While the various tools on Facebook may give you a semblance of privacy, it turns out that nothing is safe from the prying eyes of data labelers, who have access to all the intimate details you share on the social network. Reuters, which broke this news, has detailed how Facebook’s employees, including outsourced ones, go through the personal data of users on a daily basis to annotate or label it.
These staffers categorize items on Facebook and Instagram manually according to five dimensions, such as the subject of the post, the occasion, and the supposed intention of the author. Facebook says that the compiled data is used to create new features, increase user engagement, and of course, increase ad revenue. It is worth mentioning that manual analysis is not explicitly stated in Facebook’s privacy policy.
What’s even more unsettling is that the social media giant is outsourcing this work to other firms, who may or may not have strict privacy policies in place. The Reuters report specifically talks about Wipro, a company in India that has helped Facebook analyze five years worth of data since last year. Initially, as many as 260 employees were working on the project, but with the bulk of the work out of the way, the team has now been reduced to 30 and continues to label posts.
Facebook has confirmed that the Wipro labelers can access nearly everything you share on Facebook, including status updates, photos, and videos. The company also has around 200 similar projects all over the world, including countries such as Romania and the Philippines. The labelers analyze posts in all languages, including Arabic and Hindi. If it’s any solace, employees see no more than 700 items per day on average.
Apart from helping Facebook increase usage and consequently ad revenue, the data labeling process also provides fodder for training AI. This is something that companies like Google and Amazon also do. Since AI powers several different features on Facebook, the practice will most likely continue until the company devises a way to train its software without direct human involvement.
While Facebook could be violating privacy laws with the labeling process, the whole issue is something of a gray area, as the goal of the project is improve the company’s software, which can make it better at things such as cautioning advertisers against sponsoring adult videos or churning out photo descriptions for the blind.
Although the legitimacy of the practice is debatable and some users might not be comfortable with it, no one can opt out of it. Per Facebook, this allows them to gather a sample that’s reflective of the activities on Facebook and Instagram. As for the results of the project, the social media company says that it cannot offer anything right now as the analysis has not concluded yet.
Over the past few years, Facebook has been called out numerous times for its poor data protection practices and it is likely to receive flak for the data labeling projects as well, which had not been reported before.