Artificial intelligence is wonderful, it’s what’s going to make the world move forward, but it’s also immensely scary at times. Case in point, a new study from Stanford University found that an AI can correctly tell if a person is gay or straight.
Let that sink in for a bit. An AI can scan your facial features and tell what your sexual orientation is.
The AI ran a test on more than 35,000 facial images publicly posted by men and women on a US dating website. The researchers behind the AI extracted features from the images with the help of deep neural networks and ran the algorithm. In 81% of the cases, the AI chose right for men, while for women the accuracy percentage was of 74%.
This entire experiment raises numerous questions. On one side, there’s the one about the biological origins of sexual orientation, something that many have argued for or against over the years. Then, there’s the entire ethics of the facial-detection technology, which is already widely used.
Furthermore, this particular AI can be a threat to LGBTQ+ people. There are nations around the world that outlaw same-sex relations. How many? Well, one study found 74. That’s 74 countries that could run this AI or a similar one, scan their entire ID database and see who needs to be put in jail, or worse. In 13 of these nations, being gay is punishable by death. Whether the AI’s accuracy can be improved or not is beyond the point, although that’s a large number of misses if you’re going to use this type of technology to tell what someone’s sexual orientation is. This is a tool that oppressive regimes can use to abuse citizens.
The risk of AI abuse
AI is a technology that can certainly be abused. It is something wonderful that humanity has come up with, something that has so many uses, from our smartphones to our search engines, but it carries a heavy dose of danger. The worst part is that there’s no way for this type of technology to be kept away from the bad guys – it’s not some sort of expensive weapon or a rare and dangerous ingredient that’s heavily restricted – it’s an AI that may require a bit more computer power, but nothing someone with deep enough pockets can’t get.
The researchers defended their study saying that this technology is already out there, and its capabilities can be the same in every corner of the world. It is important, they say, that governments and companies can consider privacy risks and take into account the need for safeguards and regulations.
They were somewhat thrilled, however, that they managed to bring more evidence to the fact that your sexual orientation is not something you choose, it’s something you are born with, and it may be all written in the features of our faces.
Researchers went a step further and said that this exact technology could be used to also find links between facial features and other things, like political views, psychological issues, or even personality. Up until now, AIs have only been involved in studying a person’s writing, social media usage and other metadata such as this to make such determinations. In fact, this is something that’s widely used as an advertising technique of sorts.
The ethical concerns regarding the study stand, however, and expand to Artificial Intelligence as a whole. We have the capability to do all these things now, but should we?