The End Of Privacy: Facebook ‘Likes’ Reveal Sensitive Personal Data

We all know that, to online marketers, we’re just the sum of our Facebook Wall posts and “Likes” – the ubiquitous, virtual “thumbs up” that we attach to all manner of online ephemera. But all those ironic comments and votes of approval may be revealing a lot more about us than we’re willing to share, according to a new report from researchers at the University of Cambridge and Microsoft Research in the UK.

Facebook Likes As Predictor

In a paper published in the Proceedings of the National Academy of Science (PNAS), the researchers demonstrated that it is possible to use knowledge of an individual user’s “Likes” on Facebook to “automatically and accurately predict a range of highly sensitive personal attributes including:  your age, and gender, you sexual orientation, ethnicity, religious and political views. The list of guessabl`e information goes on to include other less quantifiable characteristics like your personality traits, intelligence, happiness, your preference (or not) for addictive substances and whether your parents split up.

For their study, the researchers surveyed over 58,000 volunteers who provided their Facebook Likes, demographic profiles, and the results of several psychometric tests. The researchers then performed a regression analysis on the Likes and other data to predict details about the individual users. Turns out: the model works quite well. Researchers found they were able to correctly discriminates between homosexual and heterosexual men in 88% of cases, and discern the profiles oof African Americans and Caucasian Americans in 95% of cases. Political affiliation was also reliably predicted: researchers could discern between Democrat and Republican in 85% of cases, the report found.

The report makes for good reading – even if it does tend to reinforce some cultural stereotypes. For example, researchers concluded that “the best predictors of high intelligence include ‘Thunderstorms,’ ‘The Colbert Report,’ ‘Science,’ and ‘Curly Fries,’ whereas low intelligence was indicated by ‘Sephora,’ ‘I Love Being A Mom,’ ‘Harley Davidson,’ and ‘Lady Antebellum.'” Wait – Curly Fries??

Good predictors of male homosexuality included Likes on ‘No H8 Campaign,’ ‘Mac Cosmetics,’ and ‘Wicked The Musical,’ whereas strong predictors of male heterosexuality included ‘Wu-Tang Clan,’ ‘Shaq,’ and ‘Being Confused After Waking Up From Naps.'” And, while many of the connections were self evident (homosexuals supporting a campaign like No H8, which advocates marriage equality), others were ephemeral. “There is,” the researchers concluded, ” no obvious connection between Curly Fries and intelligence.” You can’t make this stuff up.

Funny though it is, the implications of the research are profound. For one: it suggests that many of the privacy protections put in place by networks like Facebook provide little meaningful protection, at least from the prying eyes of advertisers and corporations, who can leverage “big data” analytics to derive all the information you’re trying to keep hidden, anyway.

There are, of course, benefits to be had from the ability to derive knowledge of an individual from their online behavior. In fields such as psychology and medicine, automated assessments based on large samples of online behavior could be “more accurate and less prone to cheating and misrepresentation but may also permit assessment across time to detect trends,” the authors argue. Longitudinal records of online behavior captured on sites like Facebook could “open new doors for research in human psychology,” they say.

However, there are also considerable risks: from super targeted advertisements designed to exploit potential customers’ personality flaws (phobias and anxieties), to official or unofficial discrimination based on inferred attributes such as intelligence, sexual orientation, or political views. And, if users begin to feel as if their online behaviors are being watched, it may damage trust in online services or drive them away from digital technology altogether. Check out the full report here (PDF).