Writing and Thinking
2023 is the year, apparently, of “enshittification”—of the internet, of platforms, of all sorts of things in our digital lives. This is also my sabbatical year from SFU, an opportunity to stand back and think a little more than I normally would. And it’s also a year when some bits of the digital world started to fragment and peel off in interesting (if not hopeful) directions. Partially because a lot of my reading of late has been about the rise of humanism in the Renaissance, I’m interested in what writing has come to mean today, and whether there’s any point in doing it anymore.
On the stochastic parrots
Re: enshitification: the rise, or at least hype, of ChatGPT and other Large Language Model (LLM) text generators is a significant part of that—or at least it is if we believe that LLM-generated text will soon become a large part of what’s visible and discoverable on the web. Perhaps this is already the case and we’re already there; indeed, when I Google up something now, I am conscious (er, paranoid) that what I’m finding might be GPT-generated pseudo-nonsense – the creeping “gray goo”.
Witness at the same time the prognostications about the impact of “AI” on the “content industries” (to toss out two bullshit labels at once) that suggest that huge chunks of marketing copy, social media content, corporate whitepapers, and even scholarly publications will be machine-generated, at the obvious loss of jobs and work for the people who formerly wrote such things, and the less obvious loss of the trustworthiness of the network as a source of information, inspiration, enlightenment. This is a big concern to many people right now. But such prognostications—to leave aside for a moment the question of their likelihood or accuracy—make a couple of key assumptions. One is that LLM text generators can compose copy as well as the average person employed to do so, which must be at least mostly true, otherwise there’d be no market for the technology, right? Now, if you’ve been paying attention to the debates, you’ve learned that LLMs are not “artificial intelligence” in any form that resembles what we call “intelligence” in humans (or other animals for that matter), that they rather “stochastic parrots” or “text-extrusion machines” able to digest large quantities of textual material, and re-assemble it in (often astonishingly well-structured) textual forms. LLMs do not think, nor do they compose text in the way that we use the term in reference to human rhetoric and composition—at least not unless we begin with a frighteningly impoverished appreciation of human cognition and development.
That second assumption, which underlies and also necessarily follows from the first, is that much of the “content” that we create and circulate online in the economy of ideas and attention and business is pretty dull stuff: it consists fundamentally of “information,” expressed in highly conventional genres and forms of rhetoric and argumentation. If an LLM can score highly on the SATs, if an LLM can write a scientific paper that passes peer review, if it can generate most or all of the “content marketing” bumf that a corporation needs in a year, this tells us that yes, these things can generate text pretty damn well; but it also tells us that these forms of text aren’t actually all that interesting: they are templated, formalized, generic, and largely regurgitated from earlier models. Those models might be the result of a test-taking training set; it might be a set of instrumentation data in a scientific experiment; it might be a CRM-generated, keyword-heavy sense of what one’s customers want to hear about a product. Which is to say that if you think about it for just a second, you can see that the actual intelligence that drives this composition lies elsewhere, outside the system, either upstream or straightforwardly contextual.
In short, if “AI” is going to decimate the “content industries,” two things seem to be the case: first, that the “AI” doesn’t actually have to be very intelligent at all (indeed, LLMs seem to be up to the task); second, “content” is pretty dull stuff: it consists of information (er, data), expressed in formulaic, generic ways.
What a sad indictment! How, given the glorious history of the humanities and the literary tradition, did we get to this place? How is this the world we now live in? Is the 2000-yr history of Western literate culture so easily reduced to chaff?
One might object: “You’re just talking about nonfiction. Fiction is still a creative realm, the expression of the human soul. The AIs aren’t going to write a Booker Prize winner any time soon.” But this is underwhelming, and facile on a number of levels. In the first place, to carve literature into fiction = creative and nonfiction = information is nonsense from the get-go, as a second’s thought about the history of literature demonstrates. In the second place, I’m not convinced we won’t in fact see AI-generated novels, even ones that become bestsellers—which then forces us to distinguish between “popular” and “good,” and we’ve been down such rabbit holes many times before without much benefit.
So, I Googled “greatest works of nonfiction” and Google provided me with an already-digested (by AI?) list, at the top of which was Montaigne’s Essays. That seems to me a good place to start. What is Montaigne doing in those essays? They’re nonfiction, in the sense that they don’t present a made-up world or made-up events, but rather are a reflection of this world and his actual experience and reflections on it—and, importantly, these are the reflections of a literate thinker, steeped in a tradition of writing about experience. Already we’re not far off the kind of prose that GPT is supposedly capable of producing. But while GPT may be “steeped” in a literate tradition (by virtue of its training sets), it has no “experience” nor “reflections” on anything. Montaigne’s essays, then, seem relatively safe from being replaced by AI anytime soon. So what’s different about the Essays from the “content” that GPT threatens? Wouldn’t we also expect marketing copy to be reflections on the real world (or parts of it) and our experience of them? Well, apparently not. Rather large swaths of marketing bumf fail to reach even this basic threshold of written quality; this in itself is not surprising. But that’s an important threshold to hold in mind when we think about AI’s role in our literary universe going forward.
