AI can tell if you’re gay or straight just by your face - study
A total of 35,326 pictures of 14,776 people, obtained from an unidentified dating website containing both straight and gay men and women, were analyzed by the software known as VGG-Face, which turned their image into a string of numbers to represent their face.
An algorithm compared facial features to predefined parameters to determine if they were gay or straight, with 81 percent accuracy in men and 74 percent in women.
Their human counterparts trailed behind with 61 percent for men and 54 percent for women.
When the dataset was increased to five images per subject the software’s accuracy increased to 91 percent for men and 83 percent for women.
The authors of the study, published in the Open Science Framework, warned their findings “expose a threat to the privacy and safety of gay men and women,” in an age where companies and governments are increasingly using computer algorithms to detect people’s intimate traits.
A “gender-atypical facial morphology” was associated with gay men, according to the study, including grooming styles and expressions deemed more feminine, as well as narrower jaws, longer noses and larger foreheads.
Gay women were deemed to have smaller foreheads and larger jaws than straight women.
The authors of the study say the findings support the parental hormone theory (PHT) of sexual orientation, which states that exposure of fetuses to androgens determines their sexual orientation, confirming the widely acknowledged idea that sexuality is determined before birth and is not a choice. Androgens are also associated with the determination of facial features.
People of color were not included in the study as the authors could not obtain a sufficient sample size. Bisexuality was not included in the list of results.
“We hope that our findings will inform the public and policymakers, and inspire them to design technologies and write policies that reduce the risks faced by homosexual communities across the world,” the authors said.