Poetry and Artificial Intelligence
with Campbell Mcgrath
Today’s commentary is about AI (obviously I am not speaking about Adobe Illustrator or Almost Island) and about young poets. This train of thought actually began when I recently read a poem on “Poetry Daily”1 . I share it here below.
I love the absolute disdain in the poet’s voice. Hatred is good. If you hate things hard enough, perhaps you could learn to be creative enough with your hatred, to turn it into art, or better still, resistance. Art, I find, sieves out the violence from hatred, and distills it into passion. Most of the artists I know hate AI. Some of the younger ones are coming around to see its benefits (for eg. AI helps you extract text from an image, AI could help you with your literature review if you teach it to cite all the information it gives you). I hated AI too. In fact, I still hate AI for how good it is at learning, copying, and also editing. It is the first instance that is somewhat worrisome. But then again, AI can’t replicate human error or spontaneity. The implications of this are deeper than the average commentator’s insights about AI’s inability to “feel” and “sense”. One doesn’t always need to “feel” to be a good archivist, or a good researcher, even. I’m from the generation that learnt to use Google and Wikipedia in my research - in my graduation years, both these internet tools were becoming easier to access (it was the time when Orkut was still a thing). Most of us understood that these were tools - they didn’t provide the answers. This debate comes up again when we talk about “Photoshop” and its influence on Photographic images. I know that most photographers who see their work as “art” swear by Adobe. The camera applies a filter on reality - Adobe Illustrator or Photoshop is just another filter.
Faculties in universities are driving themselves crazy trying to figure out ways in which they can reduce students’ usage of AI. I’ve seen many teachers re-introduce classroom tests - with pen and paper - in an attempt to negate the impact of AI while doing student assessments. This is a stopgap. Perhaps our efforts should be better directed towards teaching young people to use AI in more informed and sensitive ways. It has taken me very long to embrace AI, and I still hardly use it in my work on poetry. I was angry with chatgpt for a whole month after it produced a passable poem when I gave it 3 words as a prompt. I came to the difficult conclusion that AI can write decent poetry, and it’s definitely impossible to tell whether phrases or even entire stanzas of verse have been written by a human or a machine. I know this from my experience as an editor of poetry, and an anthologist. I’m still trying to figure out a way to address this while putting out open calls.
I still believe I can tell whether a poem has been written by a bot or not. A popular Indian poetry page on Instagram that was sharing prompts through poetry month (April?2) recently put out a post about a poet’s work that they loved, and had been sharing on their page. Apparently the poet was using AI to churn out poems that responded to their daily prompts. In their statement the page mentioned that they had pulled down the posts that featured this particular poet, and that they had taken notice after several of its followers had written in alleging that this poet used AI. I was so happy to see this act of community justice, noting that, in the last analysis, it takes human intuition, perhaps, to detect the workings of a machine.
With student answers it’s easy. AI is good at generalisations and producing sentences that sound scholarly, as if they’ve emerged from “deep” insight (ever noticed how language employed by AI - even as you wait for it to respond to you - is punctuated by words like “deep” etc. that assure you it is working very hard or it’s learning from you; much like an earnest employee trying to please their boss). In fact, AI text is better than good. Most of it is formal, confident, and sophisticated enought to fool the average browser. It is usually the confidence that does AI in. It is a unique confidence - the kind of confidence a young marketing executive puts on at the drop of a hat; especially when they are speaking to a client (it’s called client “servicing” for a reason). This is why AI is considered god’s gift to advertising and marketing, and corporate cultures have embraced it unquestioningly. Many of my friends in this professional domain tell me that their companies encourage their employees to invest in paid versions of chatgpt (or other AI tools). They also grudgingly admit the limits of the software, complaining that “last main apun log ko haath maarna hi pad’te”.
Coming back to Art - we’ve established that perhaps AI can write a passable poem -still cheap, derivate stuff in my opinion. (I NEVER ask AI to write poetry - it is an affront to poetry, and AI3). But AI, I must admit, is good at producing poetry prompts. Contemporary poetry pedagogy is heavily reliant on the use of poetry prompts. The American poet Joseph Fasano is well known in English poetry circles for his lovely book The Magic Words: Simple Poetry Prompts That Unlock the Creativity in Everyone. I believe there’s a pdf lying around somewhere on the internet as well. So AI can learn from that book, and from the million other websites on the internet (usually started by poets) that have been around for more than a decade, devotedly producing prompts for young poets to create around. I’m sure you could use chatgpt to help you seek these out, if you’re interested, but haven’t explored this “deeply” enough. So there you have it - for those readers who occasionally write in with questions such as “What is your take on AI as a poet?”: I’d say poets, artists, and creative people of all kinds, should pay attention to this technology. They should learn how it works, and what makes it better. The most rudimentary form of technology is the pen, or the printing press. These technologies are at the root of revolutionary transformations in knowledge production.
