Re-sourcing the Mind

The Convivial Society: Vol. 5, No. 9

Welcome to the Convivial Society, a newsletter exploring the relationship between technology, culture, and the moral life. This post about LLMs, the labor of articulation, and memory began as what I thought would be a brief installment. As if to prove one of the core claims of the essay, that the labor of articulation is itself generative, it grew in the writing. I hope you’ll find some things of use in it.

Cheers,

Michael

---

The founding text of technology criticism is found in one of Plato’s better-known dialogues, the Phaedrus.¹ During the course of Socrates’s conversation about love and rhetoric, he recounts the legend of an Egyptian king named Thamus and an inventor-god named Theuth. Theuth presents a number of inventions to Thamus for his consideration, touting their benefits for the Egyptian people. Among these was the gift of writing, but, surprisingly to Theuth, Thamus was less than enthused about this particular invention.

Here’s how the relevant portion of the dialogue goes. It begins with Theuth declaring,“Here is an accomplishment, my lord the King, which will improve both the wisdom and the memory of the Egyptians. I have discovered a sure receipt for memory and wisdom.”

And here is Thamus’s reply:

“Theuth, my paragon of inventors, the discoverer of an art is not the best judge of the good or harm which will accrue to those who practice it. So it is in this; you, who are the father of writing, have out of fondness for your off-spring attributed to it quite the opposite of its real function. Those who acquire it will cease to exercise their memory and become forgetful; they will rely on writing to bring things to their remembrance by external signs instead of by their own internal resources. What you have discovered is a receipt for recollection, not for memory. And as for wisdom, your pupils will have the reputation for it without the reality: they will receive a quantity of information without proper instruction, and in consequence be thought very knowledgeable when they are for the most part quite ignorant. And because they are filled with the conceit of wisdom instead of real wisdom they will be a burden to society.”

There are two typical responses to the critique of writing Plato here expresses through Socrates. The first is to see this as the prototypical “moral panic” about a new technology. If one takes this view, the best use of this text is to demonstrate how all contemporary tech criticism is similarly misguided and short-sighted. Plato was wrong about writing, thus contemporary critics who adopt the same pattern of analysis are likewise wrong about whatever novel technology they happen to be complaining about.²

The second typical response would be, “Yep, Plato was basically right.”

In this way the passage serves as a Rorschach test for fundamental attitudes about technology.

But there is a third way, of course. Neil Postman, for example, began his discussion of this story by explaining the error of Thamus³:

“The error is not in his claim that writing will damage memory and create false wisdom. It is demonstrable that writing has had such an effect. Thamus’ error is in his believing that writing will be a burden to society and nothing but a burden. For all his wisdom, he fails to imagine what writing’s benefits might be, which, as we know, have been considerable.”

Postman refers to Thamus as a “one-eyed prophet,” seeing only the harms and burdens that a new technology brings. In Postman’s view, however, “We are currently surrounded by throngs of zealous Theuths, one-eyed prophets who see only what new technologies can do and are incapable of imagining what they will undo.”

The point, Postman argued, was to see with both eyes. To recognize both the gains and the losses, the benefits and the burdens. Only then would we be able to judge soundly and wisely. This is, as it turns out, easier said than done. Cycles of hype and criti-hype tend to obscure our collective vision, and we seem to have a predilection for one-eyed prophets.⁴

That said, my purpose in recalling Plato’s critique of writing is to set up a brief consideration of the work that large language models (LLM) like Chat GPT or Gemini promise to do for us, which I take to be, in short, the work of helping us say what we need to say.

I’ve started with Plato because my thesis here is roughly this: the use of LLMs is rendered plausible by the externalization and outsourcing of memory initiated by writing.

Maybe that sounds like an inelegant way of stating something rather obvious, but there are two claims in that thesis, the obvious one and another less obvious, possibly more contentious claim.

First, the obvious one. LLMs work, in part, by mining massive datasets of the written (and then digitized) word and drawing mathematical correlations among the words in these massive datasets in order to make predictions about what words should follow other words in a string. (There are other critical inputs, but this is the relevant bit for now.) Frankly, it is hard not to be impressed by what can be achieved through this method, which I have described inadequately, to be sure. There can be errors of fact, or what are called hallucinations, and the outputs are often soulless. Nonetheless, while breathless agitation about super-intelligence and x-risk is, in my view, misguided, it would be disingenuous to simply shrug a shoulder at the technical achievement. But the key point here is that none of this would have been possible had we not first received the gift of Theuth, the invention of writing, which, as Plato correctly observed, amounts to the externalization of memory.

So, then, in an obvious and uninteresting sense, externalized memory in the form of writing can be understood as the technical precondition of LLMs. But there’s a second, I think more interesting, way of framing externalized memory as a plausibility structure for the use of LLMs.

I’m more interested in what renders the use of LLMs plausible than in what makes them technically possible. The concept of a plausibility structure, drawn from the sociology of religion, is meant to describe social contexts, structures, or conditions that make it easier to hold certain beliefs.⁵ Apart from such structures, a belief may become implausible or untenable. Relatedly, I sometimes find it useful to ask, “What do I have to believe to adopt this or that new technology?”⁶ Or, to put it somewhat differently, “What facts about my social world incline me to adopt a new technology?”

So, in the case of LLMs, we might say that the existing soulless and bureaucratic context of much of our writing—the filling out of forms, thoughtless school exercises, endless email—constitutes a plausibility structure for LLMs. Under such conditions, of course, it becomes perfectly reasonable to adopt a new technology that promised to relieve us of such tasks.

