Talk about technology’s encroachment into the creative industries is everywhere right now. But a year-long residency in robotic painting revealed to me just how inflated and imprecise the public conversation around A.I. in the arts has become. As an artist who has been using A.I. in making physical artworks for a decade, I thought I was fairly knowledgeable, but it has been a whiplash experience to reconcile the expansive claims about robotics in both the art world and tech industry with the reality of the technology in the lab. The public discourse is misleading—and we must get better at filtering the technological hype.

Early on, I noticed that even well-informed colleagues were conflating the term “A.I. painting”—used to describe algorithmically generated digital images—with traditional notions of painting as a physical, materially executed medium. This confusion is amplified by the way such images are labeled and discussed in public forums and by institutions: for example, Christie’s described the printed output of the A.I.-generated image Edmond de Belamy as a “painting,” despite it being a digital file transferred to canvas via printing. This blurring of terminology reflects how we increasingly consume art through digital interfaces, making distinctions between virtual and physical objects less immediately apparent. In physical settings, however—where robotic painting arms might be involved—the material differences reassert themselves.

Robots are the body to A.I.’s brain, so to speak, lending the ability to manipulate physical materials. This “embodiment,” as it is termed, is the subject of the Embodied Agents of Contemporary Visual Arts (EACVA) research group spanning Goldsmiths University in London and the University of Konstanz in southern Germany. When I began the residency there in May 2024, I had no idea how good—or bad—the robots would be. Could they paint as well as me? Better?

Even the most evangelical robot promoters wouldn’t yet claim that the robot walks into the art supplies store to purchase supplies, even if it does have a humanoid form. But what about once the materials are in the robotic lab? Do you imagine robots stretching and priming their own canvases? Mixing their own paints? Getting into the details of the painterly process is how we discern technological theater—essentially, robotic performance art—from actual robotic painting, or perhaps it would be more accurate to say robot-assisted painting.

Painting requires almost infinite flexibility and nuance in movement, and utilizes an array of sensory feedback at extreme levels of sensitivity. During my residency, I’ve learned to marvel at the inherent complexity of tasks humans, and human artists, take for granted like holding a paintbrush, dipping it in paint, and applying it to a surface. Confronted with the limitations of the technology, I began to question what it means to say a robot painted a painting.

When we hear and read these stories, most of us probably imagine a robotic hand holding a brush and painting in the familiar human way. But if it is claimed that a robot painted a painting, how much of the actual painting process should we expect to have been completed by the robot itself? This is not an academic question or a matter of semantics: it’s about the mechanics of tech hype.

Robotic arms are advanced and efficient in a very limited set of circumstances, mostly revolving around tightly defined, repetitive movements. If you’ve seen impressive videos of robots completing complex tasks, the strong likelihood is that those were completely pre-programmed routines. In unscripted situations, the robot’s capacities are much more limited. Painting is by nature unscripted, which is why it’s so fascinating to push the boundaries of the technology by trying to paint with robots.

In the Visual Computing Department at the University of Konstanz, the e-David, as Professor Oliver Deussen’s team dubbed their robot, paints without the intervention of human or printing technologies on the canvas, using a camera to observe and adapt progress. I watched e-David attempt to fill a small empty spot on the canvas for a quarter of an hour, involving numerous iterations of visual feedback, calculation of brushstroke placement, and physical paint application. It was a task that would have taken me all of five seconds. This discrepancy does not stem from the robot being bad at its job—it’s just that these tasks are anything but trivial.

Generating a digital image is one thing, but materializing it in the physical world is a whole other matter. Translating an input image into a set of brushstrokes that can be executed by a robotic arm is complicated, involving layers of algorithmic processing from depth and color field to defining the borders of objects in the image—but the materialization of that image in paint introduces immeasurable variation and organic “error.”

To highlight just a couple of examples, if the paintbrush doesn’t pick up enough paint or the paint is too dry, the planned brushstrokes will not be executed accurately on the canvas, regardless of how well the digital simulation might match the input image. If the paint is too wet, there will be bleeding or dripping of the color, which might obscure or contaminate other brushstrokes. Or, if the pressure has bent the bristles on the brush, then even with visual feedback systems the robot will make repeated attempts to fill a spot, missing ever so slightly every time.

Artists have always engaged with the technologies of their time, and this has in turn shaped both public discourse around those technologies and their ongoing evolution. There is real value in integrating artists into scientific research. By working with computer scientist Michael Stroh this past year, the e-David robot’s abilities have improved substantially. The series of oil paintings, in which each new canvas is a version of the previous one, modified by A.I., and then painted alternately by the robot and then by me, captures the technological improvements and the way that the robotic and human painters influence each other. It exemplifies how errors or irregularities can become new features in the next piece in the series.

It is this human-robot collaboration that remains the most interesting. I am convinced that we must not frame the robot as the artist but understand it as an extension of the human artist—a tool akin to painting or drawing machines found throughout art history. This is far more interesting than vague assertions of “autonomous” robot painters. In my own work, I have found myself thinking completely differently about the physical act of painting, my own embodiment, and the ways in which machine processes and digital aesthetics are shaping contemporary painting.

A flashy statement like “the robot painted a painting” can obscure more than it describes. Is the “painting” digital (in other words, a simulation) or physical? If physical, then what was the scope of robotic creation and what assistance was provided by humans? When the role of human directors and assistants is sidelined or left out entirely, the assumption is that the robot executes all the steps involved in completing the task. This is a fairy tale. And it is risky: When these assumptions go unchallenged, it builds a perception that A.I. and robotics technologies are much more advanced, independent, and adaptable than they are. This has a very real impact on important public policy, shaping the future in real ways.

Robotic painting is fascinating precisely because it is so hard. There are important lessons for artists in that revelation alone. Embodiment is an immense challenge, and simplified narratives about painting robots flatten the artistic output into a curiosity. The true value of robotic painting lies not in technological spectacle, but in the complexity and collaboration that allows it to emerge.

Gretta Louw is a multinational artist whose practice spans textiles, painting, digital media, and installation. Her work investigates the tensions between sensuality and efficiency, craft and automation within the entangled interrelationships between culture, nature, and technology—in short, the technosphere and the biosphere. She lives and works between Munich, Germany, and Perth, Australia. The EACVA project is a collaboration between artists and academics at Goldsmiths University of London and Universität Konstanz, supported by the Arts and Humanities Research Council and the Deutsche Forschungsgemeinschaft.