On Robots, Emotions, and the Enigma of Good Art Writing
Image generated by deepai.org with the prompt, “Art Writing Robot.”
[Image description: A Cubism-inspired image in blue, red, yellow, and black, featuring a vaguely mechanical-looking device with arms, coils, and a nib.]
by Justin Duyao
Here is an example of a thing a robot could never write: “A white onion / freshly washed. / Feeling of cold.” [1]
Here is another example, this time about art: “Something in the lens remembers its forgetting as someone stumbling through a forest with his hands outstretched,” as wrote Cole Swenson about Sally Mann’s photograph, Deep South, Untitled (Scarred Tree), 1998. [2]
I don’t think that sentence is impossible for robots to write because they don’t have hands. If you search “Robot” in Google Images, all you’ll find are images of robots with bodies, arms, faces, hands. I don’t think that sentence is fundamentally human because humans understand better than robots the sensation of stumbling through a forest. In fact, I’m willing to bet that a robot could find the words to describe that feeling faster than I could. The thing about art that robots don’t get is the thing that separates art from math.
Numbers are countable, in the same way that questions about numbers can be calculated and answered. But art exists on a different plane. And good art, like Olivia Laing wrote about Agnes Martin’s 2015 retrospective at Tate Modern, withholds, evades disclosure, declines to give itself away.
Laing went on:
“It isn’t easy to catch the workings of those paintings in words, since they were designed to dodge the burden of representation, to stymie the viewer in their incorrigible habit of searching for recognizable forms in the abstract field. They aren’t made to be read, but rather responded to, enigmatic triggers for a spontaneous upwelling of pure emotion.” [3]
In other words, it’s all the things we find ourselves unable to express about art that makes art so enigmatic. It’s the ways we become an onion in our minds, when we read a haiku about a freshly washed one, that separates humans from robots.
The problem with robots is their function is fundamentally computational. They can’t help but calculate, process data, and produce mathematically sound answers. ChatGPT, for example—the AI chatbot launched by OpenAI in November 2022 that has made the hair on copywriters’ necks stand straight up—can at best mimic human conversation, feeling, reaction, emotion, because for ChatGPT human emotion is mere data.
For humans, however, emotions are a wild and illusive frontier.
The Enigma of Emotion
Hyperallergic published an article last month titled “We Asked AI to Review Refik Anadol’s “Unsupervised” at MOMA.” In its response to the prompt, ChatGPT began, “As an AI language model, I do not have personal emotions or opinions. However, I can give a general review.” [4] And it did just that. One perfectly acceptable observation after another, ChatGPT leaned on words like “captivating” and “stimulating” and “pleasing” to describe Anadol’s AI-based work. It even offered a critique of the exhibition, arguing “it may be too esoteric for some viewers.” I have to admit, as somebody who knows next to nothing about the intersection between technology and art, I agree with ChatGPT.
At first glance, those descriptions of Anadol’s work feel oddly human. To be captivated, stimulated, and pleased, after all, are principally human faculties. To betray the wizard behind the curtain, however, ChatGPT wasn’t strictly the one to experience any of those things.
On February 15th, Alina Cohen described Refik Anadol’s “mesmerizing data paintings” as “captivating” in a review for Artsy. [5] On December 27, 2022, Emily Herbert described Anadol’s exhibition as “conceptually stimulating” on her blog. [6] Writers for Vulture, ArtNet, The New Yorker, and even Reddit described Anadol’s work as “crowd-pleasing.” I cannot say for certain whether it was those online sources that ChatGPT drew from to come up with its evaluation of Anadol’s work. But that is how the program works.
“There’s a saying that an infinite number of monkeys will eventually give you Shakespeare,” wrote Matthew Sag, an Emory law professor who studies the copyright implications associated with language models like ChatGPT. “There’s a large number of monkeys here, giving you things that are impressive—but there is intrinsically a difference between the way that humans produce language, and the way that large language models do it.” [7]
The unsurprising irony is that many of those latter reviews that repeated the descriptor “crowd-pleasing” read a lot like ChatGPT’s unfeeling amalgamation of words. Unfortunately, that’s part of the problem with contemporary art writing—so much of it really is robotic. The one writer whose review actually resonated with me was Jerry Saltz, who described the spectacle as a “glorified lava lamp.” [8]
In my mind, the best way to put to bed any fears that ChatGPT is going to take over the art writing world is to remind you (and myself, first) that the best art writing is as far from robotic as you can get.
“We must play with our food,” said Wayne Koestenbaum. “Producing language, we want, we eat, we regurgitate, we research, we demonstrate, we expel; with what has been expelled we repaper our bodily walls, and this wallpaper is intricate, befouled, and potentially asemic.” [9]
If I were to eat and regurgitate that sentence myself, I’d say that we can only really write after we’ve lived a little life. First, we have to devour the world. And then we can understand it enough to write about it.
What Writing Isn’t
That’s the thing about good writing. You can tell it’s good because it tells you itself. It doesn’t “show” you, it doesn’t parade across the page with all the right words in the right order. Rather, good writing speaks to you, unsettles you, raises the hair on your neck—for real. Good writing opens old wounds. Really good writing gashes new ones. It breaks all the rules AI has learned diligently to follow. And it, most importantly, reassures you, constantly, that there is more to life—more to art, more to love, more to pain—than you thought before you read it.
I don’t have any beef with robots. Not today, anyway. In the end, I defer to Adrienne Rich, whose anxiety about her own shortcomings as a writer I think better fit the robots voyaging into the blank page for the very first time: “and I have fears that you will cease to be / before your pen has glean’d your teeming brain.” [10]
Notes:
Haiku by Matsuo Bashō (1644-1694).
Cole Swensen, Art in Time, Nightboat Books (2021), 40.
Olivia Laing, Funny Weather, Norton (2020), 24.
Hrag Vartanian, “We Asked AI to Review Refik Anadol’s “Unsupervised” at MoMA,” Hyperallergic, February 2023.
Alina Cohen, “Refik Anadol’s Mesmerizing Data Paintings Are Captivating Audiences Worldwide,” Artsy, February 2023.
Emily Herbert, “MOMA MUST SEE,” eherblife.com, December 2022.
Sindhu Sundar, “If you still aren't sure what ChatGPT is, this is your guide to the viral chatbot that everyone is talking about,” Business Insider, March 2023.
Jerry Saltz, “MoMA’s Glorified Lava Lamp,” Vulture, February 2023.
Wayne Koestenbaum, Figure It Out, First Soft Skull (2020), 81.
Adrienne Rich, The Dream of a Common Language, Norton (1978), 15.