Dan Power is the founding editor of the AI Literary Review. A poet, studying for a PhD in Creative Writing at Lancaster University, his works include Memory Foam (Doomsday Press, 2023), Selected Dreams (Steel Incisors, 2021), Predictive Text Poems (Spam Press, 2016). He is also the founding editor of Trickhouse Press. Issue One of the AI Literary Review, published 1 July 2024, can be read here and our review of it here.
The Indiependent spoke to Dan about the AI Literary Review and AI’s impact on poetry more widely.
Just to get this going – could you tell us a little about yourself and what led you to start the AI Literary Review?
Sure thing, my name’s Dan and I’m a poet. I’ve been interested in digital technologies and what they’re doing to us for ages, how they change language and the way we interact with each other. This is something I often end up exploring in poetry, intentionally and unintentionally. That started in around 2015 or 2016 when I came across Spam and discovered that there were loads of people writing with and through and about the internet in so many fresh and exciting ways. I loved it, caught the post-internet poetry bug straight away, and never looked back.
AI has not just appeared, so what makes now the right time to launch such a project?
Because it’s exciting! AI becoming publicly available is a sea-change – what does it mean to be creative, to have authorship, to be original, authentic, poetic? It feels like everything is in flux now, and whenever things are in flux there’s the potential for them to transform.
I saw a lot of fear and anger and mistrust towards AI language tools online – a lot of it very valid – but there’s great creative potential here too. I didn’t want that potential to go unnoticed, and I knew I couldn’t uncover it or discover it on my own, so it seemed natural that there needed to be a journal or magazine to document and share these early experiments with AI and poetry.
I think in that way I’m trying to follow in the footsteps of the folks at Spam – they’re not just publishing good work, they’re creating a centre of gravity for all this exciting and like-minded work to orbit around, they’re curating and defining and pushing things forward. I’d like it if the AI Lit Review could do something similar, by getting people to engage with AI’s creative/collaborative potential in an open-minded way, or by getting poets to disrupt or reformulate their normal writing practice by throwing something new into the mix.
I get why a lot of people are repulsed by AI-generated art. A big reason is that there’s no intention, no humanity behind what it generates. It’s soulless, corporate, and because it can be generated so quickly and cheaply it threatens to put so many creative people out of work if it’s left unchecked. So we have to check it. AI isn’t going away any time soon, and as writers and readers, if we care about the future of writing at all, we have to face AI head-on. We have to find ways of working with generative AI that keep human authorship and experience intact. AI should be a tool we can either use or not use, rather than something that eclipses us. It’s important to neutralise that very real threat.
But also, it’s more than that – AI can be genuinely useful, it can lead us to poetic conclusions that we wouldn’t have reached without it. This thing is bristling with potential. I want to know every way of using AI to write, to draft or edit, disrupt or polish – I want to see what everyone is doing with AI and how they’re doing it. This is why the journal doesn’t just publish poems, but also a brief description of what software was used and how it was used in the creative process.
After your first submission period, are you surprised by any of the submissions? Have any of the submissions sparked ideas for how you might use AI differently in future?
Absolutely! We’ve had a few working with speech-to-text programmes and AI voice assistants, which adds a whole new dimension to the human-AI interaction and opens up so many experimental doors. In my own work I’ve been largely zeroed-in on text-based AI, but there’s so many AI tools which operate in different ways – using voice or image or raw data or physical motion – and surely each of these can be exploited in some way to produce poetry.
Another avenue to explore would be creating my own AI programme, or training a Large Language Model (LLM) with hand-picked data, in order to generate more bespoke first drafts. This kind of gets around all the issues of plagiarism and authorship, especially if you train the AI on your own words, and I guess would make the trainer the author of the AI’s first drafts and eliminate the necessity to re-draft each generation until it becomes your own? That’s all beyond me anyway. We had one submission where the poet performed a Markov chain with AI-generated data. The poem is wicked, but I truly don’t understand how she did it.
How have you found using AI in your own poetry – such as your collection Memory Foam?
Fun – I’m treating it like a shiny new toy at the moment, and always finding new ways to play with it! So the novelty hasn’t worn off yet, and maybe never will, because there’s as many new ways of using LLMs in writing as there are of writing full stop.
