In this post I explore the contours of the rapidly accelerating automation of writing in the era of the generative LLM (Large Language Model – tools like ChatGPT). Namely, I want to call attention to how we’re being encouraged to automate what we might call “low prestige” writing, while presumably keeping more prestigious forms of writing human. In using this term, I want to focus on the social assignment of value to writing, and not its inherent value. In fact, I want to discuss how the assignment of prestige vastly underestimates how much low prestige writing matters, and may encourage us to automate away exactly the wrong kind of writing. In other cases, I think our focus on prestige makes us look for the likely impact of automated writing in the wrong places.
The first form of low prestige writing I notice us being encouraged to automate is drafting. I’ve seen any number of academics on social media sharing policies for using ChatGPT that go something like this: “You may use AI tools to help you generate ideas and draft, but not to write final language.” For example, Carl Bergstrom posted on Mastodon that he told his students “that while I do not forbid the use of generative AI in research, I consider it misconduct to turn in AI-generated text as their own writing.” The producers of LLM writing tools, for their part, also seem to embrace this approach. In the press release announcing their new integration of a generative LLM into their Docs product, Google writes that the new tool makes it so that “you can simply type in a topic you’d like to write about, and a draft will be instantly generated for you.” Other language by toolmakers suggests that the drafting or idea generating stage of writing is the correct stage for the use of LLMs. ChatGPT’s home screen suggests “got any creative ideas for a 10 year old’s birthday?” as a possible prompt.
This sort of approach is understandable, and perhaps reflects both the academic custom of asking students to summarize the ideas of others “in their own words” as a test of the students’ understanding and the idea/expression divide in copyright law (in which expressions are protected property but ideas are not, more on that later). However, it tends to reify a status hierarchy in which the finished product of one’s “own writing” is valuable, but all the stages that lead up to that writing are not valuable. We hand in the “finished” draft, and throw the other “messy” drafts away.
In an age of LLM writing, I would argue this status hierarchy is exactly backwards. We know what LLMs are pretty good at: accurately reproducing the formal features of writing, and we know what they are pretty bad at: accurately reproducing factual information. Wouldn’t it be better, in an world with such tools available, to emphasize the importance of writing as thinking and to encourage students (and people more broadly) to draft things out as a way of building their own understanding? Wouldn’t it be better, as educators, to ask students to write messy, unkempt drafts of their own ideas, and then allow them to feed those drafts into an LLM and let the machine help them adapt to unfamiliar genre conventions?
Another sort of low-status writing that we seem eager to automate is the sort of quotidian writing of everyday contact. The Google press release cited above goes on to suggest that their LLM writing tool could be used by “a manager onboarding a new employee, Workspace saves you the time and effort involved in writing that first welcome email.” Microsoft suggests possible prompts like “Write a poem that rhymes for my 8-year-old Jake. He loves dogs and facts about the ocean.” for it’s Bing LLM integration. ChatGPT uses “help me write a short note to introduce myself to my neighbor” as one of the sample prompts for their tool in the blog post announcing it.
This kind of writing: everyday emails, notes, interactions with children is low prestige because it isn’t perceived as “special” or “unique.” Rather, it’s seen as something that “could be done by anyone.” This sort of writing almost never has direct economic value (no one is likely to buy any of the “songs” I made up on the fly for my son when he was an infant, no one is likely to buy an anthology of my work emails) and rarely is seen as having “artistic” merit (nobody is likely to collect either of the examples above for a museum).
And yet, this kind of writing has tremendous meaning. It’s part of the everyday work of care that binds us all together (and which our society routinely undervalues). Do we really want to automate our communications with the people we share our day-to-day lives with? Isn’t it more important that a rhyming poem for an 8 year old be written by someone who loves them, then that it be “formally correct” or even “good?” Isn’t part of the point of an email welcoming a new employee just to show that someone has put some time into acknowledging their existence?
Finally, I think we may be paying too much attention to the possibility of LLM writing replacing human writing in high-prestige spaces of the arts and not enough attention to it’s likely use in much lower prestige spaces. I see any number of major name authors/creators on social media expressing significant concern about the encroachment of LLMs (or other forms of generative machine learning in other media) into their creative fields. To put it bluntly, I don’t think they have much to be worried about. People value the parasocial relationships they have with their favorite authors. They are unlikely to give that up for machine-generated content anytime soon.
At least, in the spaces where author’s names have meaning and prestige associated with them. In other spaces, like fan production and high-volume commercial production (the vast, mostly under-recognized army of authors churning out genre books for the direct to Amazon Kindle market) it seems much more likely that Generative AI will become a significant part of the ecosystem of authorship. Indeed, it’s well on its way to already being that. Fans are eager to use Stable Diffusion and other forms of image generating AI to create fan art. Kindle authors have been engaging with ChatGPT.
Ideas that circulate in these low-prestige spaces are rarely recognized for their cultural contributions, but we know that there is a cycle of appropriation by which they influence more visible and high-prestige forms of culture. George Lucas re-imagined low status serials as “Star Wars,” for example.
What happens to culture when these sorts of creative spaces become semi-automated (this seems likely to happen, fans have eagerly embraced tools for re-appropriating and remixing culture in the past, and Generative AI is mostly another form of remix)? I’m not sure of that at all, but it seems like an important question to be asking.
To sum up, then, I think we need to be thinking more critically about the intersection of prestige, writing, and Generative AI. I would urge us not to simply automate writing tasks away because they are low-prestige, but to think critically about our application of the technology. At the same time, the likely uptake of Generative AI in less visible, less prestigious creative spaces will need to be paid attention to and investigated more thoroughly.