2021

TEAM
Alexa Steinbrück
Natalie Sontoski
Amelie Goldfuß
(Moving Target Collective)

PRESENTED AT
MozFest 2021  ➡️ more Info
New Now Festival, Zeche Zollverein Essen 2021
Kunstverein Meißen 2022

➡️ Website

Latent Riot is a series of artificial protest signs produced by a generative neural network. We trained a StyleGAN with images of hand-drawn street protest signs. We used the dataset Art of the March, an online archive of protest signs and posters collected in the aftermath of the historic Boston Women’s March in 2017. The signs are mostly drawn by hand, sometimes stenciled or made with sticky tape. The messages are witty, angry, often on point. Some protesters added drawings or collage to illustrate their message.
Generative neural network models like StyleGANs are good at capturing the visual properties of a training dataset. This makes it useful in creating a variety of copies that mimic its aesthetics on a shallow level.
However, like all machine learning models, the discriminator network of a GAN does not have a real understanding of what it sees: The text on these protest signs is analyzed pixel by pixel with respect to its visual features – contours and curves – not as words with meaning that reference the world around us, a world that is inherently political. AI systems are known to reproduce racial or gender biases while pretending to be a neutral entity.
The use of skewed datasets, the learning algorithms, the work force of engineers who conceive them and the wider infrastructure behind AI systems all contribute to bias.
Is it possible to make AI systems our allies in the fight for a more just world or will the boundaries and flaws of the technology (like racial or gender bias) continue perpetuate inequality?
The generated protest signs are currently presented as a virtual march on the website latent-riot.space