In Memory Foam I was interested in the conversational aspect of working with an LLM, the contrast between my imperfect human prompts and the sanitised, soulless professionalism of ChatGPT’s replies, so the pamphlet takes on a sort of call-and-response form. The poems are often transcripts of our conversations, either with my questions left in or omitted, and the “writing” in that case was more like steering a conversation than it was generating something from scratch. It felt like a collaboration – and it also felt volatile. There was no way of pre-determining what a poem would feel like or be “about” because, even with specific prompts, ChatGPT has this limitless capacity to misunderstand you, or give you excess unwanted information, which takes things in unexpected directions. Very quickly I realised that to work with ChatGPT you have to be willing to give up some control – not just of your voice, but of the things it says as well. Angling for any specific ideas is tricky.
More recently I’ve been experimenting with ChatGPT (et cetera) as draft-generators. I’ve been asking for sonnets with impossible conditions, like a sonnet that takes place in a forest but also in a bath, to get more interesting first drafts. ChatGPT always takes the path of least resistance, and will give you the most probable response to whatever you ask, so asking for something impossible forces it to improvise, and makes its “poems” less predictable or cliched or boring. Of course, the drafts it generates are still lacking some heart and need a lot of work.
My PhD research at the moment explores how these AI-generated poems operate, why they feel lifeless, and what human poets can do to revive them. So far, it’s looking like the answer is to rewrite them massively – but not to the point that it makes the AI step redundant! You still get a first draft to work with, which saves a lot of time, and often the draft has images and ideas that you weren’t expecting, which helps you avoid that classic human pitfall of writing the same poem over and over again. As you draft and redraft and redraft it the AI’s sanitised voice gets squeezed further and further out, its general images of moons and trees are replaced by more specific images that are personal to you, and at some point it stops being a poem generated by AI and becomes a poem authored entirely by a human, you, who just used ChatGPT to get the ball rolling. Not sure exactly where that line is yet, of course. But I should have more information on that soon.
You talk about AI as a “muse” – what do you mean by that?
I’m not sure yet – that’s something I’m trying to work out with the PhD. I think traditionally a muse was a figure who inspired a creator to create something, and you can take AI’s generations as a source of inspiration for sure. But also, to get an AI to generate anything you have to prompt it, you say give me a picture of a dog and the AI produces a picture of a dog, so in that sense maybe the human is the muse and the AI is the creator? But then if you work creatively from its output then the AI is your muse… it’s a bit of a two-way thing. Humans and AI programmes are inspiring each other.
Is there a conflict between its nature as a tool, that is responsive to commands, and a “muse”, that creates a response in writers?
Yeah, I think so. A tool is kind of inert until someone picks it up and uses it. Is a muse like that? A muse who’s a human is definitely a sentient thing moving on their own, but are they only a muse when a poet comes along and muses with them? Does a muse muse actively? I think an LLM does – you prompt it and then it prompts you. Maybe the AI-as-muse idea doesn’t really work. Or maybe I don’t know enough about what a muse is or does yet – there’s a deep history to explore there. I guess muse should be more of a verb anyway. Maybe the AI isn’t a muse, but it’s a tool we can muse with, something to help you generate or reformulate ideas? It greases the wheels of the mind. It provides material that you can then sculpt into a poem.
You also talk about imbuing AI art with a soul – is that a need inherent to AI or a product of the people who have pioneered AI art disregarding its importance?
I think, fundamentally, nobody wants to read anything written by a computer. We want writing to meet us somewhere – we want to see through someone else’s mind, think through someone else’s thoughts, and in any process-driven writing there’s a big, palpable void where the author should be, unless a human has intervened on the process somewhere along the way.
This limits what you can do with AI in writing – I think if you’re not adapting the AI’s output, or really specifically prompting the AI, or training your own LLM, then you’re not bringing anything human to the table, and you’re not offering the reader any way to find you in the work. So for the reader to be able to connect to an AI-generated text I think it has to be “ensouled”, or revived, by a human author.
As well as making a poem or other text more authentic, this human intervention on the AI generation can make the work more interesting and surprising too! And also, if you’re re-working and re-writing the AI’s text significantly you also remove the risk of accidentally plagiarising something that the AI has regurgitated without referencing.
Do you think AI should be viewed differently to other technological changes in terms of how it alters creative practices? For example, you mention the change from quill to pen in a recent interview with nb.
