In a recent scientific study, researchers in Denmark used deep learning algorithms to predict the political ideology of individuals based on their facial characteristics. Machine learning techniques were applied to photographs of Danish politicians’ faces, determining whether their political leanings were left- or right-wing. The accuracy of these predictions was 61%. It was found that right-wing politicians were more likely to have happy facial expressions, whereas neutral expressions were more common among those on the left. Interestingly, attractive women were more likely to be right-wing, while women showing expressions of contempt were more likely to be left-wing.

The human face, with its complex network of muscles, can express a wide variety of emotions and information. People often infer many aspects about others, such as personality, intelligence, and even political ideology, based on facial characteristics. The study used computational neural networks to learn patterns and relationships within the facial data. These networks, inspired by the structure and function of biological brains, adjust their connections between artificial neurons based on the data they process.

The researchers trained the neural network using publicly available photos of candidates from the 2017 Danish Municipal elections. These images were provided by the candidates themselves for public communication. The initial dataset was 5,230 facial photographs. The researchers excluded photos of candidates representing parties with less-defined ideologies, those that could not be classified as left- or right-wing, and images not suitable for machine processing or not in color.

Furthermore, the researchers removed photos of candidates who appeared to be of non-European ethnic origin and those with beards, citing their potential to interfere with the detection of facial expressions. The final dataset consisted of 4647 photos, with 1442 being female. The researchers also tested the accuracy of the algorithm on an additional sample of Danish parliamentarians, with no exclusions other than dividing the sample by gender.

The study revealed that the neural network was 61% accurate in predicting ideology based on a facial photograph for both males and females. In terms of facial characteristics, attractiveness and masculinity did not correlate with ideology in males. However, more attractive females were more likely to be representatives of right-wing parties. Happy faces, irrespective of gender, were more likely to be right-wing, while neutral expressions were more likely to be left-wing. Contemptuous expressions, although rare, were more common among left-wing female politicians.

This study offers a dystopian possibility. The researchers acknowledge that their results confirmed the threat to privacy posed by deep learning approaches. They were able to predict the ideology of a person in two samples around 60% of the time using publicly available data. It’s a demonstration of how facial features, which were previously considered personal and subjective, can now be quantified and used to predict personal beliefs and ideologies.

However, the study has its limitations. The authors did not provide the percentages of right-wing and left-wing politicians in the sample, which could influence the accuracy of the results. Furthermore, all politicians whose photographs were included in the study were Danish, and it’s possible that the results might not apply to other populations. Nonetheless, this research offers a glimpse into the potential of deep learning and its implications for personal privacy and political dynamics.

The study was published in Scientific Reports.

Photo by Tim Mossholder