Facial recognition fails on trans and non-binary people

*By Rodrigo Almeida, Public Relations, Master in Management and Technology; and Director of the agency CRIATIVOS

The use of facial recognition is still a challenge for technology companies. Allowing for “fast, secure and practical” authentication for the user, the tool, when used on mobile and/or authentication devices, usually requires some simple care, such as adequate lighting and a neutral environment, minimum imperatives compared to what really affects us.

In 2022, I wrote the article “Algorithmic Racism and the Challenges of Artificial Intelligence” for the Digital Look. In this article, I highlight that, “according to data from the Security Observatory Network, in 2019, around 90.5% of people wrongly detained by facial recognition in Brazil were black”.

With much controversy after publication, I provoke in it the importance of developing different standards at the time of construction and validation of artificial intelligence (AI), so that, in this way, we have more assertiveness, impartiality and equality in its use.

Now, in 2024, I return to this subject to, once again, provoke insufficient training data, but, this time, with another focus: the error in facial recognition in the gender of trans people.

According to a study by the University of Colorado in the United States, the accuracy rate in facial recognition for trans and non-binary people is less than 60%, compared to the accuracy rates of 98.3% for cisgender women and 97.6% for cisgender men.

Yes, this discussion is not simple.

According to journalist Pedro S. Teixeira, in the article “Facial recognition gets gender of trans people wrong” for Folha de S.Paulo, “the error occurs because the identification algorithm is based on measurements of the face, such as the distance between the eyes or between the nose and the upper lip, and can confuse the gender of the person identified”.

Coherent, but not inclusive, for some people, the solution may be for the person to update their data after the transition, right? No! The transition process for trans people involves time and phases. According to experts, the hormone phase is the one that causes the most system failures, which would require the person to continually update their image in ALL companies and devices that use this intelligence. And that's where we'll answer together: how healthy, agile and inclusive is this?

Read more:

Apparently without a solution, the Publicis Groupe trans affinity group (Égalitrans), in partnership with the NGO Casarão, developed the “I exist” API, an application programming interface that aims to train the algorithm of facial recognition systems and, thus, end, or at least considerably reduce, failure situations.

The challenge now is to convince companies about the importance of technology, the inclusion of diversity in artificial intelligence mechanisms and the social demand that all people feel included – respected and considered – in any technological and social advancement.