There is no limit to the possibilities of machine learning in research. AI works on the basis of what it is fed. (The gorgeous film Humans in the Loop shows the “human cost” of AI really well.) While concept building and analysis is an integral part of good research work, much of academia hinges on “contribution”. “What is the “gap in literature” that you are filling?” (Entitled Academics sitting on Doctoral committees ask) AI can tell you, my dear friend. With a little nudging, AI is a great research assistant. You will still have to pore over everything that it tells you, and chances are you will still have to do your own data collection or “fieldwork”. But then don’t you have to do this with the work of a research assistant anyway? You don’t expect a research assistant to do your research - they are meant to assist you, and in the case of most academics, they are there so you can take advantage of their insights, and take credits for their work. You can now do this with AI - without any ethical hang-ups (apart from the climate cost - allegedly just writing “please” while prompting AI makes it consume huge amounts of energy. I am seeking to educate myself more about this - any help in this direction would be much appreciated).
To cut a long story short, our response to AI must be attuned to its process. Like any other new technology, an understanding of the steps it takes to produce, usually assures best practices in its usage. The problem is that AI creates its own manuals, and the makers of AI are smart enough to not really tell you the specifics about its limits. We don’t really know about the checks and balances, because even the makers are at the stage of “Oooh what can this button do”. Twitter’s Grok is a perfect example. I recently read a case about a reputed senator formally lodging a complaint against Grok. This is perhaps the first recorded case against a machine. (I’m not completely sure about this claim. Still, it is fascinating.)
My emphasis on process - tareeka - is deliberate. I want to draw your attention briefly to Campbell Mcgrath’s writing and editing routine. This is also directed at readers who ask poets questions like “What’s your routine?”, “Do you write everyday?” etc.
“Most of my poems begin in a notebook, and then work their way through revision moving between the computer screen and more hand-written versions on the page. That has been a constant for me, over thirty years as a writer. But much of the rest of my process as a writer has evolved over time, as my life has changed. When one is busy with a family life, a full time job, etc., then writing time has to become flexible. Some days, or weeks, I have time to write all day; often, I can only write for an hour or two. Sometimes, nothing at all gets done on the creative front—but even those slack periods have their value, as one returns to one’s work with a fresh mind. Poetry is so variable that it can be written in big gulps or tiny nibbles, just as poems themselves can be as small as a haiku or as large as the Odyssey.”
From “Process of Creation, History, and Location: An Interview with Campbell McGrath”
I like Mcgrath’s answer - especially the part about “moving between the computer screen” and more “hand-written versions” on the page. My own process is similar. My poems go through several rounds of editing - the words touch different interfaces. One round of editing takes place when the words travel from the mind to the hand. The pen (or even the keyboard) allows a second round while writing. Another round of editing usually happens when I transfer the poem from one app/software to another (Evernote to MS Word, for eg., or ppt to email - by the way it can be a lot of fun to write directly on different apps. Ever tried writing a poem on Microsoft Powerpoint? or Composing a poem, that is actually an angry reply to someone, on gmail?) Inevitably, another round of editing takes place when I share the poem (with a friend, the internet, or as part of a submission to a poetry journal). These edits are natural. They are a part of my attempt to be as precise as possible. There always comes a point in the writing, when I say… “Hmmm I have said exactly what I wanted to say in this poem”, or, “This is exactly what happened”. This is the point at which poetry becomes testimony.
Naomi Shihab Nye’s poem “How Do I Know When a Poem Is Finished?” is about this process. But then again, it happens sometimes that a fragment of a poem lies abandoned in my archive for years, and then a phrase from the poem speaks to me like it didn’t at first, and this phrase or sentence becomes a part of something new. Another beautiful mistake. Perhaps we can start to worry about artificial intelligence when we teach it about mistakes. But we haven’t managed to teach machines randomness - real randomness - how will we teach them to make mistakes. Mistakes can’t be taught, or learnt. Mistakes are made. For now, this kind of certainty is enough to curtail any paranoia I might have about machine learning, artificial intelligence, and the whole shebang.
I return, for one last time, to the poem I shared at the beginning.
A string of other poems came to mind - poems that I have shared on this newsletter before. The direct address to “young” poets reminded me of a Nicanor Parra poem - Young Poets. In the same commentary, I also refer to one of my all time favourite poems by the dazzling Kent Johnson (look him up if you haven’t read him. I relate to his playfulness with “poet selves” and pseudonyms). The poem is called 33 Rules of Poetry - It is an “after” poem; a direct response to Nicanor Parra’s poem. The last line in Parra’s poem about improving on “the blank page” reminds me of another all time favourite; Thoughtfox by Ted Hughes was the first poem I shared on this newsletter, I think - back when Poetly was a website, before I shut it down, and moved to substack.
I doubt any of you readers would remember that… If you do - then thankyou for still being around!
Incidentally one interpretation of “Poetly” is supposed to be an abbreviated form of Poetry Daily - i didn’t know then, of course, that such a page existed
Extremely tempted to put a T.S. Eliot Wasteland joke here.
I urge my poet-readers to try this. Find a poem from your personal archives - a poem you consider a “good” poem. Write down its theme, or its title. Use this to ask AI to produce a poem around the same theme, or with the same title. Tell me your response to the AI produced poem. I’m very interested to see both your response, and the comparison of the two poems. You can write in to poetly@pm.me, if you want to discuss this further fosho.
One hopes that the art always finds a way to reach you. Do write to poetly@pm.me if you have any questions, queries, or comments. I will write back as soon as I find the space, and the time.
If you like what you read, do consider ‘buying me a coffee’.