Experiments have confirmed that the face is indeed the mirror of the soul.
We know from our own experience that there are people who can actually read faces. However, physiognomy, a method for determining the internal qualities of a person based on an analysis of his external appearance, is still considered a pseudoscience. Although Charles Darwin, who can in no way be called a frivolous person, wrote: “What is scientific in the so-called science of physiognomy?.. Each individual mainly contracts only certain facial muscles, following his personal inclinations. These muscles can be more developed, and therefore the lines and wrinkles of the face, formed by their usual contraction, can become deeper and more prominent."
A group of specialists, which included scientists from two Russian universities - the Higher School of Economics (HSE) and the Open University of the Humanities and Economics, decided to test how artificial intelligence (AI) algorithms can cope with the task of determining a person's character by their appearance. Scientists have hypothesized that a real photograph contains clues about personality traits that can be deciphered using machine learning.
Theoretical prerequisites for this existed, a number of scientific works have shown that the face can really be a mirror of the soul. For example, the ratio of face width to face height is associated with personality traits such as aggressiveness, risk taking, and dominance. And from an evolutionary point of view, it can be assumed that some facial features began to be perceived as attractive, since they played an important role in sexual selection - they signaled the sociability, benevolence and reliability of the partner.
To test the hypothesis in practice, scientists trained a cascade of artificial neural networks on a large statistical sample. In social networks, 12,447 volunteers were selected, who submitted 31,367 of their photos.
Participants in the experiment were asked to take an online test to determine their personality traits. For each, character traits from the so-called "Big Five" were determined - these are 5 key factors that are used in psychology to describe the personality model:
- belonging to extroverts or introverts
- friendliness, ability to compromise
- conscientiousness (the ability to act consciously)
- emotional stability
- openness to new experience (intelligence)
The data (volunteers, their photographs and questionnaires) were divided into two groups. On a large sample (it accounted for 90 percent of the data), they trained a neural network: algorithms searched for a connection between appearance and personal data. The results were then compared with the control group. If the patterns identified by the AI coincided with the real features of people, they were fixed. Erroneous connections were discarded and such training cycles repeated countless times.
As a result, Artificial Intelligence has learned to correctly determine the personal qualities of a person in 58% of cases. This is not much more accurate than a random probability (it is 50%). But the researchers are confident that such a result proves the correctness of the hypothesis itself. With proper training, the neural network will be able to give much more accurate estimates in the future.
This means that all efforts aimed at creating an export external image, which are now being undertaken by the queens and kings of Instagram, Twitter and other social networks, will go to dust. Artificial intelligence will soon tell users that this “heavenly angel” is actually a capricious and greedy selfish woman. And this "tough macho" is an envious hypocrite. And then, in order to be appreciated by those around you, you will have to squeeze out of yourself a slave, a fool, a liar and a villain on a daily basis using the old method of Anton Chekhov (not to be confused with Anfisa Chekhova).