Do plagiarists dream of electric feats?
The plagiarism that seemed to be all over the poetry world a few years back seems to be less of an issue today. But has the rise of AI helped similar shenanigans to go unnoticed?
It's still easy to tell when a poem has been written by an AI in response to a prompt, though the standard has been raised somewhat in the past year. Back then, I made a video aand wrote a post where I critiqued a sonnet written by Gemini. I specifically chose Gemini because the more widely used and well-known ChatGPT was unable to write a sonnet. It would always reply with fourteen lines of iambic tetrameter and couldn't shift to pentameter when I asked it to. But now, it appears to be the case that Chat-GPT can write a sonnet in pristine I.P.
As a creative writing lecturer, I don't want to give away all the tell-tale signs of a poem that has been written by an AI. One that I will point out is that the AIs have nailed the metre, but the metre is now too perfect. If you read any classic sonnet, historical or contemporary, there are always little modulations – hypermetric beats and caesuras – that add texture and variety to the poem. It takes subtlety and a trust in the musicality of human speech to carry this off.
As far as poems that AIs produce from scratch on the basis of a prompt are concerned – an editor, judge, lecturer or experienced reader or writer of poetry will sniff it out with relatively little effort. But there are other ways of using AI in the composition of poetry that could be harder to detect, albeit one that takes some previous knowledge and experience of poetry.
Note: At this point I think it’s important to emphasise that I am not saying that previous perpetrators of plagiarism are now back to their old tricks with AI. Hence why I won’t be referring to anybody by name.
About a decade ago, there were a spate of cases of plagiarism from winners of poetry competitions as well as poets publishing with small to medium publishers. I'm not going to reignite the coals in order to rake the perpetrators back over them. But I will mention the general technique that they used in order to end up on competition shortlists or get their books published. In its most crude form, the culprit would take a poem from a successful poet, one that had already placed well in competitions or been nominated for prizes. Then, to differing degrees, they would swap out words and phrases for parallel or similar ones, rejigging it enough to not be immediately recognisable to the judge's eye.
Sometimes, the plagiarist would be discovered by the poet themselves; other times by their readers. A few poets with a natural instinct for the offending works had found themselves in the role of poetry sleuths. Occasionally, the poem had been changed quite dramatically but was given away by structural similarities; other times the plagiarist had only swapped out a handful of nouns and verbs.
Whenever they were exposed, the culprits wouldn't hesitate to offer the same line of defence – that they had used the original poem as a model and had accidentally submitted an earlier draft. Some red-faced editors went to ridiculous lengths to defend their own lack of insight, from citing Barthes's Death of the Author to contrasting the individualism of authorship with the social aspects of oral cultures.
However, one thing that has been of note is that there haven't been as many cases of plagiarism sending wine-glass-trembling shockwaves through the worlds of poetry prizes and competitions of late. Could it be that the plagiarists have been sniffed out or that future culprits have been put off by the high profile unmasking of others? Or could it be because other tools that have become available?
While AI tends to have a cut off point with regard to the information it draws from wholesale theft of web content and books, there are ways to train an AI within the confines of specific threads that aren't fed back to the main database. If there's one thing that separates the plagiarist from the lazy student, it's a sense of what makes a good poem and a knowledge of what judges and editors are looking for.
I have sometimes advised students who want to submit to competitions that it's worth checking out the judges of competitions they send off to. Not just for the kind of poetry that the judges write but also the books that they have reviewed and the poems that they have publicly admired. As cynical as this might sound, the pointer was more for the student to find something within the body of their own work that might chime with the tastes of the judge.
A plagiarist, with the same sense of what the judges might be looking for, can use their knowledge to this effect. Only this time, instead of Ship-of-Theseus-ing their way towards an entry that is similar enough without arousing suspicion, they can train an AI to produce works that seem original.
The Large Language Model (or LLM – the thing that most of us mislabel as AI) works by presenting the user with the kind of output that most likely matches their query. It takes work from its database, smooshes it together and then re-writes it in a way that smooths out the joints. AI art works in a similar way in the sense that it combines previous work it has scraped, stitches it together before knocking it out of focus and then re-sharpening it. As you can see, the tools that we call AIs and the oldschool plagiarists with their method of starting from a model seem to be very similar, with the AI outcompeting the human in terms of sheer iterative horsepower.
But the thing that the AI and the plagiarist also have in common is that lack of immersion in the act of creation – in the interplay of the imagination and the unconscious. The LLM and the plagiarist miss out on is the joy of the process itself The end result might be an awful poem, but the process of making it was also a kind of transformation – something that the poet might feel compelled to repeat for its own sake and might even produce a better poem somewhere down the line.
There have also been a number recent cases where authors have left AI prompts within their manuscripts. These continuity foghorns seemed to bypass the editorial process (which might itself have been a runthrough of Grammarly or another AI model) and only came to light after being discovered by bemused readers. My feeling is that poems written by trained LLMs (instead of the more naive products of simple prompts) are already rife within submission piles and are going to be much harder to identify. If previous plagiarists were deft enough to bypass the filters of editors and judges before coming a cropper through the awareness of readers, authors and sleuths – it's going to be lot trickier to trace LLM plagiarism. Unnoticed prompts and other tells might become apparent within longer forms but such carelessness will probably not repeat in a one-off poem submitted to a competition.
If someone does achieve success through this, they’re more likely to come undone the higher they climb in the literary field. Sloppiness and delusion creep in eventually. But, in my modestly informed opinion, I think this is going to become a bigger thing at a more modest level, if it isn't that way already.
If a culture revolves around prizes and output, rather than the reward of the process itself, then those who are more focused on reward will rise to the top. While some might do it through blunter transgressions such as plagiarism, others will do just as well through the more subtle strategies of schmoozing, nepotism and derivativeness. At least one high profile case of plagiarism came from a bright young thing who knew all the plays when it came to being seen as a big prospect within their literary scene but had to resort to stealing someone else's work when it came to supplying the material for their literary debut. Even without the plagiarism, whenever I see that someone has won a big prize I immediately play a little game of finding the professional or personal connections between them and one of the judges. Business partners, proteges, work colleagues and former housemates have featured heavily.
I don't often have an answer to the big problems that I have a tendency to chew on in these posts, but this is a relatively easy nut to crack. If you want people to appreciate the process as much as the output, then poets should share their process with others.
Not in a "tell your insider secrets to people that shell out for your literary retreat" type way, but rather in a "here's something I'm working on, here's how it started and here's how it's going" type way. This would, of course, rule you out of contention for all those anonymous poetry competitions and small journals that insist on work has never been shared elsewhere, but it might encourage readers and fellow poets to value what is uniquely human and present about the process. Let the machines and narcissists worry about the outcome.
Rusty Niall has settled into a pattern of longer monthly essays (like this one) and shorter, informal weekly posts. If you enjoy them, please consider sharing them with someone you think might like them too. You can also support my work through some of the options listed below. Cheers.
Niall