Generative AI and thriving in creative darkness
What would I do if my work was automated? I was asked this question a year after OpenAI launched ChatGPT and it’s one I’ve returned to frequently as academic use of large language models (LLMs) has become the core focus of my research. If I found myself in a technological utopia where the automation of existing roles meant I could choose freely how to spend my time, I am certain I would continue to write. In fact not only would I continue to write, I suspect what I wrote would be transformed by its changed connection to paid work. There are things we write as academics which are hard to find enjoyment in, even if we recognise them as important.
For example I try to give my students helpful feedback when I grade but it’s certainly not a source of creative fulfilment in my life. I don’t particularly enjoy writing for journals, finding the genre involved limiting and the gatekeeping frustrating, even if I accept it as a required scholarly contribution. There are lots of things about what I wrote, how I wrote it and where chose to publish it which are unavoidably tied up with the requirements of being employed as an academic.
But the writing itself? That has never been work. Nor has it been a hobby. It is something I feel the need to do, a lived imperative which would manifest itself regardless of how I made my living.
This distinction between writing as obligation and writing as imperative lies at the heart of how we might respond to automation in academic life. My relationship with writing has always been tied up with work in complex and sometimes contradictory ways, providing a structure which enabled me to sustain a continual stream of outputs while also trapping me within genres and outlets which were constraining. I was expected to produce certain kinds of academic texts as a condition of my employment but I would undoubtedly continue to write in academic ways, even if automation meant this was no longer required of me. There would be more freedom in this writing, liberating from expectations of genre and the pressing requirement to keep one eye on my REF-ability.
The machine writing which was rapidly becoming a routine feature of my working life, as well as the topic of the keynote which provoked this question, raised the possibility that the production of academic texts (monographs, articles, chapters) on which my job depends could be an increasingly automated process. Even if I remained responsible for developing the ideas and directing the process, the conversational agents could undertake the writing on my behalf. It would be like a superstar professor having access to a vast team of research assistants to support their output. Except in this case my team of assistants could work all day and night, churning out text at a speed which no postdoc could possibly match. There would no expectation of mentoring or support, no claims for co-authorship or recognition, just the constant churning of a figurative printing press pushing out a stream of intellectual commodities with my name on it.
But this scenario raises a fundamental question: if writing is a lived imperative for me, why would I outsource it to machines? This thought experiment forces us to confront what we truly value about academic work.
Why would I do this? I’ve frequently heard myself described as ‘prolific’ as someone who has produced thousands of blog posts, alongside a a non-trivial volume of articles, chapters and monographs. As far as I’ve been able to discern there’s a diffusely positive valence attached to this object. There’s a sense in which quantity, particularly if it is intellectually varied, works as a proxy for creativity. I appear creative to people who might superficially encounter my work because of the range and quantity of my writing. I do think I am creative but as a sociologist I’m attuned to the proxies people rely on when conferring such an honorific onto others. There are judgements we make about perceived patterns which express assumptions about how those patterns come about, even if the underlying reality might prove far more prosaic. There’s a tendency to regard people who produce a lot, in an apparently scatter-gun way, as creative but unfocused. In contrast there are those who are prolific in a hyper focused manner, producing a vast series of tightly connected outputs, which often generates more intellectual respect within their intellectual community. The notion of ‘prolific’ as something which is inherently positive, as opposed to being a contingent outcome of our qualities which are themselves positive, should make us uncomfortable in its implicitly quantitative notion of success. Surely what matters is what and why we produce rather than how much we produce?
This question becomes even more pressing when AI can amplify our productivity. If being ‘prolific’ is valued in academia, what happens when anyone can be prolific with AI assistance? Perhaps this technology is forcing us to reconsider what we actually value about scholarly production.
As Elbow (1981: loc 216) puts it, “Nor can we write a lot unless we get some pleasure from it, and pleasure is unavailable if we wince at everything bad that comes out and stop and try to fix it”. There is a minimal condition for pleasure in writing found in overcoming self-censorship, the impulse to fix and control to avoid imagined slights from others. In his The Interpretation of Dream Freud (1900: 37) approvingly quotes a letter from the poet-philosopher Friedrich Schiller about the “constraint which your intellect imposes upon your imagination”:
Apparently it is not good – and indeed it hinders the creative work of the mind – if the intellect examines too closely the ideas already pouring in, as it were, at the gates. Regarded in isolation, an idea may be quite insignificant, and venturesome in the extreme, but it may acquire importance from an idea which follows it; perhaps, in a certain collocation with other ideas, which may seem equally absurd, it may be capable of furnishing a very serviceable link. The intellect cannot judge all these ideas unless it can retain them until it has considered them in connection with these the other ideas. In the case of a creative mind, it seems to me, the intellect has withdrawn its watches from the gates, and the ideas rush in pell-mell, and only then does it review and inspect the multitude. You worthy critics, or whatever you may call yourselves, are ashamed or afraid of the momentary and passing madness which is found in all real creators, the longer or shorter duration of which distinguishes the thinking artist from the dreamer.
Are you aware of your ‘watchers at the gate’? How much influence do they exercise over your creative process? How does it feel if you encourage them to down their tools for a period of time? Or perhaps see them off all together? The distinction Elbow (1981) draws between the writing and the editing illustrates how the watchers neither could nor should be dispensed with all together. He describes this in terms of the distinct skills of creating and criticising: “writing calls on the ability to create words and ideas out of yourself, but it also calls on the ability to criticize them in order to decide which ones to use” (Elbow 1981: 7). The problem with the watchers on the gate is not the fact they are watching but rather the fact they are on the gate. To the extent they withhold access from the unconventional, surprising or unsettling they are impeding the creative dimension of writing, rendering it impossible to cast the net widely enough that your authorship is supported by the variety of ideas it needs to reach its potential. Their critical role is not to be dispensed with but rather to be tamed, to be cast in a more specialised function which comes later in the writing process.
If you come to Conversational Agents with a sense of what you are trying to say, they can help you say it. If you come to them without knowing what you’re trying to say, the risk is you will rely on them in lieu of your own creativity. It can sound simplistic to state the dichotomy so concretely, yet I will argue this is the dilemma which GenAI poses for creative work. You can either use these as tools for thinking or you can use them as substitutes. It matters which of these we choose as individuals and as professionals.
The real promise for academic writing comes from the capacity to help us elaborate our thinking when those thoughts are already in motion. To take advantage of this requires a comfort with improvisation. It requires a willingness to sit with the discomfort involved in trying to make the inarticulate articulate. It involves an approach to thinking as an iterative process in which we try recurrently to say what we’re trying to say, becoming more comfortable and aware of our intentions in the process. If you can sit with this discomfort, if you can, as Haraway would put it, stay with the trouble, this can be enormously creatively enriching. But if you think in terms of fixed outputs to be pursued as efficiently and effectively as possible, it can in fact shut down this space.
Perhaps the true challenge of AI for academics isn’t whether our work will be automated, but whether we can maintain our relationship to writing as a lived imperative rather than just another task to be optimized.
