August 17, 2023
by Lynnell Edwards, poetry faculty and associate programs director
Poetry is having a moment. Yes, yes, we’ve heard this before—usually during National Poetry Month in April. Or the inauguration of a president or the selection of a new poet laureate. But in this case the moment is coming from an unlikely source: the meteoric emergence of artificial intelligence large language models and the remarkable text they can generate.
For doomsayers, the future is clear: AI will destroy humanity.
Champions of these new technologies, however, claim that they will free us from the often tedious work of rote reporting and summarizing information and will enhance human capacity to synthesize knowledge and thus find better solutions to everyday and extraordinary problems more quickly and authoritatively. Huzzah! But among the most curious and perhaps frequently noted claims from these same cheerleaders is that AI can also . . . write poetry!!!
In an article titled “Our New Promethean Moment” in the New York Times, pundit Thomas Friedman expressed astonishment at the abcedarian poem generated by AI during a private demonstration with the former chief research and strategy officer at Microsoft. The poem, in his words, exhibited nothing less than “stunning creativity.”
An abcedarian! So complicated! So nuanced! You can be your own judge of nuance in the first few lines here.
Satya Nadella, CEO of Microsoft, among his many claims and apologies for AI on a recent episode of the podcast Freakonomics, also reserved his highest praise for AI’s apparent ability to write and translate poetry. Asked by host Stephen Dubner about his favorite uses of AI, Nadella cited his own experience wanting to read Rumi translated into his native Urdu language from the originals in Persian. He discovered that AI could perform exactly those kinds of translations, concluding “. . . the most interesting thing about that is it captures the depth of poetry.”
Nadella goes on to say some interesting things about potential uses of AI, but perhaps he’s also out of his depth here: “But I must say, you know, going and asking ChatGPT or Bing Chat to summarize Heidegger is the best way to read Heidegger.”
Hmmmm. Maybe we should just sit with that for a minute.
Forgive me if I think these apologists’ examples of AI-generated poetry (Abcedarians! Rumi!) feel like a desperate defense of AI’s claims to nuance and humanity. In fact, it seems more like an unusually impressive party trick trotted out like a precocious child to perform for the adults before being sent off to bed.
But the claims continue, suggesting AI’s capacity to write poetry is somehow evidence of its humanity. Publications no less auspicious than the Washington Post, The Guardian, The New Yorker, The Atlantic, and research from MIT Press have all specifically examined the potential and actual poetic output from AI.
When did poetry become the new Turing test?
The conclusions are various as to whether the poetic texts generated by AI are interesting or compelling as poems. And no doubt this will be one more challenge for high school and college teachers of expository and creative writing. But the sheer volume of arguments discussing whether or not AI can write actual, real, nuanced poetry suggests something larger both about the debate and the current state of poetry.
On the one hand these repeated proofs that AI is now smart enough to write poetry feels like an opportunity for the tech-bros and political pundits to flash their humanities creds: “See! I love poetry! I’m not just trying to monetize your data!”
But of course they are. And as arguments about writers’ rights have made clear—most concretely by the Writers Guild of America strike and demands for AI transparency—it is exactly the words we writers have written that are being reassembled as (free) content.
On the other, perhaps AI-generated text is simply a new medium—like photography or fresco painting—that could open up how and why we make poetry. Poets find language everywhere. And AI can certainly find lots of words and quickly. Arguably, the text generated by AI in response to a prompt is a kind of “found poetry” (see William Carlos Williams’ famous note on the icebox about having eaten all plums “so sweet / and so cold” ) that the poet then manipulates to her own delight. More recently, Lillian-Yvonne Bertram’s 2019 collection Travesty Generator is doing exactly that by using computational programming code to enact and reveal patterns of racism.
Certainly we poets have always argued for the value of poetry, and with bombastic claims such as Percy Bysshe Shelley’s assertion that “poets are the unacknowledged legislators of the world” and from William Carlos Williams that
It is difficult to get the news from poems
yet men die miserably every day for lack of what is found there.
That Silicon Valley has suddenly taken an interest in persuading the rest of the world that writing poetry is the unique signifier of a machine’s humanity perhaps should feel validating.
Except that it doesn’t.
AI poetry “generators” are already writing poetry and some of it will no doubt find a home fulfilling the poetry assignment for the creative writing unit in sophomore English. Some of it will make a beloved’s heart swell on Valentine’s Day. Some of it may be published in a literary journal, disguised as “real” poetry by a human author, thereby igniting a scandal which will amuse the general public and further balkanize the actual poetry community.
It’s not, however, going to put anyone out of a job, because poetry has been the one kind of content that has remained, since the mid-twentieth century, stubbornly resistant to general monetization. As I heard US Poet Laureate Ada Limon state in a forum on Spalding University’s campus last fall: “Poetry is not content.” And it won’t take content developers long to figure that out. This brief AI moment that poetry is enjoying seems to be entirely in service to something other than the love, appreciation, and experience of poetry. And for those of us who treasure poetry and its gifts, we need to acknowledge that truth and persist in our usual, if unacknowledged, work of singing the world’s songs.
Lynnell Edwards is faculty in poetry with the Naslund-Mann Graduate School of Writing, where she is also book reviews editor of Good River Review. Her fifth collection of poetry, The Bearable Slant of Light, documents the effects of mental illness on the family and in our wider, anxious culture and will be published by Red Hen Press in Spring 2024.