Expect more studies like this over the next few years. As artificial intelligence and Big Data start making more human-like presumptions about our activity, we will become fascinated by testing the “humanity” of machines. Even as one aspect of our culture feels threatened by automation -- its encroachment on feelings of human specialness -- another is endlessly fascinated by exactly how far automation, quantification and data can go in matching human activity. As much as behavioral tracking is registering our every move, we would like to believe that there is some interpretive or algorithmic gap there that prevents these systems from “knowing” us in any profound way.
To wit, a study of 86,220 U.S. respondents between 2007 and 2011 suggested that their activity on Facebook could be used to create a more accurate profile of the user than one made by their own friends and family, and even themselves.
The Stanford and Cambridge researchers used 100-item questionnaires to evaluate the subject for their openness, conscientiousness, extraversion, agreeableness and neuroticism. In addition to giving self-evaluative scores of one to five for themselves on various traits, the subjects also allowed researchers to access their Facebook accounts to see what content they liked. The researcher built a model that associated liking certain kinds of content with specific traits.
The report, provocatively titled “Computer-based personality judgments are more accurate than those made by humans,” explains how it correlated likes to traits this way. “Why are Likes diagnostic of personality? Exploring the Likes most predictive of a given trait shows that they represent activities, attitudes, and references highly aligned with the Big Five theory. For example, participants with high openness to experience tend to like Salvador Dalí, meditation, or TED talks; participants with high extraversion tend to like partying, Snookie (reality show star), or dancing.”
This is probably not too surprising. Facebook can essentially be taken as a massive Myer-Briggs test, yet another personality quantification method that tends to provoke humanists.
Modeling and algorithms abound in this, so areas of interpretation are rife and subject to question. I admit not even being able to follow the authors’ model for how they use “self-other agreement” and “interjudge agreement” to create a “primary criterion for judgmental accuracy.” In essence, they were comparing self-assessment with other human assessment as well as where there was agreement across two friends on a subject.
In the end, the psychologists claim that the way they used Facebook data models the actual personality of the user more accurately than most human evaluations of a personality “and comparable with an average spouse,” they say.
The computer was more accurate than human judges on 12 of 13 criteria. Interestingly, the one where it trailed was perhaps the most important of all -- life satisfaction. Still, it seems that self-assessment correlated best, with computer beating self 4 of 13 times. In areas like assessing substance use, computers were better than you.
The authors of the study rightly warn that being forewarned is being forearmed. Once consumers, policymakers and institutions understand that social media can be scraped and modeled to profile people with such accuracy, the potential for use and abuse become clearer. Online social activities are not benign. They potentially allow companies, governments, or anyone to “know” us in ways we neither anticipate nor approve. The full article for this study is here.