Avery on Audry, 'Art in the Age of Machine Learning'
Sofian Audry. Art in the Age of Machine Learning. Cambridge: MIT Press, 2021. Illustrations. 214 pp. ISBN 978-0-262-36709-7; $45.00 (cloth), ISBN 978-0-262-04618-3.
Reviewed by Kayle R. Avery (University of Delaware) Published on H-Sci-Med-Tech (March, 2023) Commissioned by Penelope K. Hardy (University of Wisconsin-La Crosse)
Printable Version: https://www.h-net.org/reviews/showpdf.php?id=58148
In 2022, the Colorado State Fair attracted national headlines when an image generated by artificial intelligence (AI) won first place in the emerging artist section of its “digital arts/digitally-manipulated photography” competition. Théâtre D'opéra Spatial was created by Jason M. Allen, a game designer from West Pueblo, using Midjourney—a mass-market AI similar to DALL-E and Stable Diffusion. The generated image depicts several robed figures with dark hair gazing into a large glowing portal that contains a landscape of jumbled rectilinear forms along a sloping hillside. From the portal, a soft white light illuminates the onlooking figures and the perplexing architecture surrounding them. The vaulted ceiling above contains splotchy sections of gilded geometric ornamentation and a single line of looping forms that resembles calligraphy. The structure’s shadowy ribs veer in surprising directions, terminating suddenly into columns of smeared pixels and scratchy sections of digital noise.
The image sparked a heated debate over the merit of so-called AI art on social media. While a vocal minority expressed enthusiasm, many attacked AI art for the danger it posed to digital artists and their communities. While some artists already struggle to make a living online, they argued that AI’s ability to rapidly prototype and mass-produce digital artwork threatens artists in largely aspirational positions at large corporations. For these users, fears of human obsolescence overshadow the tech sector’s yearning for a so-called Fourth Industrial Revolution. In response to the controversy, some users offered in-depth critique of the image itself. Jay Dragon, a tabletop game designer from Philadelphia, cited Edward Said in a Twitter thread that drew attention to the image’s striking formal relationship with both romantic European paintings and those in the French Orientalist style. Not unlike the late eighteenth- and nineteenth-century European paintings depicting North Africa, Egypt, and the Levant, Dragon argued that Midjourney similarly “dazzled” with intricate detail and other-worldly imagery that, in effect, reproduced a racist visual rhetoric by incorporating the same “orientalist symbolic language.”
Art in the Age of Machine Learning, by Sofian Audry, offers an excellent toolkit for analyzing works of art like Allen’s award-winning piece. The slate of free and inexpensive AI programs that recently hit the market rely on advances in the science of machine learning. Inspired by the structure of the human brain, machine learning is a method for teaching computers to learn using rigorous training algorithms (training), layers of interconnected neurons (models), and enormous amounts of information (data). Each of these three main components of machine learning and their histories comprise the book’s primary sections. Therein, Audry explores the aesthetics of art made in conjunction with AI, paying particular attention to how the technology’s inner workings and materiality shape the artistic practice. As an artist trained in computer science, who also develops machine-learning systems, Audry offers a refreshingly literal approach to understanding AI artistry.
“These simulations of perceptive systems follow the rule of the average,” Audry explains, “compressing the outliers and making them disappear in favor of the norm, and are strongly biased” (p. 114). With mass-market appeal in mind, programs like Midjourney target center mass in extremely large datasets. After scraping billions of images from the web, Midjourney trained an AI system to associate words with their most commonly associated visual representations. While the company sourced some of its images from the public domain—problematic historical works included—the bulk came from public image hosts like Deviantart and ArtStation, which it processed without artists' consent. A significant portion of these images were digital paintings—a genre known for depicting otherwordly interiors and awe-inspiring fantasy and science fiction landscapes. This "concept" art, as it is often called, invokes the sublime by highlighting the vastness of a scene through low perspectives, infinite horizons, and meticulously crafted details. These images essentially adopt and repurpose the same formal strategies that were once used to socially construct the West alongside the non-Western other. While these digital paintings may have once offered a subversive context for historically pervasive visual tropes, that nuance is lost in processing. Midjourney flattens countless works together in order to extract their visual similarity. As a result, the “orientalist symbolic language” compiled by Midjourney ultimately becomes inseparable from its racist historic roots.
Allen used Midjourney as intended. He fed a series of prompts to Midjourney’s bot on Discord, a popular chat platform, and tweaked the results using image-editing software. In his reproduction of the already overrepresented, Allen’s piece drew attention to AI’s heavy reliance on pervasive systems of visual representation. However, the piece itself, and the publicity it generated, served Midjourney by promoting it to a much wider audience. In many ways, the entire process mirrored a previous attempt by a multinational corporation to popularize art made using machine-learning algorithms, Google’s release of DeepDream in 2015. In reference to the psychedelic animations produced by DeepDream, Audry argues, “insofar as artistic work produced using DeepDream algorithms is easily recognizable as such, any artistic work created through use of these algorithms immediately perpetuates and refers to its originators, hence inevitably contributing to Google’s social media campaign” (p. 114).
Although useful for making sense of images generated using “off-the-shelf systems,” these are not the book’s primary concern. Rather, the book draws into focus the complex relationship between human artists and learning machines that act together to generate meaningful works within an emerging art form. The art considered often highlights an intriguing “tension between the system’s autonomy and the artist’s control,” and rarely are these works relegated to the digital image alone (p. 77). Poetry, performance, music, and large-scale installations convey the expansive and largely untapped potential of artistic expression through machine-learning technology. For Audry, traditional researchers seek market value in their development of AI systems. As Audry often repeats, artists, on the other hand, “are more interested in processes; they attribute more importance to the creative context; and they seek the unexpected rather than the average” (p. 121).
The book serves as a crucial first step for researchers interested in machine learning. If the early adopters are to be believed, AI stands to fundamentally reshape labor as we know it. Even if the heavy promotion of AI resembles the rise and fall of crypto currency, non-fungible tokens, and virtual reality, corporations have already begun heavily investing in and aggressively implementing AI as part of a broader effort to cut costs, increase efficiency, and gain an advantage over their competitors. At this critical juncture, art not only draws attention to the technology’s implications but also makes evident the underlying structures it works to reproduce. The book should prove useful for artists seriously engaging with the medium and computer scientists looking to explore its potential for meaning-making. Perhaps most of all, curators and art historians will find the book a vital companion for the assessment and successful display of works made in conjunction with AI.
. Hey Kid (aka Jay Dragon) (@jdragsky), "There's been a lot of conversation around 'Théâtre D'opéra Spatial,'" Twitter, September 4, 2022, 7:22 p.m., https://twitter.com/jdragsky/status/1566567421955244033.
Citation: Kayle R. Avery. Review of Audry, Sofian, Art in the Age of Machine Learning. H-Sci-Med-Tech, H-Net Reviews. March, 2023. URL: https://www.h-net.org/reviews/showrev.php?id=58148This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 United States License.