260,000 micro-casual workers in France, this is one of the figures produced by the team of researchers at Diplab (Télécom ParisTech and CNRS), published in the report "micro-work in France". This form of work, more and more used with the popularization of artificial intelligence, consists in carrying out simplistic tasks. All for a low fee per piece and without form of conventional contract.
For the sponsors, the micro-work makes it possible to generate the databases that feed the algorithms, or to check the results. Machine learning, intended to automate and rid humans of repetitive tasks, is therefore often built from … a repetitive work done by humans. The report led by Antonio A. Cassily (Télécom ParisTech) and Paola Tubaro (CNRS) is the first of its kind in France, while the International Labor Organization published its report on the subject last summer. It makes a state of this particular population and points to new problems to address. Badly paid, carried out by precarious populations, the micro-work has no legal framework in France.
Precarious humans at the service of machines
The more a machine learning algorithm is fed into data, the more its results will be relevant. While some large groups can generate databases with those they collect through their products or services, they are less easily accessible to smaller structures. The latter therefore rely on micro-workers (or crowdworkers in English) to create these bases.
The tasks ordered for a few hundred euros to specialized platforms do not require any training in the majority of cases, and are very repetitive: sorting data, labeling images, pronouncing words. In other cases, humans intervene to confirm or refute the result produced by the AI. The DiPLab team has identified 23 platforms, including Amazon Mechanical Turk, launched in 2005 by the e-commerce giant, or the French leader Foule Factory (2014) and its subsidiary dedicated to IA Wirk.io.
22% of regular users of the Foule Factory platform surveyed by the DiPLab, have a standard of living below the poverty line (defined as half the median wage). 36% of them were inactive at the time of their interview. But it is not the micro-work that could get them out of this situation: according to a report of 2017 submitted to the Directorate-General for Internal Policies of the European Parliament, French micro-workers are paid only 54.1% minimum wage.
" Like T., between the ages of 50 and 60, many micro-workers live below the poverty line. (…) Listed on three US platforms similar to Crowd Factory (Microworkers, Clickworker and Clixsense), she starts her day at 14h and finishes at 2am, for a supplement of income that varies between 130 and 170 euros per month : "what to buy my cigarettes (75 € per month)" " report the researchers.
Workers invisible and without status
" Almost no platform names its "workers" providers: this would imply the choice of statutes defined by the legislation on paid employment. Instead, micro-workers are called "sellers" (of data), "providers", or even "freelancers" the "authors" of the report, the "contracts" consist mostly in a simple acceptance of the conditions of the site on a web page.The work is thus located in a "gray area " of legality.
Isolated, micro-workers do their work on their spare time, usually from home. While their interactions with other micro-workers are best limited to forums, they are even smaller with sponsors. The report points out that in the majority of cases, workers do not know the ultimate goal of their work. As a result, their work is sometimes refused by the client, in more than half of the cases without justification.
This opacity of the reports serves in some cases to hide the human work under an IA cladding. The tasks performed by micro-workers can sometimes compensate for "smart" technology gaps already on the market. The CEO of a company specializing in accounting software testifies in the report, taking as an example the AIs for making appointments: " It's human-based (done by humans, ed). (…) They (companies, ed) have to create the data of appointments to then make machine learning and hope, one day, that the process is automated. " The fantasy of artificial intelligence happening to humans is still a long way off.
For more information, do not hesitate to consult the full report, available for free in pdf.