In his excellent recent book On Digital Humanities: Essays and Provocations Stephen Ramsay offers a justification of the humanities as the solution to the troubles posed by the “unexamined life.” This grated on me when I first read it, and I’ve been thinking about it ever since. Because, historically, wasn’t the point of the humanities about the interrogation of the record of written culture—beginning with the early Renaissance rediscovery and championing of the works of classical era? The early humanists’ concern was with language, with philology, with the proper representation of human experience in the written word, whether that was in Greek, classical Latin, medieval Latin, Tuscan, or the myriad vernacular literatures that sprung up in the 16th century and beyond. Being a humanist meant being obsessed with how (and what) to read and interpret correctly, and how to write so that your words would be interpreted correctly and be durably part of this written tradition. It was about reading and writing. If that is, ultimately, equivalent to “the examined life” then my apologies to Stephen Ramsay, but I think there’s a nuance here worth dwelling on.
A focus on writing, on literature, is a particular kind of focus on culture. It is not the same as culture in the wider sense: I mean things like how we cook, how we eat, where we sleep, how we organize markets and exchanges. Literary culture may reflect and express those broader concerns, but if it’s literate, then the task of interpretation (and enjoyment, and of writing the next thing) isn’t just about thinking these thoughts; it’s thinking about how you express them. It’s about considering words, and sequences of words, and style, and genre, and indeed form and circulation (a ‘sociology of texts’). Those aspects of expression are layers of culture in themselves.
Now, we live in a world—for hundreds years already—where our modes of expression have expanded far beyond just the written word, such that image, performance, video, audio, interactivity, and so on can all be considered as part of this (humanist) tradition. And to the extent that people have expanded the sense of “text” to include all sort of other media, I think it’s easier to just include “media studies” in general in the humanities, or (maybe riskier) to say that humanists are media studies scholars. But back to the written word, which although it has many neighbours still has some unique qualities.
It is the unique quality of written language that makes things like LLMs possible. Text—written language—is subject to extremely well-developed formalistic rules that, while not being the whole story by a long shot (and frankly I’m not at all qualified to comment on that), have allowed us for thousands of years to develop curriculum for teaching languages and formally analyzing them, up to and including computational linguistics today, if not LLMs themselves. Text is an odd beast; it’s an expression of “natural” language, which suggests that it is innate and organic and ineffably expressionistic, but it is also formal and rule-based enough that we’ve been able to invent machines—books, printing presses, typewriters, digital computers, and so on—to do things with it.
The application of machines of all these kinds to text has been great for capitalism, and arguably good for democracy, but it may not have that much significance for writing as a mode of thinking, which is what the early humanists were on about, and what I still think is important, even today in the face of LLMs and machine plagiarism. The old cognitive-science model of the “mind as an information-processing device” lingers still, and the AI boosters that are willing to stand up and suggest that maybe we’re just stochastic parrots too are still in thrall to this old idea—as I say, the mechanization of mind is good for capitalism. But the illusion that we unproblematically see the world as it actually is, and merely process our sensory inputs in service of instrumental goals, while pervasive, is OMG such a tragically impoverished conception of human experience.
Learning to write again
I mentioned I’m on sabbatical this year. One of my goals is to get back in the habit of writing again. Because post-pandemic, I have really fallen down in this area: the day-to-day demands of my (now relinquished!) administrative post and the general increase in the cost of everything (I don’t even mean money) made a major dent in my capacity, and desire, to write. So here I am (literally): launching this blog in the summer of 2023 was a direct attempt to scaffold myself back into writing again: writing as practice, writing as exercise, writing as a means to sorting out what I think about things.
In the winter of 2022–23, I also noticed a shift, a response perhaps to the ‘enshittification’ that many of us noticed when Musk took over Twitter. I’d given up Twitter years ago, but when a number of my circle moved in to Mastodon last winter, I did too, and have been happily ‘tooting’ (sorry) there ever since. The migration to Mastodon was partly inspired by the appeal of a decentralized (or federated) system instead of a centralized system revolving around a billionaire or a nest of venture capitalists. This isn’t simply a lefty move; those of us who remember the early Web remember that decentralization was a big part of the appeal of the online world in the days before Facebook, etc. That idea that anyone and everyone could have a website, and that the network would make it so that we existed in a complex web of connections to one another, was a powerful idea—one that will be familiar to anyone who was online in the 90s, and perhaps not familiar to those who came online in the era of social media.
