(ORDO NEWS) — Scientists have shown that artificial intelligence can determine the character traits of a person from photographs much better than people. An article about this development is published in the scientific publication Scientific Reports.
Since the heyday of ancient Greek civilization, scientists have been trying to find a connection between appearance and personal qualities. But most of the hypotheses they put forward were ultimately refuted. The few established correlations of specific facial features (such as the ratio of the width of the face to the height of a person) with personality traits are rather weak. All studies in which people made judgments about the personality of others based on photographs yielded extremely contradictory results.
Nevertheless, there are strong theoretical and evolutionary arguments that suggest that the human face contains information about personal characteristics, in particular about those qualities that are necessary for social interaction. In the end, both appearance and behavior are determined by genes and hormones.
However, recent data from neurobiology indicate that the human brain processes images of faces in a complex manner, and does not consider individual specific features. Researchers from two Moscow universities, HSE (Higher School of Economics) and the Open University for the Humanities and Economics, together with the Russian-British startup BestFitMe, solved this problem with the help of artificial intelligence. They trained cascading neural networks to develop reliable personal judgments based on photographs of human faces.
The productivity of the model obtained in this work was higher than in all previous studies, in which machine learning or human ratings were used to solve similar problems. Artificial intelligence could make reliable judgments about good faith, neuroticism, extroversion, goodwill and openness based on selfies downloaded online. At the same time, “neural network judgments” were the same for different photographs of the same people.
The study was performed on a sample of 12 thousand volunteers who preliminarily completed a self-report questionnaire created on the basis of a five-factor personality test, including the qualities listed above. Also, these volunteers uploaded a total of 31 thousand images to the project database.
All photographs were randomly divided into training and test samples. For pre-processing of images, a number of neural networks were used, which excluded photos with a vivid expression of emotions, as well as images of celebrities and cats.
AI, developed by Russian scientists, can make the correct assumption about the relative position of two randomly selected individuals on the personality scale in 58% of cases (recall, if randomly guessing, this figure is 50%). This indicates that an artificial neural network based on static images of a face is superior in such an assessment to the average person who assesses the appearance of a stranger.
Identity recognition from real photographs can complement traditional approaches to assessing personality in situations where high speed and low cost of assessment are more important than high accuracy. Artificial intelligence can be used in marketing, to offer products that are best suited to the personality of the client, or to select possible “best matches” among people involved in pairing interactions – for example, in customer service and in online training.
Contact us: [email protected]