Microworkers: These invisible workers who train artificial intelligence

They don't drive ride-hailing services or deliver meals, but their work has become essential to the functioning of many artificial intelligence systems. Microworkers remotely perform a multitude of digital tasks: identifying objects in photographs, transcribing recordings, verifying information, classifying data, and moderating content. Often carried out from a personal computer, these tasks help improve search engines, voice assistants, translation software, and image recognition systems.

A study conducted in France as part of the DiPLab project estimated the number of highly active micro-workers at nearly 15,000, in addition to more than 50,000 regular users and several hundred thousand people who had already carried out this type of task on an occasional basis. This data gives an idea of ​​the scale of the phenomenon, even if this workforce remains difficult to quantify. A large proportion of workers use these platforms to supplement their income, finance their studies, or cope with periods of unemployment.

An essential but poorly protected workforce

The system relies on platforms that break down large projects into thousands of very short tasks, sometimes paid only a few cents. By accumulating these tasks, companies obtain the annotated data necessary to train their algorithms. Behind the apparent automation of artificial intelligence, therefore, are often people tasked with correcting, sorting, and interpreting information that machines cannot yet process on their own.

This activity, however, remains characterized by irregular income, weak social protection, and significant opacity regarding the final use of the work performed. Certain tasks can also expose workers to violent images or difficult content, particularly in the area of ​​content moderation. As artificial intelligence develops, the question of the status, remuneration, and rights of these invisible workers is becoming an increasingly important social issue.