Generative Adversarial Artworks

Generative Adversarial Artworks (2019 – 2020)
Multimedia Art Created Using AI

G.A.A. is a collection of computer generated artworks created using a Generative Adversarial Network or GAN that interrogates questions over authorship and attribution in AI-generated works as well as exploring the similarities and differences between evolutionary and fine art processes such as selection, attraction, avoidance, adaptation and mutation.

A GAN relies on two opposing neural networks to selectively combine and produce new images. This technology is often employed in the production of deep fakes which use artificial intelligence and deep learning methods to produce realistic (but false) images.

For this project, each artwork was created by feeding different combinations and percentages of images into the GAN, then curating the outputs according to personal aesthetic impulses. The source images depict biological structures (plants, animals, people), textures (fabrics, clothing, foods), and forms (architecture, objects) to explore a wide range of outputs.

In this work, I investigate: what processes or considerations including values are applied to the creation of a work of fine art either through production or curation / exhibition that mimic evolutionary processes of selection, adaptation and mutation? In the case of computer or machine-generated artworks, what roles do we assign to man vs. machine (or program) in relation to ownership and artist identity? To whom do we assign the role of artist, artist’s assistant, tool, curator? By selecting images rather than creating them, does the artist become merely a curator? Is the GAN the artist, the artist’s assistant, or merely a tool commensurate to acrylic paint? What of the data set which is used to generate the resulting images? Are the photographers behind those images willingly or unwillingly co-artists? In what ways should we consider attribution to be ethical or necessary when concerning computer-generated works?

The ingredients of each image as well as their percentage / weight in the total resulting image are listed in the caption below it. All images were generated between 2018 to present using ArtBreeder / BigGAN / ImageNet using a data set which is not expressly attributed.