Over the summer I was reminded of something profound through my experience of writing a book about academic writing and AI. It’s an argument I already wanted to make, but recent events have left me more convinced than ever of its significance for academic writers. This is a project I’d made rapid progress with through various strategies and techniques, using conversational agents as interlocutors without relying on them for machine writing. It left me with a draft of 30,000 words, albeit a messy one, within a few months of starting the project. There were also around 10,000 words of blog posts through which I had initially sketched out the themes of the book.
With a chapter framework and half the book written, I felt it was time to talk to Claude about what exactly I was trying to say. After an initial conversation in which I reflected with Claude on the themes which had emerged in the writing, I asked it to “produce five propositions about generative AI and academic writing, building entirely on the contents of my book, replicating my own writing voice to the greatest extent possible.” I experienced what I was presented with as utterly uncanny, in the sense that these were my arguments written in my style but I did not in fact write them:
- The advent of generative AI doesn’t just pose practical challenges for academic writing; it forces us to confront fundamental questions about why we write and what we hope to achieve through our writing. These are questions which have always been there, lurking beneath the surface of our practice, but which we’ve too often neglected to articulate clearly.
- There’s a real risk that the instrumental use of generative AI will exacerbate existing pathologies within academic writing culture, leading to a flood of mediocre outputs which are ‘good enough’ to meet institutional metrics but lack the spark of genuine intellectual engagement. This isn’t just about the quality of individual pieces; it’s about the cumulative effect on our knowledge systems.
- The distinction between functional and expressive documents in academic writing becomes crucial when considering the role of generative AI. While there might be a place for AI assistance in producing functional texts, like abstracts or summaries, we should be extremely wary of outsourcing the creative, expressive aspects of our writing to these systems.
- Generative AI offers us an opportunity to recalibrate our relationship with writing, potentially helping us to cultivate what I call an ‘ecology of ideas’. Used thoughtfully, these tools can support us in capturing fleeting thoughts, making connections, and elaborating on our ideas in ways that enhance rather than replace our own creative processes.
- The joyfulness we find (or don’t find) in our writing practice isn’t just a personal matter; it’s likely to be a key factor in determining whether generative AI becomes a tool for creative augmentation or a crutch for increased productivity at the expense of quality. Cultivating this joy isn’t a luxury – it’s an essential part of maintaining the integrity of academic knowledge production in the face of technological change.
I thought this was an outstanding summary of what I’d argued over the messy draft I had provided to Claude. It picked out the most salient features of my developing manuscript and distilled the core arguments I was trying to make, with a clarity and concision which eluded me at that stage of the project. What was odd though was seeing the characteristics of my own writing reflected back at me, from the writing tics (e.g. ‘real risk’) through to my favourite words (e.g. “cultivating”) and even the rhythms of my sentences. It was the first time I had the experience of recognizing myself in writing that I hadn’t actually written.
It conflicts with what Connor terms writing as autonomous knowledge, in which ideas come to be lodged in more or less impersonal forms through being written down. In this case the writing is impersonal but the manner in which it is recognizably my ideas, in my style, without it feeling like plagiarism complicates the assumptions we bring to our writing. It doesn’t feel like Claude has stolen my ideas, as much as that it has occupied my position as author. I found it profoundly eery, particularly when I gave it some critical feedback partly motivated I think by a desire to explain away what it presented me with. The result was a longer list of arguments which even more vividly felt like my own writing.
I mean eery here in Mark Fisher’s sense of a “failure of presence” in which “there is nothing present when there should be something.” He illustrates the point with the example of eery scenes in post-apocalyptic science fiction, in which we are presented with vivid scenes of depopulated cities without a clear explanation of where the people have gone. He suggests the eeriness of these scenes diminishes in proportion to the explanation we are given. It is the mystery of the absence, what should be there yet isn’t, which makes them so powerful. I found this experience of being presented with ‘my’ arguments eery because it felt I should be there as an author behind what I was reading. Yet I wasn’t there. It was mine yet I wasn’t there.
In case this sounds unhelpfully oblique, I’ll get to the point of why I’m sharing the experience. In this confrontation with the fact that ‘my’ ideas could be written in ‘my’ style by the machine, I could immediately see how Claude could with minimal prompting finish this book for me.
In the moment the project felt utterly tractable to me. The sprawling mess of the first half of a manuscript, in which I could barely see the wood for the trees, immediately gave way to a feeling of the project being within my grasp. It felt so deeply and profoundly achievable through Claude’s intervention. This felt like it should have been a positive feeling but it really wasn’t. In that moment I felt my motivation for the project drain away, almost as if I could literally feel it leaving my body. That sense of the difficulty, the struggle with inarticulacy which characterizes the writing process for me, suddenly appeared as something which I could switch off at will.
Even if I could see that intellectually the reality was more mundane than that, it felt in that moment that something profound had shifted in how I related to this writing project. It was not all writing projects because I could see that Claude could only produce what it had because of the sheer amount of content which I had provided it with. Not just the draft manuscript and the stack of blog posts which preceded it but the intellectual conversations which had accompanied these. This left my grasp on the early stages of a writing project fully intact, in which I could revel in the feeling of making connections and putting things into words. But after a certain point, I could now see it as a technical exercise in which I simply had to provide enough context and raw material for machine writing to bring things to a close.
There was a sharp sense of potential relief, as if I was on the verge of being able to take off overly tight shoes which had been troubling me throughout the day, coupled with a dawning melancholy. While the prospect of unburdening was enticing it also appeared to me as a profound loss, in which something would be gone which I might never recover. As if once I’d taken those shoes off I could never put them back on, nor could I walk outside in them or explore terrain that I had once taken for granted.
If we lack joy in our writing then we will find machine writing particularly enticing, offering an immediate solution to our difficulties and enabling us to produce more in less time. The problem is that our joy will further recede if we rely upon machine writing in this way, depriving us of the difficulties through which we struggle to develop as authors. Furthermore, it will entrench the instrumentalism which collapses the horizon of that joy by making writing a matter of producing things to be counted rather than to be read.
This experience has convinced me that we stand at a crossroads. The path of least resistance, using AI to circumvent the struggle of writing, may seem appealing, but leads to a diminished relationship with our own ideas. The harder path, engaging with AI as a partner while maintaining our commitment to the difficult joy of writing, offers something richer: a way to preserve what’s valuable about academic writing while embracing new possibilities. The eeriness I felt when seeing my voice reflected back at me now serves as a useful warning: when we outsource too much of ourselves to the machine, something essential threatens to disappear.
Mark Carrigan 