Especially since the arrival of LLMs, there are quite a few pieces that claim “writing is thinking” or portray writing as thinking in a certain way (even Nature featured a piece on it). The idea is that you don’t just jot down what you thought out beforehand. Rather, the writing and rewriting process often produces thoughts that have not been anticipated by the writer. This is crucial in processes of learning, knowledge-production, artistic endeavours, or therapeutic ambitions. Although I don’t seem to have put it quite that way, I am myself a vocal advocate of that claim. In what follows, I don’t want to revoke my earlier claims. But still: I want to convince you that it’s even worthwhile to entertain the idea that much writing isn’t thinking in this sense, and that that might just be fine.

My idea goes back to recent exchanges with my colleagues Lavinia Marin and Timm Lampert. Lavinia pointed out to me that writing papers in the so-called hard and even some social sciences has no epistemic function. Papers get written to communicate work set-ups and results at conferences, to publish and to get funding. Today, Lavinia detailed this and a number of further points in an intriguing blog post (which I advise you to read as a whole). About writing in these sciences, she says the following: “Writing is just this process where you sit down and add up words. The hardest part is sitting down. … The way many academics treat writing today is that writing is not supposed to produce any new knowledge, insights or thinking. Writing is the wrapping of your results. This is at least for large parts of science, so alpha sciences, beta sciences, all that empirical stuff. “Apply ass to chair”, as Austin K Leon likes to say.” Timm took up this idea and suggested that much applied work based on logic, mathematics, and indeed programming with the aid of LLMs might fit this account.

This is of course somewhat anectdotal, but if this picture isn’t entirely off the mark, it does reveal some insights, not least in explaining why the academic responses to Large Language Models are such a mixed bag. The philosopher in me immediately wants to raise his voice and declare how much gets lost in this view on writing. Yet the historian in me cautions me to wait and listen. What does this mean? Here is a hunch: It doesn’t mean that people in the sciences don’t care about thinking. It just might mean that the real (epistemic) action is elsewhere. Not that long ago I stumbled across some intriguing work by Dana Jalobeanu on enacting experiments and recipes, inspiring me to rethink assumptions about early modern texts. As Dana points out in a paper on Della Porta and Bacon, experiments and recipes are records of what has been tried out in a laboratory or kitchen. The actual epistemic work is not in crafting the written text (which might sometimes just be a list of items and acts), but in all the work that precedes the writing and the work or reenactment inspired in the recipient who might try and replicate the promised results, with all the twisting and turning and trying and failing and trying again slightly differently. Once you see this, it becomes clear that the status and expectations from writing in certain sciences are vastly different from the status of writing in the humanities. In the former, much writing seems a much more social and epistemic action oriented endeavour where texts are just intermediearies and starting points for the actual work. In fact, once you see this, it becomes clear that even philosophical classics like the Descartes’ Meditations or Locke’s Essay heavily rely on the collaboration of the readers.

As I see it, this is quite a positive view of non-epistemic writing. And if writing works really like that in a number of domains, then there do seem to be good reasons to outsource such tedious exercises to LLMs. Recognising this rather strong difference between writing in the humanities and a number of sciences is really helpful to understand the different reactions to the advent of a lot of AI developments. Even if many colleagues in the humanities are not entirely skeptical about the merits of LLMs, they are worried about outsourcing too much and giving way to pernicious deskilling. Rightly so, because writing in the humanities really is different.

However, the process of introducing non-epistemic writing practices into the humanities has not begun with LLMs. It began much earlier. As most of us know all too well, there is the endless repetition of talk of the “crisis of humanties” – to which many administrators (it’s always the administrators!) responded, at least since the New Public Management developments in the 80s, by introducing success measures from the sciences into the humanities. Before we knew it, we began to write papers instead of books, grant applications, performance reviews, explained the value of potential applications of our work to various stakeholders and the average taxpayer. While there is nothing wrong with accountability per se, many complained – rightly I think – that the humanities’ competition with the hard and many social sciences is a game we can’t win. As I see it, this development lead to forms of formulaic writing in the humanities that tip the balance in favour of the (in some respects non-epistemic) paper model which I think is pernicious for many endeavours in our field (and should be countered by the reintroduction of the genre of commentary, as I argued earlier). So I think that many worries about recent technological developments are not inspired by LLMs per se, but by the way in which they accelerate an aleady problematic development in the humanities.

What’s more, the message that writing as such can be shortened and largely outsourced is not lost on students who are often already too busy. I am old enough to have grown up in a world where I learned writing by hand and the first thing I typed on a computer was my MA thesis, the authoritative book-like page on the screen making me feel at once proud and like an impostor whose scribblings were seemingly unduly dignified by the miraculous word-processor. But I keep wondering what it’s like for people who have never made this transition and are born into a world where pens and paper are part of a posh hobby like collecting vinyl records perhaps.

I’m not sure how big of a difference this makes. But what I’m reminded of is that we all (still) write, even if we happen to write non-epistemically and outsource like hell, in an environment in which those who are happy enough to have learned the art of reading and writing at school. This social environment with all its implicit forms of collective knowledge matters greatly. For the forms of writing we (used to?) learn – all the different genres of reports, discussions, commentaries, tables, proofs, translations, essays etc., all that jazz – is epistemic writing in that it trained us to think and imagine. Being trained this way, we can also enjoy the luxury of writing non-epistemically now and then, for instance when the action of thinking is happening elsewhere.

