We re-mixed copyright-free pictures of men and women and study how algorithms might respond to these fictitious images. Using fake “non-binary” images created by mixing images of men and women is of course different than using pictures of actual trans and non-binary people. Many of the artists we encountered while researching this project put their own bodies on the frontlines for their art. As neither of us identifies as trans, non-binary or gender-fluid, we chose against using our own images. Using photos of real trans and non-binary individuals without the proper permissions, frameworks and policies to guarantee their safety was also not advisable as we didn’t want to increase the risks these minority groups are already subject to in online spaces.

Artbreeder was chosen as the best free tool to achieve our goal and resulted in the creation of over 100 images, which we passed through an open-source facial recognition algorithm. Ageitgey/face_recognition is a freely available face recognition library and we used 1,200 photos from Kaggle’s female/male dataset to train this algorithm. We realise that by pre-labelling the dataset pictures into a binary of male or female we were perpetuating the same gender stereotypes that we were trying to challenge. We tried to mitigate this by using a different dataset to train the algorithm from the one we used to create the Artbreeder images, so that an exact match was never possible.

The Ageitgey/face_recognition algorithm allows for more than one face to be recognised in one picture, resulting in individuals that were identified as both male and female. The sample dataset might be too small to draw definitive conclusions on the algorithm behaviour. However, it was interesting to observe how certain pictures were identified as ‘unknown’: not belonging to either category, unrecognisable by the system. This fluid invisibility is what we tried to recreate with our main artefact: an animation of all the Artbreeder images seamlessly morphing into one another, being multiplicitous and uncategorisable.

With this project we sought to follow the example of certain queer artists who challenge the notion that inclusion in systems of domination is the end goal of the fight for LGBTQ+ liberation. These artists understand that whilst their algorithmic invisibility is a result of oppression and exclusion, this freedom from accurate surveillance and data extraction can be reappropriated in the service of freedom of self-expression.

The video documentation records our process, forefronting the “gender-fluid” personas we created and the animation showing the faces fluidly becoming each other.