They found that if Friendster had used certain state-of-the-art prediction algorithms, it could have divined sensitive information about non-members, including their sexual orientation. “At the time, it was possible for Friendster to predict the sexual orientation of people who did not have an account on Friendster,”
The problem Garcia identifies lies in something called “shadow profiles,” and as a consequence, we all could be intimately profiled by the Facebooks and Googles and LinkedIns of the world—whether we agree to it or not.
Garcia says this kind of statistical analysis—essentially using machine learning to study the known tastes and relationships of one person’s contacts, and making a guess about who they are likely to be—could be used to build disturbingly detailed profiles of people who do not even use the social network
We learned about shadow profiles last year when security researchers at a company called Packetstorm discovered Facebook was maintaining its own files on users’ contacts. For example, if Facebook found two users were connected to a non-member—say, firstname.lastname@example.org—it would pool other information—different phone numbers, for example—into one master dossier.