14 Sep'17

Facial recognition software that “can” distinguish between homosexuals and heterosexuals


Stanford University study claims it has facial recognition software that can identify the sexual orientation of an individual. BBC News reports that the study has been accused of being dangerous and junk science.


In response, the scientist responsible has stated that this response is a knee-jerk reaction.


The researches created an algorithm based on photos of more than 14 000 white American from a dating website. They marked individuals’ sexuality as the individual reported on their dating profile.


When the software was presented both a photo of a gay man, and a heterosexual man, it was able to determine the sexuality accurately 81% of the time.


With the same process using photos of women, it was less accurate at 71% percent.


Quoted by the BBC, a researcher stated, ‘Gay faces tended to be gender atypical’. The research clarifies, ‘Gay men have narrower jaws and longer noses, while lesbians had larger jaws’.


Doesn’t this mirror the historical time when a criminal was identified based on their appearance? There is a reason why this theory was abandoned.


In another test when they presented their software with 1000 photos of men – 70 being gay, 930 being heterosexual – and asked to identify 100 men most likely to be gay, it missed 23.


These findings show that this software is flawed, and does not wholly distinguish between heterosexuals and homosexuals.


In a review of this study, The Economist highlighted the study’s flaws including the fact that the sample only included white Americans, and using images from dating sites, which were ‘likely to be particularly revealing of sexual orientation’.


The study approaches sexuality in a black and white manner, but this is not appropriate as sexuality is a spectrum.




‘facedetection’ by Caulier Gilles on Flickr. Licensed by CC by 2.0