A.I. can now correctly predict something we think about privately

Personality isn't skin deep, right? Don't be so sure. According to a group of researchers and an incredibly perceptive A.I., our faces give away something quite intimate — how we view our own personality; our own self-image.

Scientists have designed a neural network to analyze the selfies of over 12,000 anonymous volunteers and evaluate their personalities based on their neutral facial expressions alone. After processing and breaking down these photos into 128 different components, the A.I. was able to correctly match photos with the volunteer's self-reported personality trait 58 percent of the time -- higher than previous A.I. models and even human raters.

The researchers write that this better understanding of human personality perception could help A.I. better recommend your perfect match on a dating app or even improve A.I.-to-human interactions in the near future.

In a new study, published Friday in the journal Scientific Reports, a team of Russian and British psychologists chose to focus on the "Big Five" personality traits when designing their study. These traits include: openness, conscientiousness, extraversion, agreeableness, neuroticism. While you may be familiar with these traits through less scientifically rigorous personality quizzes such as Myers-Briggs or the Enneagram Test, these big five traits are also the gold-standard when it comes to personality testing.

The authors write that previous attempts in this field to determine how our faces display these characteristics have been mixed at best.

"Existing studies have revealed the links between objective facial picture cues and general personality traits based on the Five-Factor Model or the Big Five (BF) model of personality," write the authors. "However, a quick glance at the sizes of the effects found in these studies reveals much controversy. The results appear to be inconsistent across studies and hardly replicable."

The authors explain that poor results from previous studies may be linked to factors such as inconsistent methodology, small sample size, or simply the inability of human raters to be truly objective when judging photo samples.

To remedy this, the team behind the study chose to skip human raters altogether and instead designed an A.I. to do their job.

After sourcing over 31,000 selfies from 12,447 anonymous Russian volunteers, the researchers trained the first layer of a neural network to recognize faces and individuals (as well as weed out false images in the dataset like cats of celebrities.) Next, a second layer of the network was taught to deconstruct these photos into 128 invariant facial features. Then, based on these features and previously understood psychology about how facial features correlate to personality, the A.I. judged these photos based on the big five personality traits.

When comparing the volunteers' self-reported personalities with the judgment given by the A.I. based only on their selfies, the researchers found that the A.I. was correct 58 percent of the time. This percentage was higher than chance and also higher than previous A.I. and human-rated results. The researchers also report that of all the big five personality traits, conscientiousness was the easiest to correctly identify.

More work still has to be done to determine how much personality a selfie has -- versus a clinically staged photo -- and how it could potentially sway the A.I.'s perception. But, the researchers are still optimistic about the way this technology could be used in the future to improve human-to-human as well as A.I.-to-human interactions.

"There are a vast number of potential applications to be explored," write the authors. "Applications may include suggesting best-fitting products or services to customers, proposing to individuals a best match in dyadic interaction settings (such as business negotiations, online teaching, dating) or personalizing the human-computer interaction."

Abstract: There is ample evidence that a human face provides signals of human personality and behaviour. Previous studies have found associations between the features of artificial composite facial images and attributions of personality traits by human experts. We present new findings demonstrating the statistically significant prediction of a wider set of personality features (all the Big Five personality traits) for both men and women using real-life static facial images. Volunteer participants (N = 12,447) provided their face photographs (31,367 images) and completed a self-report measure of the Big Five traits. We trained a cascade of artificial neural networks (ANNs) on a large labelled dataset to predict self-reported Big Five scores. The highest correlations were found for conscientiousness (.360 for men and .335 for women), exceeding the results obtained in prior studies. The findings provide strong support for the hypothesis that it is possible to predict multidimensional personality profiles from static facial images using ANNs trained on large labelled datasets.