Explore how AI reflects societal biases in this interactive art installation, "Echoes of Exclusion." Discover the hidden prejudices within image and voice generation models – learn more about the project's concept and technology.
Abstract
This interactive installation, Echoes of Exclusion, investigates the latent biases and patterns embedded within generative artificial intelligence (AI) systems. Developed as part of the Master of Arts in Interaction Design programme at University of Applied Sciences and Arts of Southern Switzerland, the project utilises a range of AI models to explore image description, generation, and voice cloning, prompting reflection on the potential for algorithmic prejudice.
The Echoes of Exclusion Installation made for Swissnex Japan - Image courtesy of Matteo Subet under CC BY SA 4.0
Concept
The core concept behind Echoes of Exclusion centres on revealing the ‘thought processes’ of generative AI. These systems, increasingly prevalent in creative fields, are trained on vast datasets which inevitably reflect existing societal biases. By allowing users to interact directly with these models – specifically LLaVa v1.6 (leveraging Mistral 7B LLM), FLUX.1 [dev] 12B, and Coqui XTTS-v2 – the installation aims to expose how these biases manifest in generated outputs. The project encourages critical engagement with AI’s role in shaping perceptions and perpetuating exclusion.
The Echoes of Exclusion Installation made for Swissnex Japan - Image courtesy of Matteo Subet under CC BY SA 4.0
Interaction with installation
The interactive experience is initiated by scanning a QR code displayed near the main screen. This redirects the user to a web interface where they can upload their own image. Following processing by the generative AI systems, users observe the progress alongside other generated images on the display. Once audio generation is complete, a thank you page provides further information and instructions. A key element of the installation involves presenting the user’s uploaded image alongside the AI-generated interpretation, highlighting potential discrepancies and biases. The project was developed by Amina Fatou Cissè, Cathrine Hansen, and Hannah Kastrytsa, with crucial implementation and refinement support from Matteo Subet. It was undertaken during the Prototyping Interactive Installations course led by Leonardo Angelucci in 2024.
The interface of Echoes of Exclusion - Image courtesy of Matteo Subet under CC BY SA 4.0
Note
Full technical documentation for Echoes of Exclusion is available on GitHub
How to cite this page:
Subet, Matteo (2025). Echoes of Exclusion. Retrieved on 18 December 2025 from zumat.ch/projects/echoes-of-exclusion.html