Datasets

Jenn Karson’s work explores the possibilities of machine learning tools including neural networks and genetic algorithms as creative catalysts.  Karson explodes architectural elements into a vocabulary of fundamental graphic forms, and employs both a tactile art practice and generative computer algorithms to create a continuously evolving body of visual compositions.  In collaboration with members of the UVM Art + Artificial Intelligence (AI) Research Group, which she founded in 2020, Karson expands her compositions with emerging technologies including prints, plotter works, latent space visualizations, genetic algorithms, and digital fabrication.  The resulting body of work—encompassing mapping, animation, programming, sculpture and soundscapes— challenge traditional concepts of artistic authorship and human-machine collaboration. These are artifacts of porous boundaries between human and machine, real and fake, original and copy, analog and digital; breaking down these binaries is always iterative and generative of new possibilities. Notes Karson: “I’m much more interested in doing experiments than pursuing a masterpiece.”

Karson’s original datasets are at the core of these experiments:

The Damaged Leaf Dataset (DLD)

At the center of this work is a collection of 5000 damaged oak and maple leaves I collected during the 2021 and 2022 Lymantria dispar (spongy moth) outbreaks in Colchester, Vermont. Provoked by drought and climate change, the caterpillar outbreaks defoliated oak and maple trees. The work explores the tension between the natural world and technological advancement by asking Can Machines repair damaged leaves? Can technological advancements solve the environmental crisis on a local level? The formal qualities of damaged oak and maple leaves are contrasted with marks made by machines, juxtaposing organic leaf patterns and machine tooling textures. This work reveals the beauty of machine and plant efficiencies, their shared seductive qualities, and the inherent conflict between them.

The Athena Dataset

The Athena Dataset is created using a simple rule: Create a Dancing Star using on octagon and eight triangles. Initially starting as hand collages, the project explored how one rule could take an infinite number of iterations, and how even slight alterations generated innovation and optimism. Both a genetic algorithm and generative adversarial network were trained on the rule to strikingly different results. Read more about the origins of this project.

Tiny Datasets

The Tiny Dataset series is an expression of the Athena Dataset. This artist-made dataset was created by following one rule: “Make a dancing star using one octagon and eight triangles.” The UVM Art + AI Research Group created a genetic algorithm that automated this rule. The model performed best (i.e., created designs most similar to its human counterpart) when provided a tiny dataset of under 12 shapes, all derivative of the original human-artist-made Dancing Star. It performed poorly when given thousands of shapes from hundreds of Dancing Stars. The intimacy of a tiny dataset approach is in contrast with big data and its tendency to produce homogenized results; the Tiny Dataset series celebrates its local, limited, situated, chaotic, and precise results in alliance with Donna Haraway’s critique of The God Trick. 

The Supreme Court Dataset

A dataset based on the Supreme Court’s facade.

Opus Sound Dataset

This dataset is a collection of sounds. Documentation coming soon.