As part of my artist residency with Festival of Libraries, I’ve been experimenting with AI as a ‘collaborator’ or writing tool.
For this blog I’m focussing on creative experimentation and technical R&D. But! If you’re interested in the conceptual reasoning behind my approach and how I took artistic inspiration from Borges’ short story The Library of Babel – check out the full project essay on my artist website.
Right… where was I?
I’ve been interested in writing with AI for a while now, ever since attending a Next Gen Storytelling event with The Space towards the end of 2019. It was fascinating to hear from artists like Ross Goodwin, Es Devlin and Libby Heaney, who have harnessed AI in such creative ways – from using a car to write a novel, to a bot posing as Lady Chatterley’s lover wooing strangers on Tinder. If you haven’t seen it already, I highly recommend watching Ross Goodwin’s sci-fi short, Sunspring. It’s a few years old now, but a really good indicator of where AI was up to in its language processing and creative writing prowess at the time. Also, it’s just a really fun watch!
But while the event was creatively inspiring and set my brain cogs whirring, I still felt at a slight loss as to how I, a non-technologist, could get started writing with AI.
Fast forward to early 2021, when I was mentored by the rather brilliant author Lynda Clarke, who knows a thing or two about writing with technology. Lynda kindly shared with me some of her research into AI writing tools, and how to incorporate them into your process. Things like OpenAI’s GPT-2 (a language-based generative AI, trained on huge datasets of written text), and web-based text generators like TextSynth and Inferkit.
Now I had my hands on an easy-to-use, web-based text generator, I could experiment!
Unusually for me, I started off with a love poem. (Maybe somewhere in my head it was an ode to Alan Turing’s love letters.)
I typed the first romantic thing I could think of into Inferkit, then asked it to generate what might come next. I then made some minor trims and tweaks to shape something that felt a little more finished. Below you can see what came out of the text generator (my input line is in bold – yes I know, awful and not remotely poetic). And, on the second slide, how I honed the text into something that – to me at least – felt more poetic. But in truth, there’s not a huge difference between the two!
After a bit more experimentation, I learned that you get what you give with these sorts of text generators. The AI is trying its best to emulate you, so your prompt matters – try to start as close to the style and form you’re after. That said, I really love its seemingly innate tendency toward the weird and unexpected. In the above example I especially enjoyed the line “a surfeit of chlorophyll that mesmerises you” – something I would never have written on my own. It may be gibberish, but to me it suggests a feeling that is simultaneously earthly and alien… a little bit like love, no?
Emboldened by my experiments, I decided to test AI on the diet of distant librarians…
Inspired by the conversations that took place during my artist residency, I wrote (yes, all by myself) some short vignettes. I then offered these to TextSynth to munch upon, and spit out its own musings and suggestions. Reader, it was a mixed bag! Some lines made little sense (“we go off our toes in the middle of a town square”), some were strangely poignant (“we take our first breath in the sun of childhood”).
Interjection! >> It’s important to say that my goal here was ‘play and experimentation’ over ‘poetic quality’. So if you’re thinking these are awful poems – that’s fine. What I input wasn’t of the highest literary quality either. But I will say that I was impressed by the AI’s ‘imagination’ and its ability to emulate – at least in general – the rhythm and tone of what I was after.
Once I had some interesting lines, I chopped out the ones I liked, reassembled them into a new prompt and fed them back into the text generator. I did this several times over until I had a chunk of lines that I especially liked. I then took this collection of words, sentences, fragments… edited, rearranged, cut and paste… and rewrote in places until I had something resembling a poem.
My intention here was to take a sort of Dadaist approach, embracing nonsense, unexpected detours, serendipity and the awesome power of assemblage! What I ended up with were a set of collage poems, co-authored by me and AI, celebrating libraries. Below you can see a mix of my inputs (slide 1), the text generator outputs (slides 2 and 3) and one of the final poems (slide 4).
I’ve no doubt some very excellent computer brains are working on higher fidelity language generators that can pen the next great novel. But for me, I like the accidental jankiness of the current tools. They have the power to surprise, delight and free your imagination from tired metaphors. Rather than replacing us, they just might make us better, more creative writers…
Psst! You can read the full set of poems on my project page or get a deeper insight into my artist residency with Festival of Libraries over on my artist website.