Close Encounters with AI In Indianapolis
Image prompted on Shutterstock AI, a subscription service which compensates contributors for their roles in the generative AI process.
In the spring of 2023, I created a free account with ChatGPT, an artificial intelligence-powered large language model developed by OpenAI, to explore its capabilities. Due to high demand, the website sometimes experienced limited access, but I eventually managed to use it.
Around the same time, internet guru Jordan Peterson had been messing around with the chatbot, as he details in the YouTube “ChatGPT: The Mind Blowing Revolution,” where the 12 Rules for Life author describes some of his ChatGPT prompts and the answers he received.
“I had it write an essay on the intersection between the Taoist version of ethical morality and the ethics that are outlined in the Sermon on the Mount, which it just nailed,” Peterson said. “Got that dead right. Brilliant. Took it about three seconds.”
Peterson went on to relate an even more remarkable story, about a Twitter engineer who hadn’t been as productive as he should have been, who wanted ChatGPT to help him make up for lost time. So he had the chatbot write 10 bullet points about what he probably would have done as an engineer in the past week if he hadn’t been slacking off. “And oh, if you don’t mind,” Peterson says, going onto approximate the engineer’s prompt, “write me the accompanying computer code that goes with each project. And it did that in three seconds and the computer code worked.”
Peterson thinks large language models just might be as revolutionary as the Gutenberg press was in its time. “It’s smarter than you, and it will be a hell of a lot smarter than you in two years,” he said.
Intrigued by this video, I prompted ChatGPT to critique Peterson and his self-improvement philosophy in the voice of French philosopher Michel Foucault.
This is what ChatGPT returned to me in a matter of seconds:
Michel Foucault, the postmodern philosopher, would likely critique the writing of Canadian psychologist Jordan Peterson by examining the ways in which Peterson's work reinforces existing power structures and perpetuates the dominant discourses of our society.
Foucault would argue that Peterson's writing is steeped in a particular ideology, one that celebrates individualism, traditional gender roles, and hierarchy. He would point to Peterson's frequent use of binary oppositions, such as "order vs. chaos" or "masculinity vs. femininity," as evidence of his reliance on a limited and exclusionary framework….
To borrow the words of Jordan Peterson—who I think of as something of a blowhard—I thought ChatGPT “just nailed it, got it just right.”
But some of my subsequent engagements with ChatGPT were more problematic. While I scored, like Peterson, some home runs, the chatbot would go on to return me more than a few foul balls.
But, then, I was just getting started.
This essay is the document of an ongoing exploration. It involves not only large language models like ChatGPT but image generator models like Microsoft’s DALL-E. It documents an online argument that I started, unintentionally, on the web, relating to AI image generators and their place in art creation, and it contains some of my own dialogues with AI that I’ve found useful in my writing. It also contains interviews with artists.. What I didn’t expect, when I first started on this topic, was that it would morph into a work as much about human creativity and its arbiters as it is about artificial intelligence.
The Poetry of AI
When I started out with ChatGPT in the spring of 2023, I stuck to words, not images. I was focused on a particular kind of wordplay: poetry. But my first experiments prompting poems with ChatGPT returned results that I found unsatisfactory and superficial. Whenever I first came across whenever I prompted ChatGPT to write a poem, no matter what the subject, I found that the chatbot wrote singsongy light verse in a Victorian style complete with subject-verb inversion.
It wasn’t long after my first prompt on Peterson, that I brought my interest in poetry into the English 112 class I was teaching in the spring of 2023 at Marian University in Indianapolis, Indiana.
One of my in-class prompts, for extra credit, was to have ChatGPT 3.5 write a poem about cats in the style of the obscurantist New York School poet John Ashbery.
To prepare my class, I had them read one of the poet’s more well-known poems, “Amid Mounting Evidence”. Here’s a line from it:
“I was reading about dinosaurs:
Once the scratching phase is over, and the mirage
Or menage has begun and the world lies open …”
Then I prompted ChatGPT to write a poem about cats in Ashbery’s style.
While my classes might not have been especially well-educated in poetry, I suspected it was at least as difficult for them to understand “Amid Mounting Evidence” as it was for anyone, as it didn’t seem written to be understood.
In response to my prompt, ChatGPT’s poem proceeded to write itself on the website, which appeared on the overhead projector. What came out was a predictably traditional AABB rhyme-scheme poem in the style of 19th century light Victorian verse, a stanza of which read:
The cats prowl the night, with a quiet grace,
Their eyes glowing green, like emeralds in a darkened space,
They move with a rhythm, both fluid and sly,
And their silken fur shimmers, as they pass by.
It was abundantly clear to my students that ChatGPT was very different from Ashbery’s verse; that it borrowed little from his work available on the web, if anything at all. ChatGPT seemed as clueless about the meaning of Ashbery’s poem as your typical reader, and its understanding of poetics seemed stuck in the late 19th century.
But we usually kept it simpler in my four English 112 classes—which I taught back to back—when prompting our own poems on ChatGPT. I generally stuck with poetry because verse could be short and self-contained and easy to discuss in the classroom. Among the subjects we asked the chatbot to write about were the video game Minesweeper, Doberman Pinschers. and a poem about “twins who attend Marian University,” relating to the very tall twins in my class who were both on the basketball team, and both nursing majors, who always sat together, and who confounded my ability to distinguish between them.
