AI and Text Generation in Poetry
“Poetry… it’s rooted in the inspirational and the comical—the deeply human—and yet, in many of its forms, it edges toward the computational and algorithmic.” Dan Rockwell, The New Yorker, Jan 2020
Can a Computer Write a Poem?
When we worry about the dawn of AI, we often talk about how computers can’t ‘feel’ for us, and how art is one of the final things human beings have the monopoly on creating. We might not do maths as well as a neural network, but how could a neural network describe the unique pain and joy of the human experience?
The art of writing poetry with a neural network (an artificial intelligence modelled on the human mind, which ‘learns’ and can be trained to ‘think’ over time) is in its infancy – it’s harnessing young technology in a novel way. Whether you consider writing with a neural network to be co-authoring a poem with a machine or just one writer using a complex tool is an interesting question, and one which we’ll be exploring for many years to come as technology evolves.
The Digital Humanities team hosted the White Review prize winning poet, Charlotte Geater, earlier in 2022, and they explained their history of authoring and publishing poetry using neural networks. You can click through here to view the talk, which includes many examples of Charlotte’s AI poetry, written with corpora composed of a variety of poets and styles, as well as detailed discussion about the extent to which the AI can be described as ‘thinking’ and ‘writing’, and the limits of neural networks as creative writers.
More recently, we have been using Python and GPT-3 to run our own AI poets – and while they take an enormous amount of processing power, they have resulted in recognisable poems. Perhaps predictably, it seems that forms with strict rhyme schemes and structures are easier to produce, possibly because they’re more mathematical in nature than free verse.
Try it yourself - here are a few online AI powered lyric and poetry generators: