Employees of third-party companies are engaged in the layout of posts for algorithms.
Facebook Contracting Organizations use people to sort user publications to train neural networks. Employees of companies see even private records, and in some cases receive personal data of users. It is reported by Reuters, citing its own sources in one of the contractors.
Like many technology companies, Facebook uses machine learning to sort content on its platforms. However, in order for algorithms to work correctly, they first need to show examples. Therefore, people are usually engaged in the initial marking.
According to Reuters, one of the companies involved in sorting data is the Indian WiPro. Up to 260 employees are engaged in marking publications, which divide them into five categories, including the type of content (for example, selfie or food photo), occasion (birthday or wedding) and the author’s intention (joke, inspiration, call to organize a party).
Journalists found that the Indian company sorts different content from Facebook and Instagram. This includes staffing up with status updates, videos, photos, links, and stories. Each record is checked by two people for accuracy, and each employee processes up to 700 publications per day.
In a conversation with Reuters, Facebook representatives confirmed that the contractor can indeed access publications that are accessible to a limited circle of friends, and in some cases employees can see the names and data of users. The company said that they cooperate with 200 projects in the field of content markup around the world, and about a thousand people are involved in the work.
According to representatives of the social network, lawyers and privacy experts are checking the results of the work of contractors. In addition, the company noted that they recently introduced an audit system, and use data to recommend content in the marketplace, description of photos and videos for the blind and sorting posts.
As noted in The Verge, data markup is a key condition in the development of algorithms, but in recent years, similar projects have become similar to outsourcing call centers. Most often, contractors are located in countries with cheap labor, for example, in China, hundreds of employees in large offices are engaged in sorting data for electric vehicles.