Russian researchers have shown that artificial intelligence can better infer people’s personalities from ‘selfie’ photos than human reviewers. The accuracy turned out to be more easily recognizable than the other four properties. Personality predictions based on female faces were more reliable than those for male faces.
The technology can be used to find & # 39; best matches & # 39; found in customer service, dating or online classes. The article & # 39; Assessing the five major personality traits using real-life static facial images & # 39; is published in Scientific Reports. The study was conducted by researchers from HSE University and Open University Humanities. The physiognomists of ancient Greece and Cesare Lombroso tried to associate the appearance of the face with the personality, but most of their ideas did not resist the critical eye of modern science. The few established associations of specific facial features, such as the proportion of the face, with personality traits, are quite weak. Studies asking human evaluators to make personality assessments based on photos have produced inconsistent results, suggesting that our assessments are not too reliable to be of practical importance.
However, there are strong theoretical and evolutionary arguments that suggest that certain information about personality traits, particularly those essential for social communication, can be transmitted by the human face. After all, face and behavior are made up of genes and hormones, and the social experiences that result from appearance can influence personality development. However, recent evidence from neuroscience suggests that instead of examining specific facial features, the human brain processes images of faces holistically. Researchers from two Moscow universities, HSE University (Higher School of Economics) and Open University Humanities and Economics, are working with a Russian-British start-up BestFitMe to train a cascade of artificial neural networks to make reliable personality assessments Based on photos of human faces. The performance of the resulting model was higher than in previous studies, using machine learning or human evaluators. Artificial intelligence can be based on & # 39; selfies & # 39; who uploaded volunteers online, above all by making accidental estimates about awareness, neuroticism, extraversion, acceptability and openness. The resulting personality assessments were consistent with different pictures of the same individuals.
The study was conducted on a sample of 12,000 volunteers who completed a self-report questionnaire that measured personality traits based on the “Big Five” model and a total of 31,000 & # 39; selfies & # 39; loaded. Respondents were randomly divided into a training and a test group. A series of neural networks was used to pre-process the images to ensure consistent quality and features and to exclude faces with emotional expressions, as well as images of celebrities and cats. Then, a neural image classification network was trained to analyze each image in 128 immutable traits, followed by a multilayer perceptron that used image invariants to predict personality traits. The average effect size of r = 0.24 indicates that AI can correctly estimate the relative position of two individuals chosen at random in a personality dimension in 58% of cases, as opposed to the 50% expected by chance. In comparison with meta-analytical estimates of correlations between self-reported and personality trait observer classifications, this indicates that an artificial neural network based on static facial images surpasses an average human evaluator who reaches the target personally without prior knowledge. The accuracy turned out to be more easily recognizable than the other four properties. Personality predictions based on female faces were more reliable than those for male faces.
There are a large number of potential applications that need to be explored. The recognition of personality from real photos can complement traditional approaches to personality assessment in situations where high speed and low cost are more important than high accuracy. Artificial intelligence can be used to propose products that best adapt to the client’s personality or to identify possible ‘best matches’. for individuals in dyadic interactions, such as customer service, dating, or online tutoring.