Antonio A. Casilli
, Maxime Besenval
, Marion Coville
, Clément Le Ludec
, Paola Tubaro
Telecommunication College of the Paris Institute of Technologies (Telecom ParisTech)
School for Advanced Studies in the Social Sciences (EHESS)
National Center of Scientific Research (CNRS)
National Institute for Research in Computer Science and Control (INRIA)
Our paper presents the preliminary outputs of an inquiry into French micro-work and its links with automation and global virtual supply chains outsourcing digital labor. Insofar as the existing literature on these topics mainly studies English-speaking countries and platforms, our research both completes and challenges some of the previous knowledge about the links between precarization of work, global labor arbitrage, and worldwide economic dependencies.
Despite the pervasive ‘robots will steal our jobs’ rhetoric, evidence that Artificial Intelligence (AI) and data- based business models displace employment is still thin. In fact, the development of machine learning solutions is a rather labor-intensive process, though one that draws on forms of human work that depart from salaried employment. This work is often invisible and/or under-remunerated, and typically falls out of the scope of workers’ rights and social protections.
Micro-work is one of the forms of human labor that are instrumental to AI: driverless cars, smart technologies and virtual assistants are predicated on huge training and test datasets for machine-learning. The production and preparation of these data is the domain of specialized platforms (such as Amazon Mechanical Turk, perhaps the most widely known of them) that recruit large crowds of human workers to annotate, label, qualify, refine, or otherwise augment data. Crowds also perform quality checks on algorithmic outputs (for example, rating the accuracy of maps or of search results). Platforms fragment, simplify and standardize tasks, so as to massively allocate them to millions of micro-work providers. These tasks are performed in real time, rarely require qualifications, and attract remunerations that can be as low as one or two cents. Micro-work has been described as the ‘automation last mile’, covering residual tasks of larger data processing operations that human ‘common sense’ solves more cheaply and appropriately than computers. Funders and clients of micro- work platforms include tech giants such as Apple, Google and Microsoft.
Limitations of existing literature
Micro-work is difficult to observe as it escapes standard measurement instruments such as Labor Force Surveys and administrative data on employment and the state of job markets. Yet it is attracting increasing attention from researchers and major international institutions, in efforts to quantify this phenomenon, to perform comparisons across countries, and to develop fairness standards for workers. However, most extant literature entertains confusion by conflating micro-work with platform-based freelancing (“gigs” in highly-qualified areas such as design, programming and writing), and with geographically sticky on-demand services (such as urban transportation and delivery apps). Additionally, there has been a tendency to overemphasize high-profile platforms and to underspecify the processes through which requesters and myriad workers adjust to one another, sometimes through participation in multiple smaller, lesser-known platforms.
Insights from French micro-work
Our proposed contribution aims to fill these gaps, looking at the specific case of France – where the phenomenon of micro-work has remained under-researched so far, partly because it eludes the country’s
traditionally commitment to social welfare and employment protection. We report preliminary results of DiPLab (Digital Platform Labor), a mixed-methods study consisting in:
• systematic mapping of micro-work platforms based and/or operating in France;
• a questionnaire, distributed to users of a popular French micro-work platform;
• in-depth interviews with a diverse sample of micro-workers and a few platform owners and managers.
Our results enable to draw a typology of both platforms and workers. Platforms can be distinguished on the basis of their business model, the way they contractually frame labor, and the payment arrangements they adopt, more than the type of tasks they offer and the clients they serve. Another major distinction is the extent to which they avail themselves of cheaper offshore micro-workers, mostly located in French-speaking former colonies overseas.
In turn, a typology of workers can be drawn based on their degree of economic dependence on micro-work, the extent to which they construe micro-work as a wage-earning activity, the employability skills they develop from it, their capacity to monetize their platform work, and their degree of participation in discussion forums and mutual support websites. We add further complexity to this typology by taking into account the combination of micro-workers’ economic, human, and social capital, that enables to compare them to the general French population.
Conclusions and future perspectives
On this basis, we elaborate on the subtle transformations of work that AI – and more generally, the ongoing process of datafication of the economy – is bringing about: if automation does not entirely crowd out traditional jobs, the replacement of employment by unprotected, unstable and almost-invisible forms of work is an actual risk, increasingly touching on different segments of the population.