The Artist’s Atlas

Staking Territory in Latent Space, 2022

Jenn Karson

Like a cartographer documenting an unknown realm, this body of work maps the territory of multidimensional latent space – the vast mathematical domain where artificial intelligence systems encode and transform information, bringing into visible form what typically remains hidden. Using an AI system trained on the Athena Dataset (#ownyourdataset), I fabricated four distinct forms of mapping: documented ‘fakes’ that pair generated images with their technical origins (like field notes from an exploration), sequences that trace paths through this abstract space (like charting routes through unmapped terrain), 3D-printed sculptures that serve as topographical models of this abstract territory, monotype prints that traverse between digital and analog territories using both a traditional printing press and its miniature 3D-printed replica (mapping across technological epochs), and laser-engraved silicon disks that etch these surveys onto the bedrock of computation itself. 

While modernist artists like Picasso and Duchamp represented multiple dimensions on canvas, this work directly engages with the challenge of comprehending and materializing a space that’s dimensions exceed human perception. By charting these new aesthetic territories across multiple materials and dimensions, the work invites viewers to explore boundless mathematical realms that underlie contemporary artificial intelligence, while the silicon engravings remind us that even the most functional components of computing technology carry their own compelling visual language. This body of work is significant in my oeuvre, because it maps a foundational framework for works that followed. 

This work originated from the Athena Dataset. 

Field Notes from Latent Space 

Digital prints on paper (six)   22″ x 27 1/2” 

These paired images and documents serve as field notes from an exploration of AI’s latent space. Each generated image is accompanied by documentation that makes transparent not only its technical origins, but also the entire network of people, funding sources, and power structures that made it possible. Using frameworks inspired by Gebru’s data and model documentation practices, the work reveals both the computational processes and the human infrastructure – from software developers to financial backing – that underlies these AI systems. This comprehensive documentation challenges the notion of AI-generated images as autonomous creations, instead situating them within their full social, technical, and economic context.

Latent Space Walks 

video animations with original sound and music.

Animating how vector points exist in the abstract multidimensional mathematical space of the neural network, these latent space videos are made from selected vectors generated by a model trained on the Athena dataset.These image sequences chart routes through the multidimensional terrain of latent space, revealing how forms morph and transform as we move from one vector to another. By showing these gradual transitions, the work makes visible how AI systems interpolate between different points in their mathematical space. These ‘latent walks’ offer viewers a way to comprehend the continuous generative nature of this vast abstract territory. More from this series.

Topographies of an Alien Space

Stereolithography resin 3D print (six) various sizes
colored gels, light table    40″ x 10″ x 16″

These 3D-printed resin sculptures translate AI-generated images into topographical landscapes, giving physical form to the flat outputs of latent space exploration and what Luciana Parisi calls “soft thought.” Each piece represents an interpretation of these digital territories in three-dimensional space. The sculptures make tangible what normally exists only as mathematical abstractions, allowing viewers to physically encounter these computational forms.

Scratching Below the Silicon Surface

Laser engraving on silicon wafer (six)  5″ each, 2022
These laser engravings were made on rejected crystalline silicon integrated circuit (computer chip) semiconductor fabrication wafers using territories

Laser-engraved onto silicon disks, these etched pieces map AI-generated forms onto the fundamental material of computing itself. The aesthetic qualities of the silicon disks – a byproduct of their technical function – become an intentional canvas for this exploration. By merging the mapped forms with this technological substrate, the work reveals how computational materials carry their own inherent visual language while referencing the parallels between the etching techniques used in traditional art and the semiconductor manufacturing process.

Reciprocal Impressions

These monotype prints chart a journey between computational and physical realms, using both traditional and miniature printing presses to explore questions of originality and reproduction. Each print begins with an AI-generated ‘fake’ (generated from a model trained on an original artist-made dataset) transformed into a laser-cut stencil, which then becomes a tool for creating unique impressions through the centuries-old technique of monotype printing. Using both a grand 36″x50″ American French Tool Press and a 3D-printed miniature press (based on the Open Press Project), the work maps the territories between digital reproduction and analog uniqueness. Each print becomes a singular cartographic record of this crossing between realms, challenging our understanding of what constitutes an ‘original’ in both digital and physical spaces.

The Sound of High-Performance Computing: 

In 2020, during the pandemic, I found myself wondering what our high-performance computing center – the Vermont Advanced Computing Center (VACC) – actually sounded like. When I requested a recording, what came back via email was a recording of the constant hum of cooling fans required to keep the computers cool. This simple aesthetic inquiry revealed something profound about the material reality of computational systems: the vast amount of energy required to keep these machines running. The sound of those fans became a tangible reminder of the environmental impact of our work, influencing how we would approach using these resources going forward.Like my cartographic exploration of latent space, this sonic investigation made visible (or rather, audible) another hidden dimension of computational systems. We were likely the first research group to ask What does the VACC sound like? A seemingly simple question that opened up crucial considerations about sustainability in computational art and research. This exemplifies how artists working in technological spaces bring unique forms of attention to bear on these systems, revealing aspects that might remain hidden to purely technical and highly-specific investigations. Through both sound and image, through both digital and physical manifestations, my work seeks to map not just the mathematical territories of AI, but also its material, environmental, and human dimensions. 

Athena Dataset Sample  

Digital print on paper  42″ x 79″ 2022

Here is a sample of the two Athena Dataset branches that later trained StyleGAN models. The complete set is over 6000 files. Most of the examples here were created by hand and then digitized (DS). The orange boxes indicate synthetic data, those data samples created by our genetic algorithm (GA).