The online profile images the researchers used consisted of photos of users’ faces as well as non-face images like cartoons.The images contained behavioral cues such as poses, pets, objects and the presence of other people, which were linked to people’s interests, lifestyles and how they represent themselves on social networks.
Now the computer can apply this algorithm to new photos it hasn’t yet encountered — in effect, researchers could use these algorithms to predict intelligence on the basis of one’s profile photo.
The purpose was to find whether computers can succeed at estimating people’s “real” intelligence and whether computers might help humans avoid inaccurate stereotypes and value judgments on the basis of a profile photo.
IQ tests do not measure “innate intelligence.” Likewise, the fact that algorithms are mathematical in nature does not mean that they are free from bias; the algorithm used in the study relies on IQ scores and looks at data created by human behavior.)To develop the algorithm, the researchers trained the computer to conduct feature analysis with a set of images of users with measured (IQ test) and perceived (rated by humans) intelligence scores.
The computer used complex mathematical models to find correlations between intelligence scores and image features, such as which visual elements tend to show up in photos of people who score as intelligent.
The Cambridge study used the profile images of 1,120 Facebook users.