Neural network's findings are 81% accurate
Technology moves in bizarre ways – often for good, but sometimes fairly creepily. The speed at which our smartphones, tablets and other devices have taken over our lives is stunning enough – at least for those of us who didn’t see it coming – and with technology being asked to do more and more as our everyday lives are becoming increasingly autonomous, it is hardly surprising that so many people find the future of tech more than a little intimidating. This week, however, artificial intelligence just developed a whole new skill – and it’s more than a little unsettling, and possibly even debate-enraging.
Researchers have been able to build and train AI to look at patterns in people’s faces in an effort to look for correlation in sexual orientation. Taking into account more than 35,000 different faces, faces were split up into either ‘heterosexual’ or ‘homosexual’ – and the findings were scarily accurate. When asked to correctly identify the orientation of the person whose face was being analyzed, the neural network trained by the researchers managed to score a staggering 81% success rate. This means that, when presented with one image at a time, the AI was able to figure out who is heterosexual and who is homosexual with worrying accuracy.
It is worrying as it could fuel debate – should AI be able to do this? Is this technology that makes presumptions that could be considered distasteful? According to one of the study’s authors, Michal Kosinski, heterosexual men have ‘(typically) larger jaws, shorter noses, and smaller foreheads’, while homosexual males possess ‘narrower jaws, longer noses, larger foreheads and less facial hair’. The author also advised that female homosexuals generally possess ‘more masculine faces’ than heterosexual women. Could these points be generalizations – or could there be answers lying in what we are all exposed to during growth in the womb?
The study suggests that prenatal hormones and exposure during gestation could point towards the sexual orientation we grow into – though it should be noted that this is not considered the sole driver for such factors.
Does this AI’s results show that defining sexual orientation is as simple as reading faces? Does it spell a scary future for technology as we know it? Or has the study opened up a can of worms that is just getting spilled? Let’s wait and see – as fascinating as this work may be.