Who owns poetry? Or rather, who does poetry belong to? Let’s say that poetry belongs to the people who read and write it. Until recently, this group consisted solely of human beings and Koko the gorilla . Today this group has expanded to include Generative AI (GenAI) tools such as DeepSeekAI and ChatGPT, which have not only ingested colossal amounts of poetry during their training, but can also produce new poems on demand. You may have seen a recent study demonstrating that readers, without being told which poems are human-made and which are AI-generated, can’t distinguish the simulated from the sincere . While no GenAI tool has mastered poetry, most produce better poems than we’d like to believe. Today, we’re living through the transition between pre- and post-AI worlds. I believe that human-generated poetry can survive this transition, but only if we face GenAI head-on. This is not an anti-AI hit-piece, and neither is it a defence of low-effort, AI-powered plagiarism. Instead, it’s an attempt to find a sturdy middle ground on which human poetry can build its defences.
GenAIs produce large volumes of uninspired poetry at lightning speeds – a clear threat to the craft, culture, and commercial infrastructure of poetry. Vilifying GenAI won’t be enough to stop people using it in bad faith. Equally, embracing GenAI won’t make its poetry any less meaningless. Instead, if we want to prevent a deluge of AI-generated slop from lowering the quality of poetry as a whole, we should raise our expectations for what AI-powered poetry can be. How might GenAI be used ethically and artistically – not as a replacement for human creativity or effort, but as a tool to nurture these things? How can these tools co-exist with human poets peacefully, without taking any of the joy or meaning out of anyone’s poetic practice? We’ll answer these questions further down the road. But, before we get there, we’ll need to examine our ideas of authorship and ownership, and how these look in the light of GenAI.
WRITING VS GENERATING
When we write poetry, we feel that we own the words we’ve written. Nobody feels ownership over the words themselves – language is a communal resource – but we do feel attachment to the particular assemblages of words we create. Unlike GenAIs, humans can assemble words to relay or reconstruct their internal thoughts and feelings. Even when writing in a persona, or writing ironically, writers assemble words according to their own mental schematics. The feeling of attached ownership we have over our words – what we call authorship – arises because the words we use are extensions of our internal selves.
Translating ideas into language can be personally demanding, or at least time-consuming, and making this effort increases the ownership we feel for our assemblages. This makes us hostile to AI-generated poetry, assembled unconsciously and numerically, where nothing personal has been staked and little effort has been expended. Furthermore, when we read poems which resonate with our internal selves we feel a connection to their authors: we feel understood. If we believe a poem is AI-generated this connection isn’t possible, because we don’t believe there’s another conscious mind for ours to connect with.
Human-generated poems work when we recognise that the voice behind the language belongs to another person. The sense of an active agent behind the text, assembling words to create intended effects, is something GenAI can mimic but not replicate. GenAIs have internal minds, but not the kind that can dream, desire, or make autonomous decisions. Their words never correspond to real thoughts or feelings. GenAIs assemble poems without intention, whereas human poets choose their words consciously and deliberately. Agency is the catalyst of authorship – it’s what distinguishes writing a poem from generating one, and what enables humans to own their assembled words. So, while GenAIs can produce real poems, they lack the agency required to be real poets.
[2] https://www.nature.com/articles/s41598-024-76900-1
Read the full article in Magma 92:
*
Buy this issue for £8.50 in UK (including P&P) »Buy Now
Supported by Arts Council England