We know that we can judge someone’s psychological (happy, sad, mad) and demographic (gender, race, age) characteristics from a quick glance. Most people dismiss physiognomy, the science of judging personality or character by their facial appearance, as superstition, racism, and pseudoscience. And historically, they’d be right. However, while still a controversial subject new evidence is emerging that in three instances it does seem there is a connection between facial appearance and certain traits.
- Research suggests that character can influence one’s facial For examples young extroverted women are rated more attractive later in life.
- Facial appearance can alter one’s character. For example, good-looking people receive positive feedback, and thus tend to become even more
- Research has identified a range of factors influencing facial appearance and one’s traits, including; pre- and post-natal hormonal levels, environmental factors, and gene For example, testosterone levels significantly affect both: behavior and facial appearance.
In a groundbreaking study, Yilun Wang at Stanford University asked whether Artificial Intelligence (AI) can predict a private, psycho-demographic trait, Homosexuality by simply analyzing an individuals photo. Advances in AI and Deep Neural Networks (DNN), often referred to as Machine learning, the researchers extracted facial images of 35,326 gay and heterosexual men and women. DNN’s are incredible at taking large unstructured data and finding patters to make predictions, which are often more accurate than humans.
When humans are asked to predict another’s sexuality their accuracy is limited, ranging from 55 to 65%. However, as Yilun Wang states, “in 81% of randomly selected pairs—composed of one gay and one heterosexual man—gay men were correctly ranked as more likely to be gay. The accuracy grew significantly with the number of images available per person, reaching 91% for five images. The accuracy was somewhat lower for women, ranging from 71% (one image) to 83% (five images per person).” Amazingly, AI was able to predict up to 91% accurately whether an individual was homosexual.
How is this possible? The authors postulated that in line with prenatal hormone theory (PHT) “ gay men should tend to have more feminine facial features than heterosexual men, while lesbians should tend to have more masculine features than heterosexual women. Thus, gay men are predicted to have smaller jaws and chins, slimmer eyebrows, longer noses, and larger foreheads; the opposite should be true for lesbians.” The AI system was able to detect these slight and nuanced variations that are not easily discernible to the human eye.
Facial images are everywhere. From social media, to dating sites, and government databases facial images are ubiquitous. These photos are publicly available from Facebook, LinkedIn, Google, and others. In concluding the researchers point to the potential dangerous way this information could be used by governments and corporations—who likely are developing similar and likely more sophisticated algorithms to detect psychographic characteristics. The possibilities seem to write a dystopian novel, but this one is already real. This then quickly becomes not just a homosexual issue, but a human one. The need for privacy in the digital age and how this information is collected, safeguarded and used is of great consequence. The authors even considered not publishing the paper as the results could be terrifying if used inappropriately, but in the end determined “It is thus critical to inform policy makers, technology companies and, most importantly, the gay community, of how accurate face-based predictions might be.”
For an excellent in-depth discussion on this topic, check out the recently published NYT Magazine article: Can AI be Taught to Explain Itself.
The study, Deep Neural Networks Can Detect Sexual Orientation from Faces, was co-authored by Michal Kosinski.
Photo Credit: Yilun Wang & Michal Kosinski from study.