Computer vision has been around since 1958, when Frank Rosenblatt, then a researcher at Cornell Aeronautical Laboratory, unveiled the Perceptron, which programed a camera to detect the location of a shape on a flash card. Yet, after an initial period of excitement, this landmark event in the history of artificial intelligence was followed by what’s known as the “AI Winter,” a period when most computer scientists’ hopes of developing machine learning past the rudiments went into hibernation.
It was not until 50 years later that computer vision began to advance to the next level when Fei-Fei Li, a professor at Princeton, hired thousands of low-paid, anonymous laborers through Amazon’s Mechanical Turk to tag three million images, thus creating a dataset large enough to train image recognition models.
The extended caesura in this seemingly long history, then, goes some way to explaining why book-length studies of artificial intelligence as it relates to the production and analysis of art remain thin on the ground. However, there is a rich body of literature by sociologists, data journalists, and scholars in STS (science and technology studies) that, in touching on such topics as surveillance and data visualization, is useful to thinking about the relationship between AI and visual culture writ large. The following is a list of studies drawn from a broad range of disciplines that may be helpful to anyone hoping to get a handle on this subject. Seven key terms provide the bullet points for this condensed curriculum: bias, extraction, augmentation, operation, the so-called, categorization, and power.
This article is part of our latest digital issue, AI and the Art World. Follow along for more stories throughout this week and next.
-
BIAS: More Than a Glitch: Confronting Race, Gender, and Ability Bias in Tech, by Meredith Broussard
Image Credit: Amazon Meredith Broussard is one among a number of important scholars and public intellectuals—including Cathy O’Neil, Safiya Umoja Noble, Ruha Benjamin, and Virginia Eubanks, any of whose books could have made this list—whose research has drawn wide attention to the biased foundations and contemporary applications of modern computing technology and machine learning in particular. In her most recent book, a follow-up to her excellent Artificial Unintelligence, Broussard tackles technochauvinism, the belief that computational solutions are superior to all other forms of problem solving.
While this is clearly not a study devoted to art, running throughout the book is the important, polar theme of visibility and invisibility that will be of interest to anyone who studies visual culture. Through an array of case studies Broussard shows how algorithmic systems both reduce to invisibility and raise to hypervisibility those people who fall outside the tech industry’s norm of a white, able-bodied male, offering an unflinching look at such problems as algorithmic analysis of medical imaging, AI-based surveillance, and above all the cocksure insistence that machines are always the answer.
-
EXTRACTION: Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence by Kate Crawford
Image Credit: Amazon If I could recommend only one book to someone interested in learning about the impact of incorporating artificial intelligence into the infrastructure of modern life (as is happening at breakneck speed), this would be it. The narrative of the book follows the life cycle of the hardware and software necessary to machine learning, from mining the minerals required to build the computers on which AI runs to creating, processing, and then making decisions with the data humans supply it (both with and without our permission).
The book’s thesis is crystal clear: It’s extraction all the way down. But what Kate Crawford reveals is about far more than environmental devastation and human exploitation. Having partnered with artist Trevor Paglen to create the landmark installation ImageNet Roulette (2019), and steeped in scholarship on visual culture, Crawford is well attuned to art’s entanglements with emerging algorithmic regimes and in particular how, when AI models are trained on images that must be translated into code to be machine-readable, these training sets become “classification engines” that establish new, deceptive “truths.”
-
AUGMENTATION: Machine Vision: How Algorithms Are Changing the Way We See the World by Jill Walker Rettberg
Image Credit: Amazon Offering a historical perspective, Jill Walker Rettberg situates machine vision within 8,000 years of technology devised to augment or alter the way humans see, beginning with the earliest mirrors made of polished obsidian. Assemblage, a key term in Rettberg’s analysis with roots in the work of Deleuze and Guattari and subsequent scholars of post-humanism, describes “relationships and shared agency” distributed across people, objects, institutions, and systems.
One could quibble with Rettberg’s decision to incorporate machine vision into a genealogy of technologies that augment sight, uniting into one story line optical encounters in the world with data production and synthesis. Nevertheless, her book is a necessary corrective to overhyped accounts of machine vision’s unprecedented nature. And it’s an adroit demonstration of how humans work with tools to produce ways of seeing, both empowering and disempowering who gets seen, and how, based on the affordances of those tools and what we do with them.
