A conversation with ChatGPT about poetry

I teach first year composition at Marian University in Indianapolis, Indiana. When I first heard of what ChatGPT can do, I adjusted my syllabus to state that using AI bots to write papers would be treated like plagiarism. 

It will probably be difficult for me, however, to tell whether essays are written by man or machine. Even more disturbing, looking at my future prospects as an academic, chatbots apparently not only have the ability to write papers, but to grade coursework. And as chatbots are already being used to teach conversational English, who is to say they won’t soon take the place of faculty at the university level? Who’s to say that literary and scholarly journals might soon be filling up with AI-generated content (if this isn’t happening already)?

This isn’t just a problem for academics. Consider the predicament of my daughter Naomi, a freshman majoring in graphic design at IU Bloomington. In light of the recent reports on ChatGPT, which was released into the world in November, 2022, she wonders how much demand will there be for her skills when she graduates.

Because the chatbot also writes advertising copy and code—in addition to fulfilling many graphic design tasks—many coders and copywriters will wonder about their own financial security. Likewise for journalists—those who remain after the wholesale devastation of this field by social media, that is. 

And artists.  Surely you can find artists who take exception to this year’s winning entry to the Colorado State Fair digital arts competition. The artist designed his piece using an artificial intelligence program.

I first encountered a product of ChatGPT in the form of a Facebook post by Maxwell Anderson, the former director of Newfields and the Dallas Museum of Art. In his Jan. 2 post, Anderson gave ChatGPT the following task: "Write an op-ed about how fewer art museum directors are today trained in art history and the long-term implications for the museum field.” 

After deliberating on the task for all of two seconds, ChatGPT spat out a well-structured essay that could easily pass as an op-ed in, say, an online journal geared towards arts administrators. It identified one problem in particular; the perceived need for museum directors to “build relationships and secure funding,” as directors are often required to have business skills. The ChatGPT bot concludes that the acquisition of business skills “should not come at the expense of a deep understanding of art history and the cultural contexts in which art is created.”

Anderson concludes his post with the following: “Seems that insightful commentary on this topic is most easily found using OpenAI rather than headhunters and search committees.”

Anderson’s use of the term “insightful” captured my attention. I’ve never heard, or read about, anyone using the words “insight” and “AI” in a sentence before, at least not outside the realm of science fiction novels. Furthermore, it’s a term that I’ve associated exclusively with human thought.

Anderson’s comments made me think of the Turing Test. 

In 1950 Alan Turing, who is widely credited as the founder of both computer science and artificial intelligence, proposed a test to judge a computer’s ability to think like a human. The proposed test uses a human evaluator to ask a computer and a human a set of questions, without knowing which is which. If the computer’s responses are indistinguishable from the human’s, then eureka! The computer passes the test.   

If machines can think for themselves, have feelings, and even offer genuine insight—and there is at least one computer engineer who believes they can—then I wonder what the future holds for my particular field of endeavor: creative writing. I love poetry: I love both reading it and writing it, and getting lost in language every morning when I wake up. But it’s hard to get around the notion that it’s an enormous timesuck. As I’m currently trying to put my daughter through college on a limited income, I’ve started seriously re-evaluating how I spend much of my time. By having ChatGPT write my poems, I could be saving an enormous amount of time, right?

Early in the morning on January 27, I gave into my curiosity. That is, I turned to ChatGPT to see what it could do. I wanted to see if it could write a better poem than I could. I had been working on a poem about finding Edward Gibbon’s classic The History of the Decline and Fall of the Roman Empire in the Goodwill where I used to work as a team lead. But I had run into a roadblock.  

Here’s what I wrote:

I find Edward Gibbon’s Decline and Fall of the Roman Empire
in the Nora Goodwill.
But on the next shelf it’s romance writer Nora Roberts
who dominates.
Her romance books don’t have the beefcake hunks
on their covers,
but she’s written 225 of them. What’s more,
 people read them,
which is more than you than you can say about— say—   
Jorie Graham.
Still, I wonder, if the number of people who’ve read
Decline and Fall 
since 1776 adds up to Robert’s audience.