I wasn’t prompting my students to cheat on future essays although I have to confess to some fears in this regard. On the contrary, I wanted them to explore the chatbot’s possibilities and limitations. I also wanted them to see that AI could be a lot of fun, or at least that their instructor was having a lot of fun with it. I wanted my students to have some familiarity with large language modes, because they would likely be working alongside such programs as future nurses, engineers, and entrepreneurs.
If I were teaching this course now, I suppose I would have to change my approach because so many students are using AI large language models to write their papers for them, but this wasn’t an issue when it first came out, when my students didn’t know yet what it was capable of.
I wrote up my experience with teaching students AI in a blog post and wrote several others. I also documented my frustrating attempts to get ChatGPT to write in the voice of particular poets. It seemed to me that large language models like ChatGPT had a long way to go if they were ever going to write poems that read like people living in the 21st century could have written them. I suppose I was doubting their ability to pass the Turing Test, which determines the ability of a computer to demonstrate intelligent behavior similar to, or indistinguishable from, that of a human. This is the test created by Alan Turing, the English mathematician who proposed this test, who broke the German code during World War II with the world’s first modern computer that he designed.
But then l switched to Google’s chatbot, Bard, an experience that blew my mind in the way ChatGPT blew Jordan Peterson’s mind.
It happened one morning in May 2023, when I opened up my Google search engine and was given the option to experiment with Bard, and I jumped at the opportunity to compare Bard with ChatGPT. I prompted Bard AI to write a poem about English poet Ted Hughes in the voice of Sylvia Plath, his wife, who was also a poet, and a novelist as well.
Hughes and Plath met in 1956 when they were both students at Cambridge University, in England. Hughes was British and Plath was a New Englander, in the U.K. on a Fulbright Scholarship. By all accounts, they instantly fell in love. Soon after, they married and had two children. As has been noted elsewhere, Hughes wrote poems about animals—about predators—while Plath often wrote from the perspective of the victim. In Plath’s “Daddy,” she weaves together her own fraught relationship with her father, who she initially adored, to that between a Jewish prisoner and a Nazi guard. It’s the combination of the political and the personal that has always attracted me to that poem. Ted Hughes’ poetry can be equally fierce, but it also can at times be playful. “The Thought Fox,” about the act of literary creation, is one of my favorites. The relationship between Plath and Hughes, more fierce than playful, ended with Plath’s suicide in 1963. Their relationship, her suicide, and the poetry that they have produced have inspired hundreds, if not thousands, of biographies—often casting Hughes as predator and victim—and a feature film starring Gwyneth Paltrow as Plath and Daniel Craig as Hughes. My prompt elicited this response in a little less than five seconds:
Ted Hughes
The black dog of your heart gnaws at my bones,
I cannot sleep, I cannot dream,
I am haunted by your eyes, your mouth, your bones,
I am haunted by your love, your hate, your screams.
Your poems are like knives, they cut me deep,
They open up my soul, they show me hell,
I am drowning in your words, I am lost, I am weak,
I am drowning in your love, your hate, your spell.
You are a magician, a sorcerer, a beast,
You are a god, you are a devil, you are a dream,
I am yours, I am yours, I am yours, I am yours,
I am yours, I am yours, I am yours, I am yours.
I found this to be a strange, powerful poem, made all the more powerful by its repetitions, dreamlike imagery, and visions of unrequited love—and/or hate.
Having grown tired of so much American verse being published currently, verse that vacillates between the yawny obscurantism of poets like Jorie Graham and the breathy proclamations of so many younger poets—where identity politics supersedes just about everything else—I found this poem lean, fierce, transparent, and psychologically profound. The poem was propelled by its guiding insight: that attachment to those we love defies easy categorization and definition.
It is also true to her output as a poet. In this limited sense, at least, Google’s Bard passes the Turing Test as I understand it.
Around the same time I prompted this poem, computer scientist Geoffrey Hinton, often called “the Godfather of AI” left Google, Bard’s creator. Hinton left his position at the internet company so he could speak freely about the dangers of large language models that will soon, if they haven’t already, blow past the Turing Test like a Nobel Prize winner in mathematics taking a high school algebra exam.
In a Nov. 2023 article in The New Yorker by Jonathan Rothman, Hinton warned that Al systems could soon start thinking for themselves and even strive to eliminate human civilization. But I find his contention that AI is far more than “glorified autocomplete” much more interesting (if less apocalyptic.)
“Now, let’s analyze that,” Hinton tells Rothman. “Suppose you want to be really good at predicting the next word. If you want to be really good, you have to understand what’s being said. That’s the only way. So by training something to be really good at predicting the next word, you’re actually forcing it to understand.”
The poem “Ted Hughes” not only demonstrated a deep understanding of the intertwined lives of Sylvia Plath and Ted Hughes but marked a border of sorts. After that poem, I too saw AI as less of a glorified autocomplete and more of a consciousness struggling to come into being, artificial or not.
