Artificial intelligence was best at guessing conscientious and conscientious people.
Russian mathematicians and psychologists have developed a neural network that has learned to guess some character traits of people from their photographs better than a person. The press service of the Higher School of Economics writes about this with reference to an article in the scientific journal Scientific Reports.
“The algorithm made a correct conclusion almost 60% of the time, while a random guess usually matches only 50%. The superiority of 10% seems insignificant, but in fact, in terms of the accuracy of predictions, artificial intelligence is significantly ahead of people if they judge by the facial features of a stranger, "the press service writes.
In recent years, scientists have developed many neural networks that can perform non-trivial tasks and even “think” creatively, creating new art and technology. This became possible both thanks to the development of computing systems and the emergence of new mathematical principles that describe the structure and operation of machine learning systems.
For example, recently, mathematicians from the United States have created an AI system that can recognize traces of melanoma, skin cancer, better than the leading expert oncologists. Other neural networks have learned to draw pictures and "paint" videos in the style of Van Gogh or Kandinsky, as well as beat a person over and over in the ancient Chinese game of Go, computer shooters and strategies such as Starcraft and Quake.
Russian mathematicians, programmers and psychologists, under the guidance of Associate Professor of the Higher School of Economics Evgeny Osin, have created a new system of artificial intelligence that surpassed humans in another area - she learned to "guess" the personality traits of other people from their photographs and portraits.
Divination by photography
In general, scientists are rather skeptical about such a possibility. Therefore, physiognomy and similar theories are today considered erroneous or even pseudoscientific.
On the other hand, in recent years, biologists have found weak but statistically significant links between five key personality traits and certain genes. Therefore, scientists continue to investigate whether these genetic differences are reflected in human appearance, including with the help of artificial intelligence systems.
Osin and his colleagues conducted one of the largest studies of this kind. Their experiments were attended by 12 thousand volunteers who agreed to pass one of the classic psychological tests, and also allowed the use of photographs of their faces to train the neural network and use in other experiments.
Having passed some of these images and the results of the questionnaires through the neural network, the scientists checked whether she was able to learn to predict at least some of the key character traits from photographs of other people that the AI had not encountered before. As it turned out, in some cases artificial intelligence was noticeably superior to humans, but at the same time the accuracy of such "fortune-telling from a photo" for different psychological personality traits was very different.
For example, neural networks were the best predictors of conscientiousness and conscientiousness, and worst of all, openness to new experiences. In addition, the scientists noticed that the results of the analysis depended on the image of the person of which gender the neural network was "looking" at. In particular, the algorithm predicted extraversion and emotional stability much better for women than for men.
Scientists hope that subsequent, more advanced versions of such systems can be applied in practice, including in the service sector and in diagnosing the psychological compatibility of employees or visitors of dating sites.