In my best experiences, in writing poetry, I reach a moment of insight, or what seems like insight at the time.  In these moments, the meanings that the words on the page (or onscreen) convey stretch beyond the syntax. There is a mystery in the act of artistic creation, in birthing something new into the world.  But in this poem that was not happening, clearly.

The words on the page felt stale, dead, lacking not only insight but logic and coherence. A stillbirth.

Anyway, I had tried to toy with ChatGPT before, but I was unable to, as its servers were at capacity. (During these times, ChatGPT informed me of this fact by writing witty poems on its homescreen.) But this time around, I was able to give it a prompt.

That is, I asked ChatGPT to write me a poem about finding both Edward Gibbon and Nora Roberts in a Goodwill store.  I pressed “enter” and waited.  It didn’t return an answer right away, so I went to brush my teeth. (I had to get ready to go to Marian University as I was teaching that morning.) When I returned, I found this poem on my screen:  

“In Goodwill, a treasure trove of tales,
Where stories old and new prevail,
I found among the dusty stacks,
Books by Edward Gibbon and Nora Roberts, too, in packs.

Gibbon's "Decline and Fall" I spied,
A classic work, well-preserved inside,
Of Rome's decline, a grand narrative,
A history lesson, truly informative.

Roberts' romance novels, too, I found,
With covers bright, they stood out 'mongst the crowd,
Of love and passion, they speak so well,
A true escape, a literary spell.

Two authors, vastly different in style,
Yet both with stories worth the while,
In Goodwill, a chance encounter,
A literary treasure, now mine to ponder.”

While this poem lacked any great insight, it was serviceable enough. It would’ve been more serviceable, I think, in 1900 rather than in 2023. That is, it feels pre-modern not only in its use of inversion of noun and adjective (“covers bright,”), literalism, and formality, with a sing-songy rhyme scheme (“so well…/a literary spell.”) Clearly ChatGPT was relying more on the likes of Matthew Arnold and James Whitcomb Riley in its database for its modeling than the likes of John Ashbery and Ezra Pound. It leaned heavily on Victorian verse and stayed clear of the more obscurantist veins of modern and/or contemporary poetry. Considering that many contemporary poems read like dressed-up word salad, who could blame it? 

The chatbot’s reliance on form was less a revelation to me than a reminder of my difficulty in conceiving a poem outside the parameters of form. My college poetry professor Charles Martin, who has translated Ovid’s Metamorphoses and the poems of Catullus from Latin, was (and is still) big on form, told me as much in his critiques of my work decades ago.

That’s what I get, I guess, from taking a poetry course with a classicist.

My encounter with ChatGPT reminded me of this decades-old conversation, and the much larger conversation about the place(s) of free and formal verse in American poetics that has been going on for the last 100 years, ever since T.S. Eliot unleashed the freeform The Wasteland into the world.  

As for me, I love writing in both free verse and in formal verse. But I particularly like the 10 syllable line (iambic pentameter) that mimics the patterns of speech in the English language, which puts me on, or close to, the border of the formalist camp.  

To cut to the chase: the next morning, I tried again to write (on my own) a draft of my Gibbon poem, choosing an unrhymed sonnet form, figuring that the tension of squeezing syntax into a tight iambic form, would get me achieve an orgasm, as it were, of insight:

I find Edward Gibbon’s Decline and Fall 
of the Roman Empire in the Nora Goodwill
but on the next shelf it’s romance writer 
Nora Roberts who dominates. Her romance 

novels don’t have barechested hunks on their covers,
but she’s written 225 of them. What’s more,
people read them, which is more than you can say
about another Goodwill find—books by Jorie Graham

who laments American civilization in poetry 
with the densest syntax imaginable. Roberts’ men 
aren’t poets. They don’t have the densest biceps either 

but they will always be young, will always make  
their lovers moan. Civilizations come and go
but the romance genre will never decline and fall. 

 I’m not sure I came upon any great insight with this poem, but at least I wrote something serviceable, something I could use in my blog.  I suppose I have ChatGPT to thank for this. Will I bring up this encounter in my ENG 112 class? Maybe some time in the future, when I can design an exercise that can safely promote learning and creativity. But right now, probably not. Doing so might just bring my students down too many rabbit holes.

Just speaking for myself, however, I’m looking forward to more such ChatGPT conversations. 

—Dan Grossman