We need a political subject capable to think an alternative to digital labor (interview Green European Journal, vol. 17, 2018)

[Update: this interview has been translated in portuguese by Priscila Pedrosa]
An interview with Yours Truly and political activist Lorenzo Marsili, published in vol. 17 (“Work on the Horizon: Tracking Employment’s Transformation in Europe”), pp. 80-88 of the Green European Journal. You can download the entire issue here.

Earn Money Online: The Politics of Microwork and Machines

With hype around automation and robotisation at fever pitch, many argue that we will soon see mass labour disappear altogether. Sociologist Antonio Casilli begs to differ. Work is not disappearing, he argues in this interview with Lorenzo Marsili, but is being transformed by the giants of the digital economy. Understanding how the world of work is changing, and in whose interest, is the key political question of the future.

Lorenzo Marsili: You claim that fears of automation are one of the most recurrent human concerns. Do you think the alarm about “robots taking our jobs” should be toned down?

Antonio Casilli: We are afraid of a ‘great substitution’ of humans by machines. This is quite an old concept, one we can trace back to early industrial capitalism. In the 18th and 19th centuries, thinkers like Thomas Mortimer and David Ricardo asked whether the rise of steam power or mechanised mills implied the “superseding of the human race.” This vision was clearly a dystopian prophecy that was never realised in the form originally predicted.

But when jobs were lost, it was because managers and investors decided to use machines – as they still do – as a political tool to put pressure on workers. Such pressures serve to push down wages and, by extension, to expand the profits made by capital. Machines therefore have a precise ideological alignment that typically benefits the part of society which possesses financial means, at the expense of that which works. As a result, the rhetoric around machines as inevitable and neutral job destroyers has been used for two centuries to squeeze the workforce and silence its demands. The discourse that surrounds automation today, with the accompanying fear of robots, is a reproduction of this same rhetoric.

Let’s take a step back. The ‘gig economy’ has become synonymous with underpaid, precarious employment. You choose to focus on the concept of the ‘microtask’. What does this concept refer to?

Microtasks are fragmented and under-remunerated productive processes. Examples include translating one line of a one-page text, watching 10 seconds of an hour-long surveillance video, and tagging the content of five images. Microworkers are usually paid a few cents per task. These tasks are usually posted on microwork platforms which function as labour markets or job search websites. Microworkers can choose the task they want to perform and are allocated a few minutes to complete it. Microtasks are becoming increasingly important in domains as wide-ranging as marketing, computer vision, and logistics, to name just a few. One of the smallest microtasks is the single click, which can be paid as little as one thousandth of a dollar.

The rhetoric around machines as inevitable and neutral job destroyers has been used for two centuries to squeeze the workforce and silence its demands

Are we talking about a significant new phenomenon or is it more of a niche area?

We are faced with a statistical problem when investigating microwork, one shared with the gig economy and indeed every type of informal, atypical, or undeclared work. Their scale and pervasiveness are difficult to gauge with the usual statistical resources such as large-scale surveys, models like the Labour Force Survey, data from the International Labour Organization, or businesses themselves supplying information voluntarily. As far as microwork alone goes, estimates vary wildly. The most conservative, like those of the World Bank, point to just 40 million microworkers. The most exaggerated, meanwhile, describe 300 million in China alone. Personally, I would estimate that there are around 100 million such workers in the world. But the real question is whether these 100 million are the seeds of a much broader tendency. If microwork indicates a way of working that is becoming the norm, how many workers are transforming into microworkers?

And would you say that all work is starting to resemble microwork?

If we look in detail at the evolution of a few particular professions, we can see that they are becoming fragmented and standardised. Take journalists and graphic designers. Instead of producing a campaign, an investigation, or some other project, like 10 or 20 years ago, they find themselves increasingly tasked with producing a small part of a larger project. They are assigned microtasks, to edit a line or to change the colour in a logo, while the rest is distributed to other people. The future of journalism is not threatened by algorithms that write pieces in place of humans, but by the owners of ‘content mills’ that do not demand entire articles but three lines which are used to optimise algorithms. Because the websites in which these texts appear are found by search engines and not by readers, the texts are tailored with the algorithms in mind. Similar kinds of transformations seem to be taking place across a number of sectors.

