Alexander Reben’s AI-Generated Art

by Mikko Lautamo

Through the alchemy of bronze, I seek to capture the fluidity of human existence.  Each statue stands as a sentinel, a unique embodiment of our shared experiences and vulnerabilities.  I tread the boundary between the abstract and the tangible, inviting viewers to embrace the complexity of their own identity. — Corebella Torrers

Corebella Torrers, an Argentinian sculptor active in the 1980s, does not exist.  She is a fiction hallucinated by an AI text generator.  The quote above and many like it represent the bulk of the work on view at Alexander Reben’s AI AM I? at the Crocker Art Museum.  Reben uses these completely fictional descriptions to create real artwork that follows from the AI’s logorrhea.  The results, which span painting, sculpture, photography, textiles and video, are mostly vapid and generic but plausible.  In the end, I did not leave as impressed by how advanced machine-generated art has become as much as I am embarrassed that most human-made art is not much better.

The show, organized by Crocker curators Francesca Wilmott and Scott Shields, features a wide range of machine intelligence: there are robots with social skills, generative algorithms for painting random colors and machines that play with toys so we don’t have to, but the primary “AI” involved is the kind of generative artificial intelligence that can invent quotes like the one above or create images from written prompts.  When NFT markets collapsed in 2022, the gold rush of generative AI was right on its heels – another art fad that promised a revolution.  However, unlike NFTs, AI has some practical applications, so how does it work, and what is it good for?

We apply AI to all kinds of machine intelligence, but the hot new AIs are large language models (LLMs).  An LLM is “trained” on a data set – for art, imagine the night sky full of stars; every star is a human-made work of art that the LLM has been shown.  Using a statistical model, the AI can fill in the night sky, so to speak, predicting what a new artwork could look like if it were a mix of one, two or 100 nearby artworks in the same constellation.  It is not that different from the predictive text on your phone, though hopelessly more complex.  LLMs like DALL-E, Midjourney and Stable Diffusion all need galaxies of human-made work to function well, which is why they are under fire for infringing on artists’ rights.

Reben’s work evades some of the thornier moral aspects of LLMs through a clever inversion: most of what we see on view was completed by humans. For example, artisans fabricated the Sentinel of Memory in the Valley of Vulnerability in bronze from a 3D print.  It shows an amalgam of human body parts arranged in an unexpected way and topped by a bat-like ear.  It was created without intention by an AI and then selected by Reben for fabrication and inclusion in this show.  A reproduction oil painter, working from an AI image, completed Dreams of the Cheese-Faced Gentleman.  It looks like a magazine ad riff on Magritte.  The Council of Foxes is, bar none, the most vacuous image I have seen in a museum: anthropomorphic woodland creatures in suits stare blankly at the viewer, all rendered in a soulless, synthetic naturalism.  I hesitate to describe other works because there is so little there there.

An LLM absorbs imagery by reading tags associated with images until it can reliably predict the salient feature described by the tag.  For instance, there may be thousands of images in its training diet tagged “blue hat,” and by ingesting enough of them, it learns, more or less, what a blue hat is.  The process is imperfect, and often, the mistakes are more compelling than what it gets right, but with enough data, the

signal can be reliably separated from the noise.  A “hallucination” is when an AI gets something obviously wrong, like too many fingers on a hand or too many arms on a torso.  These arise because the model predicts what comes next in an image based on probability, not knowledge: four times out of five when an image has a finger, there is another finger (this is an oversimplification, but it illustrates the problem).  The

images AI creates are often uncanny but usually not inspired.  The final act of input for a human operator is curation – at this point, a person still needs to separate the wheat from the chaff.  You can do this yourself with the voice-to-prompt machine on view.  It will turn your request into four images created by an earlier version of DALL-E.  In my experience, these images were rife with flaws and not within a mile of what I wanted, but that is why prompting AIs has become its own skill.  There are sparks of brilliance in some of the work Reben has selected for inclusion, but the work he designs with his signature humor is far better.

The best work is full of errors: flaws that Reben had to select for or run with to complete the work.  A Short History of Plungers and Other Things That Go Plunge in the Night is a faux-conceptual installation arising from a mistake in the generated text.  It reads, “The sculpture contains a plunger, a toilet plunger, a plunger, a plunger, a plunger, a plunger, and a plunger, each of which has been modified.” Errors of this sort are common in earlier predictive text.  That is because rare words like “plunger” have so few likely words to go

with it that the machine repeats itself.  Here, Reben decided each plunger was a separate object and sawed off the handles of six store-bought plungers to increasingly short lengths, presenting them, tongue-in-cheek, with Kosuth-like profundity.  The AI didn’t write this joke; Reben is laughing with us at its expense.

Other examples of weird delights include the ancient AI of Deeply Artificial Trees.  Completed in 2017, this reversed video of painter Bob Ross was processed with a convoluted neural network like DeepDream.  In plain English, an automatic process distorts video frames into recognizable patterns, like photos of pets or faces.  The effect is to turn the quaint, inoffensive TV painting show into a nightmarish, technicolor landscape of centipede bodies, floating eyes and psychedelic cat pictures.  Next to the clarity of images made with newer LLM-based AIs, these earlier attempts look elementary – but it’s been less than a decade since DeepDream was state-of-the-art.  AI AM I?  work shows how artificial art has advanced from a toddler tripping over its laces to an Olympic runner in an incredibly brief time.  Seeing all this work at once is disconcerting.  At the entrance to the main gallery, a retro-inspired rotary sign displays fake and snarky comments about the show that AI generates.  Looking at one, I wondered, “Is this gallows humor?” These machines are improving; they can make up nonsense for artist statements better than most undergrads, as the quote opening this review demonstrates. 

Deception is a recurring theme with these technologies.  An LLM is no more an artist than a blender is a chef.  It is just good enough to produce images that play on our perception and imagination to convince most people there is more going on inside the black box than there is.  Blabdroid, a small robot Reben helped engineer during his time at MIT, is a cute-voice knock-off of Disney’s robotic character, WALL-E.  It is designed to get itself stuck on furniture so it appears helpless; it then asks people to give it information and perform tasks beyond its ability, like climbing stairs.  Reben’s word for this is “symbiosis,” but it might also be called exploitation.  OpenAI, creator of the DALL-E image generation tool, started as a non-profit, had a board vote go the wrong way, suffered mass resignations, and then sold itself to Microsoft.  Its technologies are likely illegal, having been created with the old Silicon Valley ethos of move fast and break things.  What it broke were artists’ rights to their work.  Alexander Reben is OpenAI’s first artist-in-residence, something tech companies have started doing for PR.  Whether his tenure there will be symbiosis or exploitation remains to be seen, but the moral questions raised by how LLMs gather data are not addressed by this show.  Instead, we are left with a lot of Reben’s humor and an open question about what artificial intelligence has become.

#  #  #

Alexander Reben: “AI AM I?” @ the Crocker Art Museum through April 28, 2024.  

About the author: Mikko Lautamo is an artist and educator from Sacramento.  Their work uses computer code to create interactive and never-repeating installations centering on blended biological, social, and economic systems.  They teach art and art history at Cosumnes River College and have exhibited work in the United States, Europe, Australia and online.  Their work can be viewed on Vimeo.