Danish study found AI is 61% accurate at predicting person's political ideology based on facial features
Artificial intelligence algorithms can help predict a person’s political ideology based on their facial characteristics, a study conducted in Denmark found.
The tech found right-wing politicians were more likely to have happy facial expressions in photos while people pictured with neutral facial expressions were more likely to identify as left-wing, the study said.
The study, "Using deep learning to predict ideology from facial photographs: expressions, beauty, and extra-facial information," found that AI can predict a person’s political ideology with 61% accuracy when analyzing a photo of a person.
Deep learning, a method in AI where computer scientists teach computers to learn and process information similar to humans, can be used to make predictions about people based on photographs alone, the researchers explained in their paper, which was published in Scientific Reports.
The scientists tried to pin down exactly "what information contributes to the predictive success of these techniques," according to researchers.
Humans are able to read another person’s face and make judgments almost immediately about personality, intelligence and even political ideology. Study author Stig Hebbelstrup Rye Rasmussen of Arhus University and his colleagues explored if computational neural networks – algorithms that mimic the structure and function of human brains – can predict a person’s political ideology based on a single photo alone.
Artificial intelligence algorithms can help predict a person’s political ideology based on their facial characteristics, a study conducted in Denmark found. (Smith Collection/Gado/Getty Images)
The scientists trained the neural network with thousands of photos of politicians from the nation’s 2017 municipal elections, noting the elections were not highly polarized nor competitive, and referred to the politicians as the "last amateurs in politics."
HOW TO REIN IN THE AI THREAT? LET THE LAWYERS LOOSE
They did away with any photos of candidates who were not explicitly left- or right-wing, were not of European ethnic origin or had been photographed with a beard. The photos only depicted the facial features of the candidates, not photos with backgrounds that could alter predictions. The researchers were then left with 4,647 photos of political candidates, 1,442 of which depicted female politicians.
Experts argue the difference between AI investment in China and the U.S. is the fact that the American model is driven by private companies whereas China takes a government approach. (Josep Lago/AFP via Getty Images)
The researchers used facial expression recognition technology from Microsoft to measure the emotional state seen in the photos, as well as other algorithms to determine attractiveness and even masculinity of the candidates. They also used a handful of photos of Danish parliamentarians to test the algorithm’s accuracy.
WHO IS WATCHING YOU? AI CAN STALK UNSUSPECTING VICTIMS WITH 'EASE AND PRECISION': EXPERTS
People vote during the European Parliament elections in Copenhagen, Denmark, May 26, 2019. (Ida Marie Odgaard via Reuters)
In all, the research found that the AI trained on the data could accurately predict ideology to the tune of 61% – showing the algorithms could predict political affiliation better than pure chance.
"Our results confirmed the threat to privacy posed by deep learning approaches," the researchers wrote. "Using a pre-developed and readily available network that was trained and validated exclusively on publicly available data, we were able to predict the ideology of the pictured person roughly 60% of the time in two samples."
The research found female politicians who were more attractive were more likely conservative, while attractiveness and masculinity for men was not tied to political ideology. Faces of both men and women who appeared happier were also more likely to be right-wing, while neutral facial expressions meant the politicians were more likely members of left-wing parties. The study added that though it was more rare, women who showed contempt on their faces were more likely left-leaning.
"We also provide the first demonstration that model-predicted ideology connects to independently classifiable features of the face," the study said. "For females (though not males), high attractiveness scores were found among those the model identified as likely to be conservative. These results are credible given that previous research using human raters has also highlighted a link between attractiveness and conservatism."