One interesting aspect of these microjobs is the symbiosis between automated and manual processes. There are jobs that require ‘teaching’ machines and algorithms to make them more efficient for a given task, such as autonomous driving or image recognition. It seems like Star Trek in reverse, where it is no longer the machines that work for the humans but the humans that work for the machines.

In a certain sense, we are seeing the old idea that computers are there for us to command overturned. What’s happening now is that these objects that are a part of our everyday lives – our smartphones, our cars, our personal computers, and many more objects in our homes – are often used to run the automatic processes we call artificial intelligence. By artificial intelligence we mean processes that take decisions in a more or less automatic manner, and which learn, solve problems, and ultimately make decisions, including purchases, in our place. But the problem is that we have this false idea that artificial intelligence is intelligent from its very inception. On the contrary, artificial intelligence needs to be trained, which is why we use terms like ‘machine learning.’ But who teaches artificial intelligence? If we still think the answer is engineers and data scientists, then we are making a big mistake. What artificial intelligence really requires is a huge quantity of examples, and these come from our own personal data. The problem is that this raw information we produce needs to be refined, cleaned, and corrected.

So this is where microwork comes in?

Yes, who wants to do this degrading, routine work? Many people recruited by microwork platforms come from developing countries where the labour market is so precarious and fragmented that they accept minimal remuneration. In return, they perform tasks that might include, for example, copying down a car license plate to provide data for the algorithm managing motorway speeding tickets, or to recognise 10 images, which might be used to provide data on pattern recognition.

But how does this expansion of microwork relate to the stagnation of labour markets in the more advanced capitalist economies? In the UK, for example, there is almost full employment but jobs are increasingly precarious and wages flat.

There is a longer-term trend here that became marked at the end of the 20th century. It consists in the segmentation of the labour market through a pronounced division between ‘insiders’, those who work in ‘formal’ jobs, and ‘outsiders’, who live on ‘odd jobs.’ The so-called outsiders, who are used to moving from one job to another, are the first candidates on microwork platforms. What’s also happening, however, is that insider jobs are becoming less and less formal. The decline of formal work is the result of a political assault on the rights and numbers of salaried workers with the goal of increasing the profit share relative to the wage share. What we see as a result in Western labour markets is an ongoing movement of people from jobs that were traditionally in the formal sector into informal work. This trend is both a result of the huge wave of layoffs seen in recent years, as well as of the outsourcing of productive processes. Outsourcing sees people leave formal jobs to become informal providers for the same company that previously employed them. These people are sometimes asked to leave companies to create their own small businesses and become subcontractors of their former employer.

So labour is not so much destroyed as transformed. Can this development be explained by today’s new monopoly capitalism, with a few large monopolies each dominating a specific platform service?

I would say that there is a process of concentration of capitalism but I don’t agree completely with the notion of monopoly capitalism. I tend to follow the school of thought presented by Nikos Smyrnaios, a Greek researcher, who wrote a book about oligopolistic capitalism, specifically regarding online and digital platforms. The point of his analysis is that there is no such thing as a monopolistic approach to the digital economy. What actually happens is that, for structural and political reasons, these platforms tend to become big oligopolistic economic agents and tend to create what economists would describe as ‘oligopsonies’, or markets dominated by a few buyers, in this case buyers of labour. Thus a handful of big platforms buys labour from a myriad of providers, as happens on microtask services like Amazon Mechanical Turk. These platforms cannot become actual monopolies because they tend to compete amongst themselves.

Citizens are facing relentless efforts deployed by digital capitalists to fragment, standardise, and ‘taskify’ their activities

One way of describing it today is by using quick acronyms like the GAFAM (Google, Apple, Facebook, Amazon, and Microsoft). There are four or five big actors, big platforms, which despite being known for a specific product – whether it is the Google search engine or the Amazon catalogue – don’t really have a ‘typical’ product either. Instead, they are ready to regularly shift to new products and new models. Look at Google’s parent company, Alphabet: it trades in everything from military robot-dogs to think-tanks to fighting corruption. The only thing that is constant for these platforms across products and services is that they rely heavily on data and automated processes, that which we now call artificial intelligence. To capture the data they need to nourish the artificial intelligence they create and sell, they need people to create and refine this data. And so we are back to our role as digital producers of data.

So you would agree with the late Stephen Hawking: the problem is not the robots, but capitalism or, put differently, whoever controls the algorithmic means of production.

