

Artificial intelligence (AI) is more human than we think. The development of many image recognition, text analysis and sound manipulation systems requires essential "hands-on workers." French sociologists Maxime Cornet, a doctoral student at the Interdisciplinary Institute of Innovation, and Clément Le Ludec (Study and Research Centre for Administrative and Political Sciences, Paris), who defended his doctorate in March, have tried to understand their role. Since 2021, they have interviewed around 20 companies operating in this sector in France. This led them to study seven of their subcontractors in Madagascar, as well as around 200 of their employees. In 2023, they published "The problem with annotation. Human labor and outsourcing between France and Madagascar" in the journal Big Data & Society, alongside Antonio Casilli.
Clément Le Ludec: This technology is used for classification and detection, based on learning principles. Large quantities of so-called training data – such as images, videos and texts – are used to fine-tune them, so that responses can be generalized to apply to new data. Humans are therefore essential for training AIs, either to generate data, for example by filming themselves in front of a camera, or to check that the model's predictions are correct. However, the main task consists of annotating texts or images, in order to build up the learning corpus, for example by indicating on a photo of a junction what the traffic signs are, or identifying traces of rust on photos of electricity poles, or spotting whether a customer is stealing in a store. Even so-called generative AI is involved. ChatGPT required a lot of annotation to teach the program what was an acceptable response and what was not, according to a certain scale of values. In our database of companies turning to these human tasks, a third belongs to the natural language processing sector.
Maxime Cornet: Among these various human tasks, we've even seen a fourth one, and the most "extreme." It involves hiring individuals to replace the software and make the customer believe there's artificial intelligence behind it.
M. C.: Some companies keep these tasks in-house, especially if the data is sensitive. However, many tell us that for this repetitive and tedious work, which can involve viewing several hundred images a day, they can't find anyone in France. Hence the outsourcing we've observed to specialized companies in Madagascar. To our knowledge, no quantitative study exists to estimate the extent of this outsourcing, but in our database of some 20 companies, two-thirds subcontract these data tasks. We also estimate that this represents 5% to 10% of the cost of AI software. The development of artificial intelligence does not mean job losses due to automation, as some have argued, but rather their relocation to developing countries.
You have 63.97% of this article left to read. The rest is for subscribers only.