-
OPERATION: Operational Images: From the Visual to the Invisual by Jussi Parikka
Image Credit: Amazon On page 65 of this study, Parikka thanks readers for their patience. It was a gratifying moment that alleviated some of my aggravation with the unnecessary density of theory, repetition, and arcane language of the book’s opening sections. This comment might undermine my recommendation, but it is just to inoculate would-be readers against the forbidding nature of the book’s early pages.
Inspired by the filmmaker Harun Farocki’s concept of the “operational image,” Parikka’s study considers “how images are operated upon and become operationalized through aggregation, algorithmic analysis, and the ensuing questions of data-driven mobilization of the mass image.” His work occupies much the same territory as Rettberg’s study, citing many of the same influences, but with a key difference.
Using scholars Adrian MacKenzie’s and Anna Munster’s term the invisual, Parikka understands images mediated by computer technologies not as images but instead as configurations of data and “statistical distributions of patterns” with a graphical interface. This is an essential point that more people need to understand. In starting with it, Parikka offers a valuable account of what happens when images are homogenized as data and put to work to reshape the world.
-
THE SO-CALLED: AI Art: Machine Visions and Warped Dreams by Joanna Zylinska
Image Credit: Amazon If we were to take a capacious approach to computer-generated art, we might venture back to the 19th century, when Jacquard programmed its looms with punch cards. Such an approach would, however, necessitate an entirely new syllabus, overflowing with the work of artists, curators, critics, media scholars, and art historians who have trained their focus on art and computational technology. While she acknowledges this lineage, Joanna Zylinska’s AI Art is refreshing in targeting the subject of its title, all the better to give the reader a pretty zippy account of what it is and does. Underpinning the book is what I came to think of generically as “the so-called,” that is, its concern with how “AI-driven” art requires us to think critically about so-called intelligence and so-called creativity.
As a number of other authors on this syllabus point out, it’s crucial not to take for granted terms whose origins and main function are commercial—like the “learning” in “machine learning” itself (coined for marketing purposes by Arthur Samuel in 1959 while at IBM). Zylinska does not retreat from what she admits is the tedious question of whether AI-driven art can be “creative”; nor does she simply turn the question on its head to ask how AI-driven art causes us to ask how humans can be creative.
Adopting a post-humanist stance, Zylinska queries the meaning of creativity for “the human-with-the-machine, or even, more radically, the human-as-a-machine.” Revelatory as it is when AI-driven art critiques AI itself and the inhumane labor practices and invasive protocols that make it possible, such critique is ultimately incapable of effecting change. Moving beyond critique, Zylinska concludes, “Intelligent work on artificial intelligence could therefore perhaps attempt to sever that link between the work of art and human vision, going beyond the mere aesthesis of human experience to open up the problem of the universe itself as sentient.”
-
CATEGORIZATION: Computational Formalism: Art History and Machine Learning by Amanda Wasielewski
Image Credit: Amazon Amanda Wasielewski’s book, which I reviewed for Art in America last year, is rare in directly examining the impact of artificial intelligence on the practice of art history and collecting. In my review, I took issue with Wasielewski’s framing. Nonetheless, its two body chapters bring together, as I put it, “the most skillfully limned assessments of [machine learning’s] functionalities and limitations when applied to works of art to date.” Art historians, collectors, and curators who are interested in the potentials of incorporating machine learning into their practice will find in these chapters an insightful guide to its promises and, mostly, its perils.
-
POWER: Techniques of the Observer: On Vision and Modernity in the Nineteenth Century by Jonathan Crary
Image Credit: Barnes and Noble A book written during the AI winter, Techniques of the Observer may seem like an odd choice for this list. But if this classic cannot stand on its own merits for its relevance to our 21st-century reality, then Jonathan Crary’s more recent manifestos, 24/7: Late Capitalism and the Ends of Sleep (Verso, 2014) and Scorched Earth: Beyond the Digital Age to a Post-Capitalist World (Verso, 2022), show that his first monograph was just an opening salvo in a career-long program to call attention to how modern institutions colonize vision, constructing the observer as both the subject and object of control.
With opening questions like “How is the body, including the observing body, becoming a component of new machines, economies, apparatuses?” and “In what ways is subjectivity becoming a precarious condition of interface between rationalized systems of exchange and networks of information?” it more than deserves the designation of prescient. Reading it more than three decades after its original publication is a bracing—and vital—experience.
-
EXTRA CREDIT: The Hundred-Page Machine Learning Book by Andriy Burkov
Image Credit: Amazon Want to learn how machine learning works? Read Andriy Burkov’s The Hundred-Page Machine Learning Book. Even the most innumerate of us (among whom I count myself) will come away with a basic understanding of its principles.
