Videa of fatal accidents, loss of family members, sexual violence: to be watched in a continuous cycle, up to 800 in a day. The objective of the work: to classify images, videos and texts reported by automatic systems as possible violations of the rules of a certain platform. An article from Guardian thus tells the life of a group of women, “ghost workers” of the AI in the Indian state of Jharkhand. It is one story among many that tells what can be hidden behind the “profession” of content moderator and data annotator (data labeler). That is to say, the precarious and underpaid workforce necessary for the development of AI.
Who are data workers: the ghost workers who train AI
These are people all over the world – many of whom earn just a few dollars or cents an hour – who they create training data for machine learning models, often under exploitative and painful conditions. According to the World Bank there are between 150 and 420 million in the world. But these are underestimated data according to Milagros Miceli, a leading expert on the topic, among the 100 most influential people in AI according to Time.
Behind the scenes of AI there are not only high-level trainers but above all a lot of laborers: those who do the dirty work, and suffer the consequences. According to researchers, all data workers face a emotional numbnessOften with delayed psychological repercussionsso much so that content moderation falls into the category of dangerous jobs. Lasting cognitive and emotional stress, increased alertness, intrusive thoughts, anxiety and sleep disturbances: these are some of the side effects.
January 12, 2026, “data labeler” Indu Nadarajan at work at the distribution center of NextWealth, an artificial intelligence services company in the Indian state of Tamil Nadu. As it reaches rural India, AI is quietly reshaping lives, particularly for women from conservative backgrounds. (Photo by Idrees MOHAMMED / AFP via Getty Images)
The invisible slaves who make the internet possible
About 80% of data annotation and content moderation workers come from rural, semi-rural or marginalized contexts. Companies deliberately choose smaller cities and towns, where rents and labor costs are lower, and take advantage of a growing pool of first-generation graduates who are looking for work. Improvements in internet connectivity have made it possible to connect these locations directly to global AI supply chains, without having to relocate workers to cities. This created an army of ghost workers (from the book Ghost work: how to stop Silicon Valley from building a new global underclassby anthropologist Mary Gray). Otherwise called invisible slaves: of the so-called crowdwork. As the title BehanBox (Voices of Sisters in Hindi) «those who make the internet possible».
What makes them invisible is the dispersed nature of their work, opaque AI supply chains, and poor policies. Suffice it to recall the dismissal without notice by Google of more than 200 AI employees: They had developed intelligent chatbots like Gemini, but then complained of low wages and poor working conditions.
Women working for AI
In this population of ghost workers, there are many, many women. Companies welcome them because they are considered reliable and attentive to details. They accept because that for AI it is a job from home or close to home, and rated as “safe”, “clean”, “respectable”. It certainly gives women a rare access to income without migration but, in fact, reinforces its marginal position.
The Data Workers’ Inquiry by Milagros Miceli
He has dedicated his research to the topic of the social consequences of algorithms, and in particular to the working life of artificial intelligence data labelers Milagros Miceli. Among its goals, the Data Workers’ Inquiry: an academic project that empowers AI data workers to publish research of and about themselves. The first group of 16 researchers worked in Kenya, Syria, Brazil and Germany. They earn the same salary – 35 euros an hour – as any other academic researcher at the Weizenbaum Institute in Germany, where Miceli works.
Among the questions they investigated were, for example, the prevalence of gender violence among data workers, including one case where two women were forced to work in the same office as their alleged rapist, and were unable to leave because their visas were tied to their employment status. This is fundamental research with a political impact: because, if on the one hand “they inform us about what is happening”, explains Miceli, “on the other they support them and their organizational efforts: they push them to collectivise and take action”.