No. It feels radical, and scary, but what doesn’t when it’s significant and new? Does it diminish our writing’s authenticity, its meaningfulness, our sense of authorship? Only if we let it. Will it ever put human writers and editors out of a job? Only if it gets significantly better – at which point I can’t imagine it would still be available for free!
The big concern is the amount of energy generative AI consumes – and the volume of water used to cool the server farms. This one worries me – the internet was already the largest polluting industry in the world before the AI-boom, creating more emissions each year than every plane in the sky put together, and it’s only getting worse. People worried about the environmental destruction the industrial revolution would bring too, and they were right to do so – it did cause destruction, is still causing destruction, and we don’t seem to be on the right track to stop it! I don’t want to dwell on the climate collapse too much in an interview about AI – one existential nightmare at a time! – but I’m quietly hopeful that things will improve and get more efficient as the technology matures. Energy sources are increasingly, and significantly, veering away from fossil fuels towards renewables. And processors are getting more efficient, less greedy for water and fuel. So while it would be reckless to pivot all writing from analogue to AI-driven overnight, I do see a future where using AI in writing practice is only marginally more or less impactful than writing on paper, or typing something up over the course of a few hours. The boffins will crack it I think. The energy will get cleaner and the processing more efficient.
And then of course there’s brain organoids! Have you heard about the organoids? They grow these tiny human brains from stem cells and use them as processors for computer data – so much more efficient than plugs and wires. They can do something similar with mushroom mycorrhizal networks I think. Maybe that’s the future – living computers – organic artificial intelligence! You can buy a brain organoid today if you want, they sell them online. Or maybe that’s just swapping one existential nightmare for another? I hope those brains aren’t alive. Born to work. Kept alive to work. Hard to imagine anything more bleak.
In that interview you also compare current backlash against AI to the Luddites and say the Luddites were right – is the AI Literary Review, in part, then an attempt to bring AI into current creative practices in a way controlled by said creatives, and so avoid the classic capitalistic process of mechanising away jobs?
Yeah I think so – there’s definitely a future where AI replaces all copywriters, then regular writers, then creative writers, and I think if we want to avoid that there has to be some intervention. Ever since they added autofill to text messages we’ve been on this slippery slope, outsourcing our expression to an algorithm, letting a computer tell us what we want to say. That’s no good – not everyone is the same person, not everyone will want to say the same thing!
Imagine a world where most of the text you encounter is AI-generated with no human intervention, it’s generic and predictable and flavourless, and kids grow up in this world thinking that that’s how expression is, so they copy it, and they start to sound more and more like soulless chatbots, and the chatbots learning from the kids keep getting their own beige language fed back to them… we could create a horrible feedback loop where all individuality and novelty is washed out of the way we talk. That would be terrible not just for poetry but for the whole world!
I realise I’ve gone off topic a bit here. Your question was about jobs – I think trying to keep language alive and personal in the face of digital reproduction is also a way of ensuring that being a poet, or a writer, is still possible in the future. If language and text become so generic, and so readily available that they lose significance then we would lose poetry, we’d lose literature, and we’d lose the ability to really understand ourselves. In that world there would be no place for a creative writer, and maybe no place for any other profession where language is your toolkit – like teaching or therapy or hostage negotiation.
What are your longer-term aims for the AI Literary Review? Do you hope to end up one of many publications focusing on AI as part of a creative practice, or do you hope that wider poetry publications will accept AI as one of many possible tools available to poets?
I think general acceptance is the end goal. And for people to be able to openly say “here’s a poem that was co-written with an AI programme, or a poem edited using AI tools”, without any stigma of being called lazy or a plagiarist – you can be lazy and steal without AI, and you can be hard-working and original with it! It’s not a factor really, and it shouldn’t factor into people’s enjoyment or appreciation of a good piece of writing.
Already in the first issue of the AI Lit Review there’s so many great ideas coming to life – poems that invent new words, poems that distort reality, poems that are funny, poems that are spooky – and it would be a shame if this work went unread because AI had been dismissed as a creative tool.
It would also be no good if a poet who used AI had to hide that fact, if they felt like their work would never be published or read because of the methods used to create it. Just seems better all-round if we accept that AI is here, see what we can do with it, and allow everyone the freedom to play how they like and share what they like.
This interview has been lightly edited for style and length.
Interview conducted by Ed Bedford
Want more Books content from The Indiependent? Click here