Most of all, we shouldn’t forget that meandering and serendipitous research processes inside and outside of the actual writing are a deeply human thing. Seen this way, there is no point in outsourcing, because writing and thinking – just like many other beautiful things that take time and practice to learn – are not there for the purpose of piling up huge quantities of ever more texts or lines in an academic CV. They have, as the saying goes, intrinsic value. As Lavinia highlights in her blog post, it doesn’t just happen in academic contexts but – now perhaps most of all – in our (popular) culture more widely understood:

“Much of the knowledge propagated in the popular culture is exactly from nerds and fans who have no profit mindset, but simply enjoy the sheer act of research. (I am inspired here by Celine Nguyen’s amazing essay). Research is a wonderful experience, just as painting, singing, staging a drama in your backyard, singing in a garage band, sculpting or knitting. It’s a particular form of leading a creative life, which I think we were all meant to lead. Any artistic practice is highly enjoyable in itself and helps us endure this life on earth. And we just need to recognise that reading and writing are such practices that make us more human and happier to be alive. And hence, the benefit, if you must use these economic terms, is simply experiential. Writing by hand, reading slowly are acts of sheer experiential enchantment. Ways of living our intellectual lives. If we insist on becoming more efficient about these practices, we will be refusing to future generations the access to these practices; we may be cutting them off from very human experiences that we shouldn’t hoard for ourselves.”

My daughter Hannah began to indulge in irony probably around the age of six or seven. Depending on the context, she could be quite frustrated or even fearful when she didn’t get it. When she finally would get it, it was often with great relief. Now, at the age of nine, she still sometimes asks, “you meant that ironically, didn’t you?” What’s so upsetting about it? Part of it is that, when she doesn’t get it, the world or I am being unpredictable to her. If she does get it, we’re in sync again. Being “in sync” doesn’t just mean that we share a harmonious mood; it means that we established a cooperative sign use. I produce the sign “I’m actually a famous rock star disguised as your father”; she interprets the sign. But if she doesn’t immediately get the non-literal sense in which the sign can be used, it can be confusing, at least for a moment. The example is light-hearted, but I’m sure you can imagine grimmer versions that are disruptive to the point of triggering feelings of existential threat. – Now, did you notice how I just placed the (cognitive) responsibility on my daughter?

I said, “she didn’t get it”. Instead, I could have said, “I didn’t produce the right kind of sign.” But this move is so common that it’s hard to notice. We assume that “irony” and other figures of speech are well established conventions. Assuming that there are stable linguistic conventions and assuming that speakers and writers used them aptly, we tend to place the burden on the listener or reader, rarely on the producer. This is particularly obvious in online communication. Writing blog posts, for instance, I often wonder why people don’t read what I wrote but rather seem to single out a side issue and run with it. Of course, I’m joking now, readers of my blog are the most competent ones. But what’s going on here anyway?

The ableism involved in charges against (online) readers doesn’t help, but it’s not the issue I want to tackle now. Nor is the point to place the burden on the writer instead. The point is that linguistic conventions don’t just exist ready to use. When I say that someone “didn’t get it” or someone “didn’t express it clearly”, I’m supposing that it is a stable convention that I can hit or miss just like I can sing in or out of tune. But isn’t that the case? Well, it is and it isn’t. Linguistic conventions, I assume with a little help of Ruth Millikan, are stabilized for a reason. They serve certain purposes for the producers and the consumers of signs. (Consumers and producers should not be seen as humans or whole organisms but as cooperative systems, but that’s by the by now.) It’s the purpose or function that keeps the convention alive. If the proximate purpose of a remark is social alignment through irony, then it’s not enough for both parties to “hit” or “get” the convention of irony as such. Rather, the purposes of social alignment have to be connected to the use of irony for both parties, too. Why? Because the purpose of social alignment is what stabilises the convention. “Getting” the convention cannot be separated from participating in the function that that keeps the convention in existence. It’s not enough then to share knowledge of linguistic conventions; we (or our cognitive systems) have to participate in a certain set of common purposes. Such participation has to be established.

Why is that so difficult in online contexts? For Hannah and me, the purposes of social alignment can be established through various ways and sign uses. If one communicative attempt fails, there are thousands of other attempts that make up for it. Generally, in familiar local contexts, conventions are commonly established in relation to joint purposes that, in turn, will stabilize even most idiosyncratic conventions or codes. We form a community, both in the sense of linguistic conventions and the purposes that stabilize the conventions. Given then that we’d interpret and produce certain linguistic items in keeping with joint purposes, we might call this community an interpretive community. By contrast, in online contexts, we often have random or very reduced encounters. Here, it’s not a given whether we share in common purposes. An ironic remark might aim at social alignment, but it might also aim at detachment. While we might recognize familiar linguistic conventions, we cannot be sure that they are used in keeping with the same functions. Someone might write an ironic remark, but for the readers randomly hitting on that remark online the irony might not be related to social alignment. In such situations, we’re at odds. But we’re at odds without noticing it, because what’s in front of us, the written ironic text, doesn’t reveal the mismatch of purposes. (There are many purposes, even purposes driving our own actions, that we are not aware of.) Neither the writers nor the readers are stupid. But they’re out of sync. They do not form an interpretive community.

In fact, online contexts are not only often populated by people out of sync. The playful recontextualization of conventions and memes lends itself not just to jokes and shitposting. It also undermines and destabilizes the cooperation required for forming interpretive communities. If the reassurance available to us in offline local contexts is entirely missing, we are often kept in limbo, yielding frustrated attempts of cooperation in sign use.