In mid-February, there were around 737,000 Instagram posts that used the hashtag #AIArtist. Many adopting the hashtag could do so thanks to text-to-image platforms that need only a few keywords or a brief description to produce hyper-realistic images that can be (and have been) mistaken for photographs.
So how soon can we expect the rise of the #AIPoet? When put to the test, OpenAI’s large language model and chatbot ChatGPT gave us what it claimed was an “original” poem.
The morning dew upon the grass
Reflects the sun’s first light at last
A world reborn, refreshed and new
A gift of beauty, just for you
We encouraged ChatGPT to break away from rhyme schemes and attempt some blank verse instead. While the chatbot was adamant that it could do it with the right prompt, it repeatedly churned out poems with the ABCB or ABCC rhyme structure— no blank verse.
ChatGPT tries to write poetry without rhymes. | Photo Credit: Screenshot from ChatGPT
When asked why it couldn’t manage any blank verse, the chatbot admitted that one reason might be because there was not enough blank verse in its training data for it to generate the same “confidently.”
Upon delving deeper, ChatGPT admitted that some of the literary works it was trained on, including poems, “may be protected by copyright.” Poems by Rupi Kaur were included in its data set, while poems by Meena Kandasamy and Ocean Vuong were “likely” included in its data set, the chatbot claimed.*
In spite of this impressive repertoire, ChatGPT struggled to change up its poetry style.
Regardless, there are mounting fears of ChatGPT and similar AI platforms usurping the position of human artists and writers.
Portrait of the artist as a large language model
Tishani Doshi is an award-winning poet, novelist, and dancer whose themes of exploration include gender-based violence, the protection of nature, and families that defy borders. Her poem, Girls Are Coming Out of the Woods (2017), starts like this:
Girls are coming out of the woods,
wrapped in cloaks and hoods,
carrying iron bars and candles
and a multitude of scars, collected
on acres of premature grass and city
buses, in temples and bars.
When ChatGPT was asked to write a poem in Ms. Doshi’s style, it generated five stanzas of rhyming verse, of which this was the first:
In the quiet of the night,
When the stars are bright,
I think of all the moments gone,
Of laughter, tears, and songs.
Ms. Doshi was shown the entire poem that ChatGPT had created in her name, during a phone interview with The Hindu.
“I don’t recognise that as my work and anybody who knows my work would not recognise it as my work,” she said, before observing that a person unfamiliar with her poetry might think that ChatGPT’s end result represented the kind of poetry she wrote.
What if she collaborated with ChatGPT to write a poem? Ms. Doshi pointed out this would keep the poet from fully confronting the issue they were writing about in the first place.
“Say I want to write a poem about climate change and I’m just going to put in a few images, here are a few words, make me a poem—it might make a fine poem but it’s not going to help my understanding of why I wanted to write about climate change. It’s not going to help me hold my fears about climate change because I have had very little to do with the making of it,” said Ms. Doshi.
“If we are going to outsource the complexity of human aliveness, then what is the point of being alive, you know?” she wondered.
Ms. Doshi said that a poem made by her in collaboration with ChatGPT would be an “approximation” of a poem by a human. She said it would not resonate with her.
However, the poet admitted that if a colleague or someone she admired tried making art with a tool like ChatGPT, she may still give it a try to satisfy her curiosity and see if their work could move her.
Writer, scholar, and translator Meena Kandasamy, a winner of the Hermann Kesten prize, also read a poem that ChatGPT had written in her “style.” The chatbot’s poem began so:
With every word I speak,
I strike a blow against the chains of tradition,
Against the patriarchy that seeks to keep me weak,
Against the forces that would silence my conviction.
“It’s cute, it’s quite nice,” Ms. Kandasamy said happily after reading the whole poem. “It’s got the heart of what I’m trying to to say. I think the content is what I write but I think what’s different is perhaps style, because my style is much more individual than this.”
“It’s me, and it’s not me,” she explained, noting that ChatGPT’s version of her poem was less artistic, less subtle, and more transparent than what she hoped her own work was like.
Ms. Kandasamy added that the poem was “beautiful” and laughed at the idea of feeling violated by ChatGPT using her work to generate poems. On the contrary, she said that her work being part of ChatGPT’s data set was “democratic.”
“This is knowledge,” she said. “Knowledge is meant to be for everyone, right?”
“My work exists like the work of every other artist—to influence others.”
However, Ms. Kandasamy said that she respected the feelings of creators who did not feel the same way.
ChatGPT discusses its data set* | Photo Credit: Screenshot from ChatGPT
The industry responds
Reaching out to a few publishing companies and literary magazines could help aspiring AI poets and AI writers gauge whether the industry deems their work original or not.
In response to The Hindu’s email query, Radhika Marwah, Executive Editor at Penguin Random House India, said, “We are not averse per say to submissions which engage any of the multitude of AI tools available today. Every book proposal is unique, so it isn’t possible to really police the form in which is written or created.”
However, Ms. Marwah pointed out that a book written with ChatGPT would probably fail the publisher’s plagiarism checks.
“More than the AI tools, it is about how an author can bend newer tools to their will and use emerging technologies to express themselves better,” she explained.
On the other hand, Clarkesworld Magazine, which publishes science fiction and fantasy submissions under 22,000 words, already has an AI policy in place. The magazine’s website said, “We are not considering stories written, co-written, or assisted by AI at this time.”
ChatGPT wants the final word
When asked whether it wrote better poetry than a human, ChatGPT told us, “As an AI language model, I don’t have emotions or experiences like humans do, so I cannot fully appreciate the beauty and meaning of poetry in the same way humans can.”
It added, “That being said, I can be a useful tool for generating poetic inspiration or exploring new forms of expression, especially when combined with human creativity and judgment.”
But ChatGPT still insisted that its poems were “original.”
*Disclaimer: AI-powered chatbots are prone to a phenomenon known as “hallucination,” where they generate logical sounding yet completely false answers. For this reason, a response generated by an AI chatbot cannot be taken as a fact at face value. This report was researched using the February and March versions of ChatGPT.
