Conservative women are more attractive than left-wing females, according to a European study of thousands of faces.
Danish and Swedish researchers tested a deep-learning artificial intelligence, called a neural network, that can predict a person’s political leanings the majority of the time, based solely on their headshot.
It found that right wing women more were attractive, based on a publicly available scoring system.
The group found no such link in men, but did determine that the left-leaning men showed more neutral, less happy faces, suggesting perhaps better skill at guarding their emotions.
The true purpose of the researchers’ study, however, was to show the alarming accuracy of off-the shelf AI, which can correctly guess a person’s political views based on limited information, like a simple selfie, posted to social media everyday.
‘Our results confirmed the threat to privacy posed by deep learning approaches,’ the researchers wrote in their published findings for the Nature-owned journal Scientific Reports.
The researchers, a trio of psychologists and political scientists from Denmark and Sweden, selected 3,323 publicly submitted headshots of political candidates for analysis in a neural network capable of facial expression coding and classification.
The photos all came from candidates running in the 2017 Danish Municipal election, submitted by the candidates themselves to the Danish Broadcasting Cooperation.
For both men and women, the AI was correct at predicting the political leanings of a person, based on just one photo of their face, 61 percent of the time.
The results were even more accurate for men, 65 percent, before the team stripped their photo dataset of any visual imagery other than the man’s face.
Given the low level of political polarization in Denmark and the lower political stakes in these municipal races, the researchers theorized that these candidates were the most likely to reflect the faces ordinary, everyday, politically partisan individuals.
The researchers did cut 188 individuals out of the original dataset based on their non-European ethnicity, saying that those candidates were more than 2.5-times more likely to represent a left-wing party and would have racially biased the AI.
Face API by Microsoft’s Azure’s Cognitive Services was used by the neural network to asses the emotional state expressed in each candidate’s photo.
Its results determined that 80 percent of the faces displayed a happy expression, while 19 percent read as neutral.
‘Using a pre-developed and readily available network that was trained and validated exclusively on publicly available data,’ they wrote in their conclusion, ‘we were able to predict the ideology of the pictured person roughly 60% of the time.
‘While the Danish team determined that this was partly due to bias in the kind of photos politicians present to the public and partly due to Face API struggling to identify certain other facial expressions.
The deep-learning algorithm did catch one unusual distinction, however: female, left-wing political candidates were slightly more likely to be read by the neural net as having a face expressing contempt, the researchers said.
The team was much more confident in the neural net’s findings linking a high attractiveness rating with conservative views.
‘These results are credible given that previous research using human raters has also highlighted a link between attractiveness and conservatism,’ the authors wrote.
While the deep learning algorithm did find attractiveness to be a predictor of political slant for women, the same could not be said for male candidates.
Nor could macho stereotypes: the neural net failed to detected any noticeable distinctions between right-wing and left-wing men based on its programmed indicators for masculine facial features.
