biAs&bIas

The work shows biases of a visual AI and thus visualises the weaknesses of the technology.


tl;dr
biAs&bIas uses sequential image generation to explore the dry abysses of a technology that's often perceived as a magical all-rounder. This approach allows to make its realistic opportunities and risks tangible and negotiable.

Concept
The fact that visual AIs can generate images in the blink of an eye. Many perceive them as magical visualisation artists. Despite they are often far away of being able to perform magic, let alone depict everything their users dream of.
It is now quite clear, there is at least as much sexism, racism, rejectionism etc. in them as is in society. It's because each of their components is influenced by human bias: their training data sets, the artificial neural networks developed for them, even the attempts to adapt them to an elusive neutrality. Because behind the scences humans make them work and each implements their very own bias of perception. As a result, perception and reality are often at odds when evaluating the visualisation capabilities of visual AIs.

biAs&bIas’ makes the gap visible. With the help of a special prompt strategy, the AI reveals its biases. ‘biAs&bIas’ bans the research results on one hand on a classic canvas and on the other captures them playfully alienated in video clips. The ‘painting’ is populated with 144 repeatedly similar image motifs: white people who embody to the ‘normal’ weight of the body mass index - the main bias of the AI. Meanwhile the video clips show biases regarding humans, flora and fauna.