FOLLOW US ON: follow us in feedly
Artificial intelligence will soon achieve near-perfect gaydar
09:00 am

We all have a friend who claims to have “perfect gaydar”—maybe in some instances we are that friend. At the risk of venturing into “special snowflake” territory, my own tendency in this area is to assume that a person’s sexuality is usually at least somewhat unknowable from surface appearances.

However, the artificial intelligence community is intent on proving me wrong! That confident friend who can claim to pick “the gays” out of any crowd…. might exist fairly soon, in the form of computer applications, which have recently seen startling success in identifying sexual preference based on a single photograph. And they are doing it without the benefit of a gif of the person doing that limp-wristed “tinker-bell” gesture that was universally acknowledged to signify “gay” in 1980s TV (watch any episode of Three’s Company).

Gay traits may mostly be a stereotype, but a Stanford University study into facial features has demonstrated that a computer could determine sexual orientation in men an astonishing 81 percent of the time and in women 74 percent of the time—on a sample size of a single image. When the program was given more than one image, the success rate increased to 91 percent and 84 percent, respectively. For some reason, it will be noticed, gay men are easier to “identify” than gay women. One theory states that perhaps (as is generally suspected) sexuality really is more “fluid” for women.

Note that for comparison, when people assessed the same images, they had a success rate of just 61 percent for men and 54 percent for women. Those numbers sort of establish that gaydar among people is an actual thing, right? 54 percent is close to a coin flip, though.

The AI was trained to assess bone structure and facial features, on the premise that gay men were more likely to have feminine features and gay women more likely to have a masculine appearance. The study looked at jawlines, hairlines, nose length, among other features. According to The Guardian, “The data also identified certain trends, including that gay men had narrower jaws, longer noses and larger foreheads than straight men, and that gay women had larger jaws and smaller foreheads compared to straight women.”

As amusing as the concept of gaydar AI is, the prospect of its existence does suggest some fairly obvious potential problems, including the possibility that organizations premised on homophobia could use such technology to discriminate against LGBTI people. Since the program appeared to use physical characteristics to make its assessments (and not aspects that are a later choice by the user), it suggests that homosexuality may be more innate than it is a product of a person’s upbringing and environment.

It’ll be very interesting indeed to track the progress of this technology over time.
via Lost at E Minor

Posted by Martin Schneider
09:00 am