In the study, which involved the examination of more than 35,000 images, it was found that the computer algorithm could correctly distinguish between gay and straight men in 81% of the instances. For women, this number was 74%. When the software reviewed five images per person, this accuracy rocketed up to 91%.
Click to read More big developments in the field of AI
In contrast to the algorithms, the human judges performed much poorly. They accurately identified the sexual orientation correctly 54% of the time for women and 61% for men.
Just in case you’re wondering, the public images used for the study were pulled from an online dating website. From the user profiles, the exact sexual preference was determined, as reported by The Economist.
These images were fed into a software named VGG-Face, which assigns a ‘faceprint’ number to each face. After that, using a predictive model named logistic regression, the correlations between a person’s facial features and sexuality is determined. The neural network is trained with tons of pictures to help it develop its own understanding.
The researchers, Dr. Kosinski and Mr. Wang, have also offered an explanation for this outcome. According to them, as a fetus grows in mother’s womb, it’s exposed to different hormones–including testosterone–that play a part in the development of facial structures. The AI model is able to pick the subtle signals of a person’s sexuality from a man’s nose, eyes, eyebrows, cheeks, hairline, and chin. For women, the most important parts are nose, mouth corners, hair, and neckline.
You can find the complete study here in PDF format.
Did you find the revelations made by this Stanford study interesting? Aren’t there some obvious ethical implications? Don’t forget to share your views and feedback.
Also Read: “A.I. Will Make Humans Completely Useless” — Warns Bill Gates-approved Historian