Who knows you better than anyone? Facebook, of course. In a recent study of 86,000 Facebook users, researchers compared the results of a personality survey completed by study participants’ family and friends to a computer analysis of the participants’ Facebook likes. The computer blew family and friends out of the water.
After 10 likes, the computer could better predict personality than a coworker; after 70 likes, the computer outperformed a friend or a roommate; after 150 likes, the computer was more accurate than a family member; and after 300 likes, even a spouse couldn’t beat Facebook.
The researchers were shocked by the results. “We were thinking, okay, maybe we will be able to report that computers are half as good as human beings,” said Michal Kosinski, a co-lead author on the study. “We were completely surprised to see that computers are already beating us. They are beating us at our own game and they are beating us very decisively.”
The study utilized psychology’s Big Five personality traits -- openness, conscientiousness, extraversion, agreeableness and neuroticism -- then linked certain Facebook preferences to corresponding personality traits. For example, according to the study, liking The Velvet Underground on Facebook is linked to lower levels of conscientiousness, while liking The Guardian is linked to higher levels of conscientiousness.
So I took the assessment myself. (You can have your own Facebook likes analyzed here.) My conscientiousness score was 46 percent -- a slightly below average score, indicating that I am spontaneous and flexible. I also happen to like both The Velvet Underground and The Guardian on Facebook. (In comparison, I scored 73 percent for conscientiousness on the corresponding 100-question self-assessment.) While I’m not sure my results prove that Facebook has cracked my personality code, the idea of being analyzed by my digital footprint is intriguing. What real-life applications could be in store for digital personality predictions?
For starters, transforming therapy. According to Kosinski, cheating is a big problem in psychology. “Psychologists joke that personality questionnaires are really IQ questionnaires,” he says. “If someone is intelligent enough, then they can misrepresent themselves in the personality questionnaire in such a way that they will get a job, or get out of the therapy program if they don't want to be there.”
Cumulative Facebook likes are much more difficult to manipulate than traditional personality tools. To fake one’s digital footprint, a prospective cheater would need to maintain her online behavior for an extended period of time. If a person can fake being consistently well-organized for five or six years, she is probably just a well-organized person, Kosinski explains.
In addition, computer personality scores were better at predicting real-life outcomes -- such as the likelihood that an individual would abuse drugs -- than family or friends. If we treat a pen and paper like a simple computer program, a lot of what therapists do is already governed by a machine, Kosinski says. "They wouldn't try to judge your depression just by looking at you. They would use a depression questionnaire."
Facebook personality analyses could also streamline the relationship between job seekers and recruiters, a notoriously inefficient process. If entry-level job seekers underwent personality assessments and were matched with potential careers that suited their personalities, it could eliminate some of turnover associated with young workers. On the recruitment side, employers could pull from a pool of individuals whose personalities match the position they are hiring for, instead weeding through thousands of resumes and trying analyze a candidate’s personality during an interview.
Perhaps the most obvious area ripe for an overhaul is online dating. Not only can users manipulate the traditional personality questionnaires of dating sites like OKCupid and eHarmony (as Chris McKinlay, a UCLA Ph.D. student, notably did in 2012), but the site’s algorithms don’t necessarily reward honest answers. People have a goal when they come to a dating website, Kosinski says. “I want the hottest person, or the wealthiest person, or the smartest person. And then they will try to fill out the personality questionnaire in such a way as to look like the best candidate. We are getting into this kind of cheating race, where if you don't cheat, no one will actually look at you.”
There may be an even more fundamental problem with matching up romantic profiles based on shared personality traits. “I haven't seen a reliable work showing which personality profiles should actually match with each other,” Kosinski says. There’s no evidence to support the idea that two emotional people, or two stoic people, for example, would make good romantic partners.
Like any new technology, the widespread analysis of Facebook likes could have negative implications, too. While Kosinski cautions against fear mongering, he also acknowledges that hyper-accurate predictions based on publicly available information poses privacy concerns, especially if those algorithms can predict sexual orientation, political or religious views.
“Our democracy, our Western civilization, is based on the idea that you have a right to not share the information about your religion. This is one of the bases of religious freedom," Kosinski says. "But now it just becomes accessible to anyone using your web browsing history or your credit card records. There's definitely an issue there.”
Correction: An earlier version of this post stated that more than 17,000 Facebook users participated in the study. In fact, 86,000 Facebook users participated in the study and 17,000 of those users were judged by one friend for the study.