 1 and 2 develop, the profitability of the platform model rests on how digital technology enables to taskify production and outsource it across a wide variety of users—paid and unpaid—whose limited purpose and inter-changeability weakens in their antagonistic relation to their ‘employer’. This leads Casilli to take position in the well-known debate about the ‘end of work’: rather than a replacement, digitalisation is a displacement of work to and beyond the fringes of traditional workplaces and work relations (p.48). In articulating the connection between fragmentation and exploitation, Casilli clearly positions himself within the lineage of the operaist theorists, who have been exploring how capital expands its circuits of valorisation through ever-growing areas of social life, blurring the lines between the workplace and its outside. However, he is careful to distant himself from what he sees as a problematic deficit of materialist thought in that tradition and its interest in “immaterial labour” (p.32). By focusing on the material conditions of the various forms of digital labour, e.g. on the various ways in which data-production can be valorised (p.94), Casilli is diving into the concrete mechanisms, places and bodies through which platforms generate capital beyond the workplace.
Automation as ideology
The other major front of Casilli’s critique targets certain discourses on digital technology as a form of ideological construct. Following him, there is a “theoretical imposture” in mainstream takes on automation, which exclusive focus is on the transformative role of potent machines and intelligent algorithms, when, in fact, these machines are mere “puppets manoeuvred through human labour” (p.68). Simply thinking about the kind of discourses on technology, and specifically AI, spread by someone like Elon Musk should suffice to see the target of Casilli’s criticism. The ideological working of those discourses consists in concealing the existence of this human labour and the conditions in which it is deployed, thus supporting the unimpeded continuation of the exploitation of users by platforms. We can also easily add the exploitation of ecosystems, I think, to this remark.
To support his point, Casilli shows that many technological developments that are celebrated as automated solutions have in fact the much more modest function to assist humans in their activities: rather than autonomous cars, we see automatic pilots that assist the human driver; rather than an AI doctor, we see a database that helps decision-making in the medical realm; etc. (p.62). All these forms of “narrow AI” involve lot of digital labour (for their development, maintenance, and operation), a labour which is glossed over by many commentators. However, why should we not expect this situation to change? Why should the robots not end up arriving, after all? For one thing, it may be technically impossible for digital labour to fully disappear in the making of these machines—the book hints in that direction but does not say very much (p.367). Instead, Casilli’s central argument for why the awaited robots won’t arrive is a political-economic one: platforms do not have reasons to automatise labour processes so long as hiring labour is cheaper than developing automates. And because digitalisation is a process that disempowers labourers in their relation to capital, the current trend and foreseeable future is one in which platforms continue to find new ways to obtain labour for little money, or even no money at all. In this context, Casilli warns us that automation is “first and foremost a tool for disciplining labour”, and “this is why it is always postponed, left to the future”[4] (p.71). This disciplining power is of course related to the processes of fragmentation mentioned above, through which the objective position of workers is weakened, but also to the ideological function of the spectre of automation and the gloomy perspectives of an ‘end of work’ with dramatic consequences for working people.
Conclusion
To resist the process of digitalisation as disempowerment that he analyses, and to give a different social fate to digital technology, Casilli emphasises the necessity for the development of a “true class consciousness as value producers” among digital labourers (p.373). To this end, he advocates the “recognition of digital use as labour” (p.32). As I already mentioned, I remain sceptical about the conceptual soundness of that proposition when understood as broadly as Casilli does, as I would also be sceptical about the political success of a strategy that would imply to enrol social media users qua users within something like a labour struggle. But here again, Casilli’s concluding remarks about ‘what to do?’ are pertinent and do not crucially depend on one’s view about whether the whole of the digital activities involved in producing value for platforms should be described in terms of labour.
In these final pages, he discusses several strategies and propositions, while emphasising that each come with their potentials and limitations. Firstly, the broad category of attempts at extending the paradigm of labour relations as “protected subordination” to digital labour, e.g. through employment rights, can be and have been successful, notably for the most ostensible types of digital labour such as on-demand labour. These legal approaches are insufficient, however, both because of their limited capacity to encompass less ostensible forms of digital labour and because they cannot fully address the drastic capacity of platforms to outsource digital labour abroad in a form of social dumping (p.377). Secondly, there have been interesting attempts to develop cooperative platform systems, meant to provide a better allocation of the value produced by its users as well as a democratic structure. However, these remain exceptions within a landscape dominated by capitalistic firms, and this environment leaves them at the perpetual risk of being outcompeted and thwarted (p.381). Finally, Casilli explores two (co-related) ideas for a more fundamental transformation of the digital economy, namely the idea that data should be a form of common good, subjected to a “bundle of rights” across members of society (p.382), and the idea that digital value creation should give rise to a “social digital income” (p.386). The first idea would lay the grounds for individuals to benefit from the socially produced wealth and resources that the digital economy represents, while the second idea aims at a form of distribution of that wealth.
These strategies all have in common to pursue a more just digital economy. But they also all have in common that none of them will find a fertile soil in the sort of technological fetishism that saturate mainstream discourses. For that reason, contributions like Casilli’s are essential in their efforts to bring the production, and not just the consumption, of digital technology at the forefront of our political attention. Whether the concept of ‘digital labour’ is sufficiently robust to support the whole of this endeavour, or whether other normative concepts must be deployed to make sense of the “value-production” of unpaid users, must still be clarified. But, as I tried to highlight here, we don’t need to have a made-up mind on this question for us to benefit from the insights that Casilli provides throughout his book. It remains a mind-opening and, last but not least, very pleasant read that I would recommend to anyone interested in labour, in digital technology, and certainly in any combination thereof.
Bibliography
Casilli, A. 2019. En attendant les robots. Enquête sur le travail du clic. Éditions du Seuil.
Casilli, A. 2021. ‘Waiting for robots: the ever-elusive myth of automation and the global exploitation of digital labor’. Sociologias. 23(57). pp.112-133. URL: https://www.scielo.br/j/soc/a/M3pMfF9nkYXnrgrQwcBCBBr/
Coutant, H. 2020. ‘Antonio A. Casilli, En attendant les robots. Enquête sur le travail du clic’. Sociologie du travail. 62(1-2). URL: https://journals.openedition.org/sdt/30398
Gandini, A. (2021). ‘Digital labour: and empty signifier?’. Media, Culture, & Society. 43(2). Pp.369-380.
Maslej, N. et al. 2023. The AI Index 2023 Annual Report. AI Index Steering Committee, Institute for Human-Centered AI, Stanford University, Stanford, CA.
[1] For example, this one, by Hadrien Coutant: https://journals.openedition.org/sdt/30398
[2] This translation of the book’s terminology is taken from an article in English by Casilli, accessible here: https://www.scielo.br/j/soc/a/M3pMfF9nkYXnrgrQwcBCBBr/ All other translations in this review are my own.
[3] This is not meant to be a crisp definition of exploitation, but only to indicate that Casilli’s analysis is compatible with various understandings of exploitation.
[4] On the idea that firms may not be incentivised to develop the automatic solutions they promise, it is worth reporting the beautiful little opening story of Casilli’s book: Simon, an intern in a start-up that proposes AI-powered services, soon realises after starting his work that the company’s alleged technology does not in fact exist and that its services are carried out manually by microworkers in Madagascar. These workers are literally impersonating the purported-AI that the company sells. The reason for this imposture is not that the technology could not be built, but simply that the company has no economic incentives to develop the AI they pretend to sell. It’s just cheaper to have the job done manually.