Will gadgets be able to determine our character from a photo?

In principle, any of us is able to determine by the appearance of the interlocutor his personal qualities, character traits. For example, to understand how friendly or emotionally stable he is. This estimate is not always true, but it is not accidental. A trained person, for example, a specialist in non-verbal communication, will cope with this task even better. It is enough for him to look at the photo.

And now a group of Russian psychologists and artificial intelligence specialists have created algorithm, which is able to recognize our personality traits from a photograph more accurately than a stranger or a specialist who saw us for the first time would.

For this, a team of Russian researchers trained several neural networks, which, interacting with each other and working in a complex, perform the same task as a person in the process of communication, namely: they evaluate the predisposition of a person in a photo to behavioral features, which are expressed as five basic traits (the so-called «big five»):

— extraversion (orientation to the outside world, sociability, assertiveness),

— friendliness and willingness to compromise,

— conscientiousness and conscientiousness (including punctuality, reliability, readiness to follow the rules),

— emotional instability (high or low levels of anxiety, self-esteem, resistance to stress),

— openness to new experience (flexibility of mind, imagination, independence of judgment, creativity).

How was the training?

“The neural network imitates the work of the human brain in some aspects and learns in much the same way as a person,” says Alexander Kachur, head of the scientific team, artificial intelligence specialist. — When training, it is given a set of input data and reference output data (dataset).

In our case, it was more than 30 photographs from 12,5 men and women sent through social networks, and the results of tests that these volunteers underwent at our request to determine their personality portrait and the severity of the Big Five traits.

One part of this data served as a training sample for the neural network. The second is a control sample to evaluate its performance. Gradually, the neural network learned to decompose each image into 128 invariant features and further calculate the patterns of their connection with the character.”

Control measurements showed that artificial intelligence correctly determines the level of expression of a character trait in an average of 58% of cases, while when randomly guessing a stranger, this figure is 50%.

All traits, except for extraversion, the stranger guesses worse than the algorithm. The difference is small, but this is only the first stage of work. The researchers are confident that the more data the algorithm compares, the more targeted the results will be in the future.

recognize the liar

It has long been known about the relationship between character and appearance — in particular, the face. Physiognomy was fond of even in Antiquity. But attempts to describe facial features and connect them with character, undertaken in the XNUMXth and early XNUMXth centuries, were unsuccessful.

“Data that certain facial features are associated with character gradually accumulated over the course of the XNUMXth century,” says one of the authors of the study, Associate Professor of the Department of Psychology at the National Research University Higher School of Economics Evgeny Osin. “But these connections were very weak, and it was impossible to rely on them to make judgments about the personality, as the physiognomists did.”

A team of Russian researchers was the first to suggest that these connections are non-linear and are based not on individual traits, but on their complex.

“Our neural network looks at the appearance like a person,” Alexander Kachur adds, “that is, it does not single out the nose or lips, but perceives the entire set of features at the same time, evaluating them in a 128-dimensional space.”

Who needs it?

The authors of the project suggest that this technology will allow us to better understand ourselves, make more informed and informed decisions in various situations, and better interact with the outside world. Did the researchers manage to discover something new in themselves?

“Technology has really helped us learn more about ourselves,” Alexander Kachur explains. — Among us, for example, there were introverts who did not know about it and were forced to lead an open, extroverted lifestyle. Having received the necessary recommendations from the program, we organized the workspace in a new way — we divided the office into two parts.

In the far quiet zone, introverts are located, where they can communicate less, plunging headlong into formulas, algorithms and writing code. And we gave the other part to extroverts, life is seething there, noisy and fun there. I think as a result we began to interact more effectively as a team.”

For business, the technology is also useful — understanding the buyer, companies can more accurately form their product offering and offer customers what they need, and not what they want to sell. Some experiments of the authors of the study have already confirmed this hypothesis.

In an experiment with finding a salesperson who would most effectively interact with a customer, sales actually increased as customers were more likely to be offered something that met their needs.

Other possible applications of the algorithm are dating applications, employment diagnostics, recommendation systems for online stores, and so on.

The path to discrimination and predestination?

Will it turn out that the use of the new algorithm will lead to discrimination, allowing, say, to cut off people who are recognized by the program as aggressive, mentally unstable and irresponsible when hiring?

“This is a very simplified view, which was guided by physiognomy, and that is exactly why it failed,” Evgeny Osin comments. — Firstly, the capabilities of the algorithm should not be overestimated: it does not give an exact diagnosis, but an approximate one, which somewhat improves the quality of our forecast.

For example, if out of 1000 people, 50% of whom are extroverts and 50% are introverts, we want to select 100 extroverts for a job, then out of 100 people whom the algorithm will advise us, about 70% will be extroverts. But this does not mean that all the others are not suitable for us.

Therefore, it is worth using this system primarily in situations where the cost of an error is not so high, and a correct hit gives a tangible gain: for example, when matching buyer-seller or teacher-student pairs.

Secondly, it seems to us that if the shape of the face is predetermined, then the personality is also predetermined. And this is also not true at all. Of course, there are some biological givens (genes, hormones) that affect both.

But in relation to personality, to use the metaphor of Dmitry Leontiev, these are just “default settings”: by working on oneself, a person can develop those character traits that seem valuable to him or others, and learn to compensate for those manifestations, the consequences of which he would like to change.

An introvert can learn to behave like an extrovert, an emotionally unstable person can learn to control himself, and so on.

Research shows that over the course of a lifetime, personality traits in people, on average, develop in an adaptive direction that is desirable for themselves and for others.

And the signals that the neural network reads from the photo are not reduced to biological data: the muscles of the face and wrinkles reflect our usual patterns of tension, and the position of the head, lighting and other characteristics of the photo are how we position ourselves.

It all depends on our personality traits and changes with them.”

Study authors: Alexander Kachur (Artificial Intelligence LLC); Evgeny Osin, Deputy Head of the International Laboratory for Positive Psychology of Personality and Motivation, Department of Psychology, National Research University Higher School of Economics; Denis Davydov (OSUE), Konstantin Shutilov and Alexey Novokshonov (BestFitMe). Source: Alexander Kachur, Evgeny Osin et al. Assessing the Big Five personality traits using real-life static facial images. Scientific Reports, 10, Article number: 8487 (2020) Published: 22 May 2020.