Two MIT students found in 2009 that it is possible to predict if a man is gay by looking only at his Facebook friends and their listed orientations. This is intriguing, because it means that one can programmatically learn about a person by looking at the structure of the social graph they inhabit. It also raises the question of what other things people are inadvertently revealing through their behavior online. This type of analysis is possible because people actively shape their social graphs to reflect their personalities.
The “Gaydar”, as the MIT students call their project, is only a specialized instance of the more general machine learning algorithms employed by advertising companies like Google and Facebook to target ads to users. These algorithms must strike a careful balance – if they are too accurate, people can get freaked out. (When making a Google+ profile, a friend of mine joked, “I this it’s cute how Google plays dumb and asks me to fill out my personal information.”) On the other hand, if advertisement is ridiculously irrelevant, it undermines the site’s credibility and hurts advertising revenue. (Another friend keeps getting advertisements for “steamy cowboy encounters” on Facebook, which he is not very interested in.)
(To be fair, the “Gaydar” was tested only on a very small sample, and I’m not sure that they managed to get it published anywhere. Still, it is an interesting concept.)