(CN) — There’s no easy way to get to know a stranger — some people are hard to read and after hours of conversation they can often remain aloof. Artificial intelligence and a selfie may be the key to cracking their shells.
According to a study published Friday in the journal Scientific Reports, a group of Russian researchers say with the help of some machine learning, they trained a series of neural computer networks to determine a person’s personality through photographs. They say the selfie betrays all and allows the program to outperform any humans tasked with trying to predict someone’s personality.
The program more reliably determined personality from women’s faces than men’s, and was able to make above-chance judgments about openness, conscientiousness, extraversion, agreeableness and neuroticism.
These personality traits make up the Big Five personality theory or the OCEAN model developed in the 1980s in the field of psychology. The traits and characteristics that fall under those categories are found to remain consistent from childhood through later in life.
The research teams who conducted the study worked with Russian-British startup BestFitMe to train their artificial intelligence program to make the personality judgments on a sample of 12,000 volunteers and their selfies, according to the study authors.
Volunteers filled out a questionnaire that measured their personality traits and then the study authors uploaded 31,000 selfies into the neural network. These volunteers had their information split into a training and test group, where a series of extra images were thrown in to make sure the program could exclude faces that had emotional expressions, pictures of celebrities and cats.
The selfies were then broken down into 128 invariant or unchangeable features that were fed into a computer algorithm to make its predictions.
According to the study authors, the AI can guess correctly about the relative standing of two randomly chosen individuals on a personality dimension in over half of the cases, as opposed to less if left up to chance.
When compared with the questionnaires and someone trying to rate a person’s personality trait, the artificial intelligence that based its predictions on static photos was able to outperform the human observer.
From the five traits, conscientiousness was easily the most recognizable and women were easier to read than men.
Researchers say they want to apply their technology to dating services, online tutoring and customer service.
The research team were from the Russian universities HSE University and Open University for the Humanities and Economics. BestFitMe describes itself as a face analysis platform that “provides an innovative and unique software solution for instant psychographic profiling from facial photo.”
The study authors did not immediately respond to a request for comment.
Facial recognition software has crept into more facets of everyday life, like at the airport or with law enforcement agencies utilizing the technology at checkpoints. Some states and cities in the U.S. have banned or restricted the use of facial recognition software in police body cameras while other nations have embraced the idea of tracking its citizens.