The appeal of a federated Mastodon was echoed in a tiny little resurgence of blogging in my social sphere, and a return to good old RSS feeds (RSS has powered podcasts for the past decade, but when Google discontinued their “Reader” software in 2013 they also largely killed off the use of RSS for blogs and news sites—there’s a lesson in platform dependency for you). I noticed a little surge of people going back to where we’d been before the days when doomscrolling became the norm. The web itself, we noticed, wasn’t actually broken. It was just that we’d been doing it wrong for some time. And there’s nothing stopping us from going to back to a decentralized, small-parts-loosely-joined model. Nothing except cynicism.
Of course, blogs have been there all along, regardless of whether we had Google Reader to follow them all. People used their social media accounts to point to new posts. Or else they just didn’t. Loads of bloggers have been just writing away, for small audiences, all along. There’s nothing wrong with that, especially if you weren’t in it solely for monetization or brand extension or whatever; maybe there are other reasons to want to write. Then again, a lot of writers have turned to things like Substack, but if the point is writing and thinking, and having a small audience who appreciates it… Substack exists for a different set of reasons having to do with making a living from writing, which is not the same thing as writing. I make a living, on some level, from writing scholarly articles (the publication of which justifies my existence as a faculty member at a university), but it would be a mistake to conflate the thinking and the publishing of these things; they are necessarily entangled because we all live in a world of money, but ends and means are not the same thing and they do not straightforwardly justify one another. Necessary but not sufficient.
Back in the day, before the internet, before the wheel, before the duchess-faced horse, people wrote letters to one another. Letters were not just a way of keeping in touch; they were a whole genre of expression. The post was a literary form, and people have, for ages, collected and curated the letters of all sorts of historical personages. We don’t write letters anymore, mostly, I think, because we’re all drowning in our respective in-boxes and can’t stomach the thought of writing another email, let alone one expressing our heartfelt sentiments and thoughtful reflections. My ideal of blogging, though, is not far off this epistolary model: our addressee might no longer be a singular person, it’s generalized and fragmented somewhat now, but the mode is still personal, somehow. It’s also social, in a way that the network makes possible and which can still be convivial—if we keep our eye on the ball.
How far have we come from LLMs and the “textpocalyse” that Matt Kirschenbaum warned us of? My bottom line is this: we are reminded that text is cheap. But we already knew that; the written word has always been in an awkward relationship with capitalism, as publishers know all too well and very much despite the rose-coloured nostalgia of authors for the good old days when the “midlist” could supposedly generate a good living and social status to boot. But no, text is cheap; I’ll embrace that, and I propose to use my not-so-precious words as means rather than ends.
PS. It occurs to me that I am neglecting the use of generative AI as a tool that provokes or helps distill actual human creativity, and I’ve read a number of intriguing perspectives on this recently. The trouble with most of them is that they have a sticky “pathetic fallacy” feel to them, in which the AI appears like a cute sidekick… the unsaid implication is that the kid will grow up to be somebody one day. It’s not that people actually believe that, but it’s hard to import half of a trope without becoming indebted to the rest of it.. But the larger story here—about “distributed cognition”—is important: we don’t think our thoughts like brains in jars; we are embodied, and we have an external world, and we have tools, and our thinking, our cognition, our mind, is based partly on what our brains can do and partly on what our environment provides, and the clever ways humans have worked out to bridge the gap. Obviously, generative text technologies open a whole new world for distributed cognition. But let’s remember where humanity is. I particularly like Sean Michaels’ warning about how and where we place our trust in AI tools: “People will be suitably skeptical if they deal with crazy AI every day; they’re more likely to remain creative if they assemble their own weird array of tools.”
Obviously this post is hardly the last word on AI, LLMs, or even how I feel about them. Seriously, ask me again next week.
The term seems to have come from Cory Doctorow, who wrote this https://pluralistic.net/2023/01/21/potemkin-ai/#hey-guys in January 2023, at the beginning of a whole wave of corporate mask-removal of this kind. ↩︎
These terms are generally credited to Dr Emily Bender, a linguist. For a gentle intro, see Elizabeth Weil’s “You Are Not a Parrot” in New York Mag March 2023. https://nymag.com/intelligencer/article/ai-artificial-intelligence-chatbots-emily-m-bender.html. ↩︎
see, unfortunately, Silicon Valley. ↩︎
When I was in grad school at UBC, Dr Karen Meyer played the role of coach: “Writing a dissertation is like running a marathon; you have to train for it!” ↩︎
“If we are to go forward, we must go back…” - Martin Luther King Jr. ↩︎
see https://www.theatlantic.com/technology/archive/2023/03/ai-chatgpt-writing-language-models/673318/ ↩︎
See Karawynn Long’s interesting “Language Is a Poor Heuristic For Intelligence,” https://karawynn.substack.com/p/language-is-a-poor-heuristic-for ↩︎
Sean Michaels, “Chat’s Entertainment” https://thebaffler.com/latest/chats-entertainment-michaels?utm_source=substack&utm_medium=email Aug 24, 2023. ↩︎