Algorithms can predict a person’s intelligence based on social network photos as accurately as humans can and without faulty stereotyping, says study from University of Cambridge Judge Business School.
Algorithms can assess a person’s intelligence based on their social network photos as accurately as humans can and without faulty stereotyping, finds a study from University of Cambridge Judge Business School being presented to a leading big data conference today (Tuesday 7 February).
Humans inaccurately stereotype people wearing glasses or smiling as being intelligent, the research shows.
The study, based on the Facebook profile photos of 1,122 people who had also completed Intelligence Quotient (IQ) tests, used machine learning to find links between the social network users’ photos and their intelligence, and also asked strangers to estimate each person’s intelligence from their photo.
The findings: machines using factors such as a photo’s colour, composition and texture predict a person’s “measured intelligence” as accurately as (or marginally better than) humans do, while humans use inaccurate cues such as eyeglasses in judging “perceived intelligence”. This has important implications for hiring and other practices in which profile photos are routinely reviewed.
“The fact that machines can accurately judge intelligence poses an obvious privacy risk, as social media profile photos are normally public by default,” says research paper co-author Dr David Stillwell, Deputy Director of the Psychometrics Centre at Cambridge Judge. “The fact that humans use inaccurate stereotypes in perceived intelligence is also significant, as hiring managers often look at social media profile photos before inviting people to interview.”
The paper is co-authored by Xingjie Wei, formerly a researcher at the Psychometrics Centre at the School and now Lecturer in Information Systems at University of Bath School of Management, and Dr David Stillwell, University Lecturer in Big Data Analytics and Quantitative Social Science at Cambridge Judge.
The paper – entitled How smart does your profile image look? Estimating intelligence from social network profile images – is being presented to the 10th ACM Web Science and Data Mining conference this week in Cambridge.
“For social media users, the study presents important practical applications, including creation of an automatic profile photo rating system to determine whether a photo ‘looks’ intelligent – as this would help people manage their profile by determining whether the image is appropriate for a CV, LinkedIn or other professional setting,” says Dr Xingjie Wei. “Put simply, the most intelligent people as captured in our dataset understand that the most effective social media profile photo includes a single person, captured in focus, and with an uncluttered background.”
The study is based on data from the myPersonality database, which now contains more than six million psychometric test results together with more than four million individual Facebook profiles of consenting users.
The study looks at the correlation of certain photo features with both measured and perceived intelligence, examining such factors as the proportions of skin and face area, colour and composition. Social media users who are too close to the camera – so their entire face doesn’t fit the image – are perceived by humans as less intelligent; likewise, those who smile and wear glasses are perceived by strangers as more intelligent, but there is no such correlation with measured intelligence – so smiling and glasses are inaccurate stereotypes.
“Given that the computer’s predictions of a user’s intelligence do not match humans’ perceptions, it begs the question of what features humans use that our machine learning algorithms do not capture,” the study says.
So the researchers looked at the top 50 images perceived as most intelligent, and found such visual cues in the photos as business clothing, books, chess, music and formal dining, while the bottom 50 images for perceived intelligence had such visual cues as colourful hair, offensive hand gestures, tattoos and heavy make-up.
The 1,122 Facebook users were 51 per cent men and 49 per cent women, with an average age of 25.9. A total of 739 independent raters, all strangers to the Facebook users, judged their perceived intelligence, with each photo image judged by at least 24 raters to weed out bias through an average score.
Dr Stillwell says:
We set out to find what sort of user’s photo is perceived as intelligent, and whether that matches actual intelligence, because a mismatch indicates an inaccurate stereotype and a faulty judgment.