This has always been the main problem. The point today is that the algorithmic means of production have become an excuse for capitalists to take certain decisions that would otherwise cause popular uproar. If I were a CEO of a big platform and I declared that my intention was to “destroy the labour market”, I would of course provoke a serious social backlash. But if I said, “I’m not destroying anything, this is just progress, and you cannot stop it”, nobody would react. Nobody wants to be identified with obscurantism or backwardness, especially on the Western Left, whose entire identity is rooted in historical materialism and social progress. So the cultural discourse of “robots who are definitely going to take our jobs” is designed to relieve industrial and political decision-makers from their responsibilities, and to defuse any criticism, reaction, or resistance.

So we need to push against the portrayal of these transformations as natural or magical events, as opposed to political choices. As you know, in the 1970s there was an early re-reading of Marx’s Fragment on Machines, led by Toni Negri and others, which developed the idea of a ‘cognitariat’ as a new political class that could rise up from new forms of immaterial labour. Where do you think that a political force to contest top-down automation might come from?

My own personal history is rooted in a specific intellectual milieu: Italian post-workerism. Nevertheless, some of its hypotheses need to be critically reappraised. I can think of three in particular. The first one is the Marxist notion of a general intellect. With today’s platforms, we are not facing such a phenomenon. Our use of contemporary digital platforms is extremely fragmented and there is no such thing as progress of the collective intelligence of the entire working class or society. Citizens are facing relentless efforts deployed by digital capitalists to fragment, standardise, and ‘taskify’ their activities and their very existences.

The second point is that the bulk of ‘Italian theory’ is based on the notion of immaterial labour. But if we look at digital platforms, and the way they command labour, we see that there is no such thing as a dematerialisation of tasks. The work of Uber drivers or Deliveroo riders relies on physical, material tasks. Even their data is produced by a very tangible process, resting on a series of clicks that an actual finger has to perform.

And finally, we need to dispute the idea that such a political entity, a class of proletarians whose work depends on their cognitive capacities, actually exists. Even if it did, can we really characterise this political subjectivity as a cognitariat? If you read Richard Barbrook’s 2006 book The Class of the New, you’ll see there’s a long list of candidates for the role of Left-sponsored ‘emerging political subjectivities’, one for each time we experience technological or economic change. Between the ‘lumpenproletariat’, the ‘cognitariat’, the ‘cybertariat’, the ‘virtual class’, and the ‘vectorialist class’, the list could go on forever. But which one of these political and social entities is best suited to defending rights and advancing the conditions of its members? And more importantly, which is able to overcome itself?

What do you mean by overcome itself?

The world doesn’t need a new class that simply establishes digital labour and the gig economy as the only way to be. We need a political subject that is able to think about an alternative.

What do you think should be the role of the state? It seems that the only two national ecosystems trying to govern artificial intelligence are the US and China: Silicon Valley and the state-driven ‘Great Firewall of China’. Where does this leave Europe?

There is a question of what the role of the nation state is in a situation where you have a dozen big players internationally whose power, influence, and economic weight are so vast that in some cases they surpass those of the states themselves. Yet states and platforms are not competitors; they collude. U.S. multinationals are just as state driven as Chinese ones. U.S. government funds and big agency contracts have been keeping Silicon Valley afloat for decades. Moreover, there’s a clear revolving door effect: Silicon Valley CEOs going to work for Washington think tanks or for the Pentagon, like Google’s Eric Schmidt for example.

To be extremely blunt, states should heavily regulate these multinationals, but at the same time they should adopt a policy of extreme laissez faire when it comes to individuals, citizens, and civil society at large. Yet so far exactly the opposite has happened: generally speaking, states are repressing any kind of development or experimentation coming from civil society. They stigmatise independent projects by accusing them of being possible receptacles for terrorists, sexual deviants, and hostiles. Meanwhile, the big platforms are left free to do whatever they want. This situation has to change if we are to have actual political and economic progress.

M, le maudit chatbot (ou, de l’impossibilité de l’automation complète au lendemain de l’échec de l’assistant virtuel de Facebook)

L’année 2018 commence fort chez Zuckerberg. Facebook a décidé de mettre fin à l’expérience de son assistant virtuel, M. Par rapport aux autres chatbots sur le marché, M avait une particularité : il ne cachait pas la présence d’humains derrière les rideaux de l’automation, pour ainsi dire. Facebook allait jusqu’à intégrer cet attribut dans ses arguments de vente. M était présenté comme une intelligence artificielle « mue par des humains » (human powered).

