The Machine Cannot See Itself

Hello {{name | Transient Family}},

This week we’re excited to share Mirror 404, a series of 50 point cloud artworks by David Lisser. Scanned over the course of two years, the pieces capture the breakdown which occurs when a camera scans a mirror, embodying the difficulty for a computer to see itself.

The artworks will be available today (September 17th) at 12 PM EDT. As interactive point clouds, the artworks are best explored directly on the exhibition page. Images are used to express the works below in the accompanying essay by Kevin Buist.

by Kevin Buist

The thing that every mirror has in common is that when you look into it you see yourself looking back out. Wherever you go, there you are–as the saying goes–and mirrors embed this truism into a shiny object. Mirrors are tools for self-perception, windows to our appearance and our soul, for good or ill. As machines become more intelligent, they also become more perceptive. The more AI advances in its ability to mimic human outputs (text, images, agentic multi-step reasoning), the more it masks its own inhuman, alien nature. To understand a machine’s way of being in the world requires us to find its edges, to bend its processes until they break. David Lisser’s new body of work, Mirror 404, does this with regard to machine vision, pushing photogrammetry to the brink in a simple and poetic way: by pointing the electronic eye at a mirror.

A tuft of verdant leaves and delicate white flowers cling to an array of free floating branches. At the center of this coarse bouquet is a small framed mirror, only a few inches across. This assemblage is not a still life in a traditional sense, rather it’s a digital 3D model created through a photogrammetry process that allows the viewer to change the angle and zoom in and out by clicking in their browser window. As we zoom in, we see that the plants and the mirror are not rendered as voluminous 3D forms, as they would be in a video game, but instead are made up of colored points floating over a black abyss. Each 3D vignette in Mirror 404 features a small mirror nestled in natural surroundings of various kinds. What’s most fascinating about these works is not the way the artist uses photogrammetry to capture these scenes, but rather the way the technology fails spectacularly in its attempt to capture the mirrors. The reflective surfaces are rendered as voids with white or blue edges with no discernible reflected images. In some cases there are bright points of reflected sky within the mirror frames that are erroneously placed far behind the mirror itself. In other cases the mirror frames are simply rendered as holes.

Lisser captures the data for these digital dioramas on walks with his dog where he brings along a bag of small mirrors and a DSLR camera. After nestling a mirror in leafy surroundings, he takes hundreds of photos from various angles. The images are later stitched together using photogrammetry software which finds common details between the photos, and uses triangulation to place those points in space, creating a 3D model of the scene. There’s a rule in photogrammetry which Lisser is deliberately breaking: you are not supposed to photograph reflective objects. Lisser is peeling apart the layers of this particular form of machine vision through creative misuse. Like a rock musician creating a squeal of feedback or Warhol overexposing screenprints, by revealing the limitations of tools we get a better sense of how they shape perception.

Mirrors appear throughout art history to serve various narrative, metaphorical, and perceptual functions. The myth of Narcissus staring at his own reflection has been depicted by Caravaggio, Dali, and many others. Jan Van Eyck’s Arnolfini Portrait (1434) famously features a convex mirror in the center of the picture, mounted on the wall behind the bride and groom who are standing for the portrait. The tiny distorted reflection shows the backs of the subjects as well as two figures who occupy the space of someone viewing the scene, one of them is presumed to be Van Eyck himself. Two centuries later Diego Velázquez painted Las Meninas (1656), a royal group portrait (and self portrait) where the artist can be seen working behind a large canvas next to a young princess and her entourage. On a mirror behind them we see King Philip IV and Queen Mariana. The position of the mirror means that we, the viewers, are seeing the scene from the same position as the king and queen. Or, if the mirror is angled slightly, we’re seeing the surface of the painting that Velázquez is working on. For both Van Eyck and Velázquez the inclusion of a mirror serves to deepen and destabilize the scenes depicted. In both cases, the mirrors give us a glimpse of the scene that exists on our side of the picture plane, providing a data point from which we can triangulate our own position. Like Lisser’s use of photogrammetry, these famous paintings extrapolate positions in 3D space using information provided by 2D images.

An even closer art historical parallel to Lisser’s Mirror 404 is Robert Smithson’s series of Mirror Displacement works, particularly the ones created in England and Yucatán, Mexico in 1969. Like Lisser, Smithson ventured into natural environments with mirrors and a camera, placed the mirrors in various settings, then photographed them. While Smithson’s photos were not stitched together to make 3D models of the scenes, he was thinking about the disruptive perceptual qualities of mirrors in a way that presages Lisser’s work. In an essay for Artforum in 1969 titled “Incidents of Mirror-Travel in the Yucatan,” Smithson writes, “Reflections fall onto the mirrors without logic, and in so doing invalidate every rational assertion. Inexpressible limits are on the other side of the incidents, and they will never be grasped.” Smithson’s pontificating could just as easily apply to the shortcomings of photogrammetry exposed by the mirrors in Lisser’s Mirror 404, when he says, “Why not reconstruct one’s inability to see? Let us give passing shape to the unconsolidated views that surround a work of art, and develop a type of ‘anti-vision’ or negative seeing.” In fact, it could be argued that Lisser’s work produces an “inability to see” even more clearly than Smithson’s, no pun intended.

Artists look at themselves and they create artworks that can lead us all to a deeper sense of self reflection. But we live in an age where our perception is augmented or even replaced by machines. Many of our tools are tools for self perception, our screens tempt us into stasis like Narcissus staring into the pool. Lisser’s work exposes the deficiencies of machine vision. Despite the fact that the artist and his camera appear in many of the source images that he shoots of his mirrors in the wild, he never ends up in the final 3D model. This is not by choice, it's simply that the photogrammetry software discards the artist along with the other points of light that appear in the mirror’s frame, unable to triangulate these chaotic bits of light into a reconstruction of the scene. The machine cannot see itself. Like AI chat bots that are sycophantic not because they’re nice, but because they’re reflecting back what the user wants to hear. Machines don’t have the presence of mind to use mirrors, they are mirrors, and imperfect ones at that.

For more information on new releases and behind-the-scenes looks into exhibitions, join the Curated newsletter below.

Sincerely,
Transient Team