It’s often through struggle that intellectual growth takes place. As Elbow (1981: 131) puts it, “new and better ideas … don’t arrive out of the blue”. Instead they “come from noticing difficulties with what you believed, small details or particular cases that don’t fit what otherwise feels right”. Unless we are willing to grapple with what feels wrong, our sense of what feels right will be limited and partial. This is why we need to cultivate “a willingness to notice and listen to these inconvenient little details, these annoying loose ends, these embarrassments or puzzles, instead of impatiently sweeping them under the rug”. What we recognise as a “good idea” will seem “obvious and inevitable after it is all worked out and the dust has settled, but in the beginning it just feels annoying and the wrong old idea seems persuasively correct” (Elbow 1981: 132).
This insight about intellectual development has profound implications for how we approach AI writing tools. Language models like ChatGPT and Claude offer us something seductively powerful: they provide immediate solutions to whatever difficulties we encounter with our writing, offering relief from authorial challenges which are creatively significant much as they can be intellectually gruelling and anxiety provoking. Even when this works well, it will tend to produce what Elbow (1981: 18) describes as “language that is clear and perfectly obedient to the intentions of the writer, but lifeless”. Though he was writing decades before language models, his emphasis upon the struggle involved in the process is deeply relevant for understanding the challenges we now face.
Language models did not create the inclination to subordinate the creative process to the output, seeking to manifest our intellectual intentions as efficiently as possible in an article or chapter which best meets the expectations we feel subject to. However they do provide a radically powerful new way of enacting that subordination, enabling us to create texts which are “perfectly obedient” to our intentions but which are likely to be, to use Elbow’s evocative term, lifeless.
The allure of these tools is particularly potent in contemporary academia. When we’re drowning in deadlines, administrative duties, and pressure to publish, the promise of smooth, effortless writing can feel like a lifeline. But what if this very efficiency undermines what makes academic thinking valuable in the first place?
The point I’m making is not that models are inherently problematic for writers but rather that the orientation we take towards them is crucial. The manner in which we approach them is not a function of the model itself, though of course it has been designed to encourage certain kinds of use. It reflects the context in which we live and work, particularly the expectations we are placed under and the incentives provided to us to meet those expectations in particular ways. I am intensely worried about the role language models could play in a deterioration of scholarship (Carrigan 2024: ch 8). But in a real sense I don’t think we should blame models for the incentive structure which guides academics within an accelerated academy (Vostal 2015, Carrigan 2016).
The reason academics are likely to approach them as a solution to their problems is that the structure of our work within the contemporary academy presents us with a lot of rather intractable problems! However if we see them as a way out of difficulty, they carry the risk of diminishing us as writers and stunting what we produce. However if we see them as a supportive presence through which we can grapple with that difficulty, helping us go deeper into it rather than seeking to evade it, they can enrich and expand our capacities as writers.
This distinction, between using LLMs to avoid difficulty versus using it to engage more deeply with difficulty, offers a framework for thinking about academic writing. The former approach seeks to outsource the creative struggle; the latter treats LLMs as a dialogic partner in that struggle.
It’s easy to slip into this worldview in which we must protect human creativity from the encroachment of external forces. It can be fun too. It offers a sense of purpose to imagine ourselves fighting a valiant rearguard action against the forces of capitalism and technology. The problem with such a view is that it obscures the extent to which creative work has always been bound up in compulsion, the manner in which creative impulses are inevitably knotted together with the practical demands of employment within an organisation and/or production for a market of cultural commodities. It is exceptionally rare to be able to produce in a way which is entirely devoid of external considerations, simply expressing the creative impulses which spontaneously emerge within oneself. The renaissance notion of the writer as a creative genius, “someone who is personally and individually responsible for her work, who by dint of extreme effort and self-control can determine the outcome of her product”, has a long and storied history (Dreyfus and Kelly 2011: 53). Even if we see it in its Nietzschean mode as a challenge to be and do more, to become as gods who stake a unique place within the universe, it’s an overwhelming task.
While I’m claiming that we risk losing something profound about the writing process if we rely on models to get us out of difficulty, the problem I’m identifying is how we use them not that we use them. It’s not that a human process is corrupted or imperilled by a technological element but rather that we should avoid drawing on that element in a hasty or careless way which changes the nature of the process. The problem with circumventing the tea ceremony by going straight for the caffeine is not that there are more potent and immediate ways of delivering caffeine, but rather that there was a purpose to the ceremony which we miss out on if we reach straight for the outcome. If we have sat through the ceremony many times, this might not be a problem at least initially. We retain the capacity to focus, to inhabit a complex and diffuse process, even if we forego the occasion to do so when there is a pressing need. Eventually though, we can expect that our engagement will change. If we try to sit through it, we will become twitchy and impatient.
It risks leaving us with a diminished view of writing as “an assembly of placeholders” which when “magically becomes more than the sum of its rote parts” when “everything is finally compiled” (Horning 2023). If the whole is simply an arrangement between the parts which is more or less functional why does it matter how we got there? It matters I would suggest because we will go on to write other things with other people. We will go on to think other thoughts about other matters. The significance of the writing is not just the finished product but rather the change it brings about in us as we become the person who wrote this and wrote that, moving through the world with a trajectory of these texts which we leave behind us. The point I’m ultimately trying to argue is that we are less likely to grow if we rely on machine writing, outsourcing elements.
Mark Carrigan 