Il s’agissait d’un logiciel qui envoyait des messages via l’application Messenger, initialement disponible pour un nombre limité d’usagers-testeurs (10 000 résidents de San Francisco) qui l’utilisaient surtout pour des opérations commerciales. La plupart des tâches réalisées par le chatbot nécessitaient de personnes pour assister, entraîner, vérifier ses réponses — et parfois se faire carrément passer pour M. D’où la suspicion que son nom ne soit pas un clin d’œil à l’assistant de James Bond, miss Moneypenny, mais une référence assez claire au “micro-travail”…

L’objectif de Facebook avec cet agent conversationnel était de développer une technologie d’intelligence artificielle capable d’automatiser presque n’importe quelle tâche (“Facebook put no bounds on what M could be asked to do“). C’est toujours la vieille promesse du machine learning : tu fais réaliser des tâches par des humains, puis “la machine” apprend et — bam ! tu te débarrasses des humains. Néanmoins, malgré les vastes ressources techniques de Facebook, le taux d’automation de M semble n’avoir jamais dépasse 30%. Presque trois quarts des tâches, donc, étaient effectuées par des petites mains payées à la pièce, entre San Francisco et Katmandou (si le bruit selon lequel elles étaient recrutées via le bureau népalais de la plateforme de micro-travail Cloudfactory était confirmé).

L’histoire de M contient une moralité pour toute entreprise qui développe ou achète des solutions de IA : quand on prend en compte le coût de the human in the loop, l’automation revient très chère. C’est à cause de l’importance de ces coûts (et de la médiocrité des résultats) que Facebook a définitivement décidé de discontinuer M en 2018.

L’autre leçon à retenir ici est que dans la mesure où l’objectif final de M était de préparer le développement de solutions IA qui auraient pu automatiser presque toutes les tâches que les usagers réalisent en ligne, l’automation complète s’avère être un fantasme de silconvallards. A chaque fois qu’on automatise une tâche, les êtres humains avec qui l’IA interagit en redemandent, et de plus en plus complexes. Comme Wired le souligne :

“Another challenge: When M could complete tasks, users asked for progressively harder tasks. A fully automated M would have to do things far beyond the capabilities of existing machine learning technology. Today’s best algorithms are a long way from being able to really understand all the nuances of natural language.”

Il faut alors recommencer et recommencer à concevoir-entraîner-tester-micro-travailler etc. Comme quoi, nous (autant les centaines de millions de tâcherons du clic sur les plateformes de micro-travail que les milliards de micro-tâcherons dissimulés que nous sommes sur les plateformes de médias sociaux) avons devant nous une longue carrière de dresseurs d’IA. Très longue. Interminable même, à en croire certains jeunes experts d’automation. Parmi eux, un nommé Ernest Mandel, lequel affirmait, dans un texte paru en… 1986 :

“Sous le capitalisme, l’automation complète, l’introduction de robots sur grande échelle sont impossibles car elles impliqueraient la disparition de l’économie de marché, de l’argent, du capital et des profits. (…) La variante la plus probable sous le capitalisme, c’est précisément la longue durée de la dépression actuelle, avec seulement le développement d’une automation partielle et d’une robotisation marginale, les deux étant accompagnées par une surcapacité de surproduction sur grande échelle (une surproduction de marchandises), un chômage sur grande échelle, une pression sur grande échelle pour extraire de plus en plus de plus-value d’un nombre de jours de travail et d’ouvriers productifs tendant à stagner et à décliner lentement. Cela équivaudrait à une augmentation de la pression à la surexploitation de la classe ouvrière (en faisant baisser les salaires réels et les prestations de Sécurité sociale), en affaiblissant ou détruisant le mouvement ouvrier organisé et en sapant les libertés démocratiques et les droits de l’homme.”

Micro-lavoratori di tutto il mondo… (Rassegna Sindacale, Italia, 17 sett. 2017)

Riassunto della mia lectio magistralis alle Giornate del Lavoro della CGIL. Qui il video del mio intervento.