A Neural Network Has Been Created That Determines Sexual Orientation From A Photograph - Alternative View

A Neural Network Has Been Created That Determines Sexual Orientation From A Photograph - Alternative View
A Neural Network Has Been Created That Determines Sexual Orientation From A Photograph - Alternative View

Video: A Neural Network Has Been Created That Determines Sexual Orientation From A Photograph - Alternative View

Video: A Neural Network Has Been Created That Determines Sexual Orientation From A Photograph - Alternative View
Video: Alice O'Toole - Understanding face representations in deep convolutional neural networks 2024, April
Anonim

According to American researchers, the neural network they developed can extremely accurately name the sexual orientation of women and men from a photo.

Stanford University representatives Michal Kosinski and Yilun Wang posted interesting material in the PsyArXiv repository dedicated to their new creation - a neural network that can accurately determine the orientation of a person by facial features. Researchers talk about the high scientific value of their development, but they fear that it can act as a tool to invade a person's privacy.

To create a neural network, Kosinski and Wong used developments in other neural networks that serve for face recognition. The scientists took the base of photographs from one of the American dating sites: in total they selected 35 thousand photographs, which captured more than 14 thousand women and men. Scientists have made it so that the number of heterosexuals and homosexuals was the same. The authors of the work determined the orientation of a person by the gender of the partner whom the person is looking for on the dating site. It is noteworthy that the researchers selected only photos of white people. The site simply did not find enough gays and lesbians with a different skin color.

The neural network was taught to compare photos of two randomly selected people, one of whom was homosexual. Already from one photo, the neural network showed the correct result for men in 81% of cases, and for women - in 74%. If the number of photographs was increased to five, then the figure was 91% and 83%, respectively. The neural network performed much better than people who were given the same task. When determined by one photo, their indicators were 61% and 54%, which is clearly not enough for successful orientation recognition.

The results show that homosexual men had, on average, narrower jaws than heterosexuals, as well as elongated noses and high foreheads. Gays had less facial hair, although it is not entirely clear whether the reason for this was biological factors or the style chosen by the person. As for lesbians, on the contrary, they had wider jaws and lower foreheads. Lesbians have darker hair, and besides, they use significantly less makeup. The data obtained indicate that the propensity for homosexuality is laid in a person even before birth.

The relationship between facial features and sexual orientation / & copy; Michal Kosinski, Yilun Wang
The relationship between facial features and sexual orientation / & copy; Michal Kosinski, Yilun Wang

The relationship between facial features and sexual orientation / & copy; Michal Kosinski, Yilun Wang

Recall that recently another team of researchers developed a neural network that is able to distinguish the faces of people wearing masks.

Ilya Vedmedenko

Promotional video:

Recommended: