https://into-the-hidden-layers.github.io/

About

With the introduction of complex AI systems, generative technologies used in creative processes have become more obscured through their many hidden layers. However, below the surface of AI technologies lie rich material characteristics, that we have yet to explore.This workshop invites the community to explore alternative modes of creating with AI by approaching the hidden layers as a mechanistic material for design. Existing XAI approaches often talk around the material of AI systems by trying to explain the behaviour of the hidden layers for reasons of efficiency, simplication, or user-friendliness. To approach this gap, this workshop proposes to explore what better conditions for such manipulation might look like, especially for creative practitioners. By centering the materiality of AI systems rather than explaining their opacity, we propose to consider the hidden layers of neural networks as directly manipulable material for creation and explore on the interventions and interfaces that enable users to manipulate these systems' inner workings.We prompt researchers ranging from the technical field of mechanistic interpretability to interaction design and beyond to explore the questions:

• How can you investigate the material properties of the hidden layers?
• How could you imagine interacting with the material hidden inside the layers of neural networks?

Call for Participation

To participate in the workshop, researchers and designers from various domains are invited to submit their work. We welcome submissions that explore co-creation via AI's hidden layers through any kind of modalities, such as music, textual, visual, voice, code-based, bio-signals, sensor data, and multimodal systems.

These can take the form of

• material investigations of technical character (e.g., using interpretability methods) that explore the internal mechanistics of AI systems. Submissions can present results of material investigations, describe the development of tools like interfaces, artworks, or other products, or document the actions taken during the process. We also encourage submissions of failed approaches documenting the difficulty of entering the hidden layers of neural networks.
• interaction imaginaries of speculative character that propose alternative forms of co-creation. Submissions can draw on metaphors to existing digital or physical tools (e.g., the layers in Photoshop or looms) or present fictive design scenarios. Dystopian imaginaries that critically explore the potential negative effects of material gateways into AI's hidden layers are also encouraged.

Submissions are accepted in the following formats:

Pictorials, up to 4 pages (e.g., see here)

Computational notebooks(e.g., see here)

Interactive web articles(e.g., see here)

Papers, up to 4 pages (e.g., see here)

Special emphasis is placed on the representation and visualization in the submissions. As the modalities of the hidden layers are not necessarily natural to human understanding and perception, we invite participants to freely explore how to represent them in various forms. In addition to new works, we also encourage authors to submit previously published work (e.g., from technical papers or products) that can be contextualized under the theme of Hidden Layer Interaction.

Submission Deadline: May 7, 2026

Please submit your contributions via EasyChair.

Submissions do not need to be anonymized. Please use the corresponding templates for papers and pictorials.