A team of Facebook workers in India has accessed millions of content published by users of the social network since 2014 to tag them manually, a work that raises questions about the treatment of data privacy.
The work of this team of 260 workers, located in Hyderabad (India), was to identify photographs, videos, states, and other content, and classify them in five dimensions, according to Reuters.
Facebook has tagging projects around the world, and in 200 of them, the Indian company Wipro participates. These workers are dedicated to training the algorithms that guide the services of the platform, such as the news bulletin board, known as the news feed.
Workers of the aforementioned company, who have chosen to remain anonymous, have detailed to Reuters that they catalog millions of content published on Facebook according to what they show (food, an animal), the reason (a party, daily activity) and the intention of the author (organize an event, make a party), among other categories.
According to Reuters, Facebook confirmed this labeling work. The manager of product management for AI in Facebook, Nipun Mathur, said that it was an essential part of what the user needed and that there were no reasons for it to disappear.
Doubts about privacy
However, and according to Reuters , this project raises questions about the privacy of the data, since many of those publications, among other things, include comments and name other users, which could clash with the General Regulations for the Protection of Data of the European Union.
The regulation establishes that companies must have the explicit consent of users for the processes of data collection and processing and their use. They must also have mechanisms that eliminate these data after their treatment.
The Indian company Wipro works on the manual labeling of Facebook content since April of last year. A team of 260 people was dedicated in 2018 to analyze the publications of users from all over the world of the last five years in the social network. In December, the team was reduced to 30 people, who went on to tag the publications of the previous month.
Among the contents that they analyze, publicly shared publications are included, but also shared private publications with a limited group of users of the platform.
There are, according to Reuters, labeling equipment also in Romania and the Philippines. Another company, Cognizant Technology Solutions, also with workers in India, is dedicated to the analysis of the language used in videos, to detect sensitive topics or violent language to train an artificial intelligence that prevents advertisers from publishing their videos in harmful content. They also analyze the publications of their Marketplace service.