Published Date : Sep 11, 2017
Stanford University researchers have authored a study that used deep neural networks, a form of technology, to detect sexual orientation with the help of just facial images. Currently available in a draft form, the details of the study have been expected to be published in the Journal of Personality and Social Psychology. The accuracy of the algorithm used has been studied to have increased to 91.0% and 83.0% when the technology was fed with five images of each person. Two types of facial features were included, i.e. transient type such as grooming style subject to change as per the individual’s preference and fixed type such as the unalterable shapes of nose and other facial parts.
Such AI Systems could also Reveal IQ or Political Views in Future
A facial-detection technology was implemented to sort images with most clarity. Finally, 35,326 images of 14,776 people were left that fairly represented both heterosexual and homosexual men and women. The images were accessed through an American dating website. The VGG-Face would first detect the face in an image and reveal a string of numbers that characterize the face. This is called faceprint. Next, statistical tests were used for analysis to detect if an individual was straight or gay. About 81.0% of the cases marked the successful differentiation between homo and heterosexual men and 74.0% of the cases between homo and heterosexual women for each image that the technology was exposed to.
As a conclusion, the study authors have added that this technique could potentially risk the safety and privacy of gay men and women in the coming days.