Computers are binary, but humans aren't: why automated gender recognition is a high-tech gender essentialist fantasy

Automated Gender Recognition (AGR), sometimes referred to as ‘gender detection software’, takes a bio-essentialist construction of gender and codifies it.

In February 2021, Roxanne Tickle uploaded a selfie as the price of admission to the women-only social network, Giggle for Girls. Using automated gender recognition software, the app determined that Tickle was a woman, and let her in. Seven months later, Giggle CEO Sall Grover manually removed Tickle from the app. “I looked at the onboarding selfie and I saw a man,” she told The Australian, “The AI software had let them through, thereby making a mistake that I rectified.” 

Now Tickle, a trans woman, is suing Giggle and Grover for alleged unlawful discrimination in a landmark gender identity discrimination case. The judgement is expected in the coming months and will likely have significant consequences for the direction of trans rights in Australia as well as potentially wide-reaching implications for sex-specfic spaces and activities.  

But what of the technology Giggle used to attempt to keep trans women out? 

Automated Gender Recognition (AGR), sometimes referred to as “gender detection software”, sits in the same family as facial recognition technology. Using AI techniques including computer vision and machine learning, AGR attempts to infer someone’s gender from their biometric data like bone structure and skin texture. Despite the name, this technology doesn’t recognise or detect gender—gender isn’t something that can be recognised or detected from external factors—it makes a prediction of gender and assigns it to people, often without their knowledge or consent.

If your understanding of gender is anything beyond short hair, strong jawline = male and long hair, high cheekbones = female, then this might sound a bit, uh, reductive.

But such is the nature of AGR: it takes a bio-essentialist construction of gender and codifies it.