AI facial recognition technology can now identify a person's political leanings from an expressionless face with surprising accuracy. A Stanford University study found that AI can precisely identify political orientation even from a seemingly expressionless face.
This study has been published in the journal American Psychologist. The research raises serious privacy concerns, especially regarding facial recognition without individual consent. The paper's author, Michal Kosinski, stated that his upbringing behind the Iron Curtain made him acutely aware of the risks of surveillance.
The research team recruited 591 participants and controlled the photography environment to ensure consistency of the photos. They used facial recognition algorithms to process the photos, extract facial feature descriptors, and use them to predict political orientation. The results showed that the algorithm could predict political orientation with a correlation coefficient of 0.22.
The study also found that 1,026 human raters could predict political orientation from neutral facial images with a correlation coefficient of 0.21, comparable to the algorithm's performance.
In another study, researchers had the model identify 3,401 photos of politicians from the US, UK, and Canada. The results indicated that facial recognition models could predict political orientation from politicians' images with a median accuracy correlation coefficient of 0.13.
A Nature study as early as 2021 pointed out that facial recognition technology could correctly predict a person's political orientation 72% of the time. Another study in 2023 showed that deep learning algorithms could predict political leanings from facial recognition with an accuracy of up to 61%.
However, some netizens believe that AI systems claiming to read people's emotions or other characteristics from facial expressions lack scientific basis. Research in this field remains controversial and requires more rigorous scientific validation.