10. Art by Algorithm? Sol LeWitt's Legacy in the Age of AI
When the idea becomes the machine that makes the art
This week there have been a lot of social posts marveling at the leaps AI has made in image generation, seemingly overnight. Looking at ChatGPT's outputs from the simple prompt "Draw a bird," made just a year apart, and you have to agree; it's wacky how far these AI tools have come.
The jump from crude geometric sketches to photorealism is certainly impressive, but it's important to remember this isn't just the result of more compute. The ability of an image model to represent the image you're thinking of is also the result of their improved understanding of the assignment given; most people want something that looks like a lifelike bird, not a set of overlapping circles and triangles that approximate "bird." And that framework–one in which a work of art is generated by clear, repeatable instructions–always makes me think of Sol LeWitt.
Solomon "Sol" LeWitt (1928-2007) was an American artist born in Hartford, Connecticut, to Russian Jewish immigrant parents. After studying at Syracuse University and serving in the Korean War, he moved to New York in the 1950s where he initially worked in an Abstract Expressionist style while taking a night job at the Museum of Modern Art—a position that proved transformative when he met fellow artists Dan Flavin, Robert Ryman, and Robert Mangold, all of whom shared his growing frustration with painting's limitations. Influenced by Eadweard Muybridge's sequential photography and seeking to move beyond the emotional intensity of Abstract Expressionism, LeWitt began developing systematic approaches to art-making in the early 1960s, eventually articulating his revolutionary philosophy in his landmark 1967 essay "Paragraphs on Conceptual Art."
One of those paragraphs that has come to my mind frequently over the course of AI image development: "The idea becomes a machine that makes the art." It's a premise that feels particularly relevant as we think about our definitions of art and creativity in the age of AI.
LeWitt's vision of creativity was in many ways iconoclastic. It cut against the grain of revered, Western definitions of creativity that were built on Renaissance notions of artistic “genius.” For instance, one reason we value a Picasso painting more than a Picasso print is because the artist’s hand has a direct, physical connection with the painting.
LeWitt's approach to authorship was radically different. Consider what it means to "own" one of his works. When collectors acquire a LeWitt wall drawing, they don't receive a physical painting or drawing in any traditional sense. Instead, they purchase a certificate—a piece of paper containing precise written instructions for creating the work. Wall Drawing #65, for instance, reads simply: "Lines not short, not straight, crossing and touching, drawn at random using four colors (yellow, black, red, blue) uniformly dispersed with maximum density, covering the entire surface of the wall." The certificate grants the owner the right to execute this work, but the physical manifestation is always temporary, painted over when exhibitions end. The art lives in the instruction itself.
Pioneering a kind of conceptual art based on ideas rather than gestures, LeWitt wanted to hold space for the "productive ambiguity" that goes into the creative process. The resulting variations weren't bugs but beautiful accidents that honored both systematic logic and individual interpretation.
This connects to philosopher Nelson Goodman's crucial distinction between "autographic" and "allographic" arts in his influential 1968 work Languages of Art: An Approach to a Theory of Symbols. Autographic arts are those where only the artist's own hand can create authentic works—you cannot make a new Vermeer painting, however skillfully you might try to imitate his technique, because only Vermeer's specific touch, developed over decades of practice, could produce his distinctive style. The authenticity resides in the physical trace of the artist's body moving through space and time.
Allographic arts, by contrast, exist as instructions or scores that can be authentically reproduced by following the creator's parameters. A Mozart symphony lives not in any single performance but in the notation—the systematic description of relationships between notes, dynamics, and timing that any skilled orchestra can interpret. Each performance will be different, but all can be authentically "Mozart" if they follow his compositional instructions faithfully.
You could say LeWitt proposed a visual equivalent of musical notation, transforming the traditionally autographic medium of drawing into something closer to allographic art. His wall drawings exist primarily as certificates containing written instructions; the physical manifestations are authentic reproductions rather than unique originals.
As a graduate student, I vividly remember watching these instructions come to life in real time on a massive scale. During my time at Williams College and Clark Art Institute, the nearby Massachusetts Museum of Contemporary Art (MASS MoCA) began a collaboration with the LeWitt Estate and the Yale University Art Gallery to launch an installation of 105 large-scale wall drawings. Occupying more than 27,000 square feet of gallery space, this retrospective is big and bold, but also full of small playful ideas that reward looking closely.
Over sixty skilled artists and art students spent months executing LeWitt's instructions with an almost liturgical precision. Each had their own artistic voice, yet they willingly subordinated their personal preferences to serve his systematic vision. Yet even within this systematic approach, LeWitt preserved space for what we might call "human error"—the inevitable variations that emerge when people interpret instructions.
They stretched strings across vast gallery walls to establish perfectly straight reference lines, then used these guides to describe gentle curves with mathematical accuracy. The process was mesmerizing: meticulous taping to achieve razor-sharp edges between different washes of color, followed by the careful application of pigment with large natural sponges, moved in specific circular motions to create exactly the right tonal gradations across the wall's surface. The resulting works ranged from layers of straight lines meticulously drawn in black graphite, to sensuous drawings created by dozens of layers of transparent washes, to bold geometric forms that seemed to meld with the structure of the historic mill walls themselves.
Watching those artists at MASS MoCA revealed something crucial about LeWitt's process that distinguishes it from our current AI moment: the irreducible role of human interpretation even in allographic artistic modes. Contemporary AI is similar, of course. But that similarity sheds an even brighter light on the differences. Even if we prompt an AI model to give us different representations of a bird, it is constantly driving us towards the most statistically probable response, reducing variations and countercurrents into a bland sea of sameness.
This raises profound questions about the future of human creativity. If ideas become machines that make art, why can’t we design those machines in ways that preserve beautiful accidents where individual human judgment drives the creative process? LeWitt's genius wasn't in systematizing art-making per se but in designing systems that still required human interpretation—algorithms that needed the serendipities of manual labor to complete.
As we stand at the threshold of AI systems that can execute our creative instructions with unprecedented sophistication, LeWitt's work reminds us that the most profound art might always require something that resists optimization: the unpredictable, unquantifiable human elements that flow from lived experience. The future of creativity may lie not in perfecting our machines but in designing them to preserve space for the beautiful uncertainties that make us human.
Perhaps that's the model we need as we navigate our own relationship with AI: not replacement but collaboration, not perfection but productive ambiguity.
References:
LeWitt, Sol. "Paragraphs on Conceptual Art." Artforum, vol. 5, no. 10, June 1967, pp. 79-83. https://monoskop.org/images/3/3d/LeWitt_Sol_1967_1999_Paragraphs_on_Conceptual_Art.pdf
Goodman, Nelson. Languages of Art: An Approach to a Theory of Symbols. Indianapolis: Hackett Publishing Company, 1968.
https://massmoca.org/event/sol-lewitt-a-wall-drawing-retrospective/
https://www.nytimes.com/2008/12/05/arts/design/05lewi.html