I’m less interested in these cases, however, than I am in the use of LLMs to accomplish what, for the lack of a better word, we might call more personal tasks. Consider, for instance, the anecdote recently shared by

Matthew B. Crawford

in an essay for the Hedgehog Review, which explores some of the same terrain I’m traversing here. Crawford tells of a recent conversation with a father who told him about how he had used Chat GPT to craft a toast for his daughter’s wedding. It’s the use of LLMs for this kind of writing that might be worth considering a bit more deeply, especially because it's abundantly clear that tech companies want us to use their products in this way.⁷

Here too, of course, a relatively straightforward consideration presents itself—writing is hard. Many people find it intimidating, perhaps especially when you’ll be expressing yourself in public as in the case of a wedding toast. As Walter Ong, among others has noted, writing is not natural. While the use of language is natural to the human animal, the emergence of writing was not, strictly speaking, necessary. So if writing does not come easily, why not take up a tool that promises to do it for us, particularly in cases that call for something more personal than inconsequential boilerplate? Part of the response to that question involves showing what might be at stake, which I attempt to do in the next two or three paragraphs. But then I’ll also come back to why I started with Plato and conclude by considering whether there is not also a case of conditioned dependence stemming from our readiness to externalize our memory.

So let’s start with the observation that in these cases LLMs are more than a tool for writing, narrowly understood, because the act of writing is also the more basic act of articulation.⁸ When we turn to an LLM to write for us, we are also inviting it to undertake the more fundamental task of articulation, and this is no small thing. Indeed, given the centrality of language to the human condition, we should wonder about the degree to which the outsourcing of the labor of articulation is the outsourcing of a fundamentally human activity.

To see this more clearly, consider what is entailed in the labor of articulation, and it often is, quite literally, a laborious activity. It is not simply the case that articulating ourselves in language is a matter of matching a set of words to a set of internal pre-existing feelings or inchoate impressions, as if the work of articulation left untouched and unchanged what it was we sought to articulate. Rather, the labor of articulation itself shapes what we think and feel. Articulation is not dictation, articulation constitutes our perception of the world.⁹ To search for a word is not merely to search for a label, the search is interwoven with the very capacity to perceive and understand the thing, idea, or feeling. It is, in fact, generative of thought and feeling, and, ultimately, of who we understand ourselves to be. To articulate is also to interpret, thus it also constitutes the experience of meaning. The labor of articulation binds us to our experience and in relationship with others. The labor of articulation always presupposes the other, and is thus an ethical act that relies on candor, honesty, and attention. And while it is, in part, for the sake of the other that I set out to articulate myself, it is in this way that I also come into focus for myself. If I might be forgiven the analogy, it is through the labor of articulation that the self is birthed.

In the essay I mentioned above, Crawford cited remarks from the philosopher Talbot Brewer in an unpublished paper about what he termed “degenerative AI.” As it happens, I’ve also had occasion to hear some unpublished remarks by Brewer through a friend who attended a recent conference. One phrase in particular caught my attention. As I understood it, Brewer argued that dependence on LLMs took the self “out of play.” This is an evocative way of getting at the matter. In the labor of articulation, we put ourselves in play, with all the risks, rewards, burdens, challenges, and consolations that entails. To outsource the labor of articulation is to sideline ourselves.

So much then for what is at stake in the outsourcing of the labor of articulation. It was an important digression establishing the stakes, but now let’s come back to the main point. When we externalized our memory in the form of writing, we began building the databases upon which LLMs rely. But we also, as Plato argued, began emptying ourselves of the resources upon which the labor of articulation works. Plato was ultimately ahead of his time. It took a good long while for writing to be widely adopted. The residue of oral culture, including its valorization of memory, lingered for millennia. But digital technologies brought us across a critical threshold. The scale and ubiquity of digital databases, the vaunted access they provide to information, the promise of having all human knowledge at our fingertips have made it increasingly likely that people will “rely on writing to bring things to their remembrance by external signs instead of by their own internal resources.”

My contention, then, is that when we are confronted with the opportunity to outsource the labor of articulation, we will find that possibility more tempting to the degree that we experience a sense of incompetency and inadequacy, a sense which may have many sources, not least among which is the failure to stock our mind, heart, and imagination. There was, after all, a reason why memory was one of the five canons of classical rhetoric.¹⁰ It was not just a matter of committing to memory what you had planned to say. It was also a matter of having internal resources to draw on in order to say anything at all. Of course, very few of us have any reason to see ourselves as rhetoricians, except that there may simply be something deeply humane and satisfying about the ability to express oneself well.¹¹

And this is to say nothing of how we might distinguish knowledge from the mere aggregation of disparate, readily accessible facts. Others may distinguish the two differently, but I think of knowledge as something more personal, something that emerges within us as we take in the world from our own unique perspective but also as members of particular communities. In doing so, we construct relationships among the things we come to know (and not merely know about), these relationships are shaped by our history and our desires. And this knowledge, carried within, shapes our ongoing encounters with the world, building a cascading experience of “understanding in light of,” a form of poetic knowledge. But this seems hardly possible if we too readily dismiss the need to curate our memory as carefully as we might curate our feeds.

I am reminded, too, of something the avant-garde playwright Richard Foreman observed many years ago¹²:

I come from a tradition of Western culture, in which the ideal (my ideal) was the complex, dense and “cathedral-like” structure of the highly educated and articulate personality—a man or woman who carried inside themselves a personally constructed and unique version of the entire heritage of the West. But today, I see within us all (myself included) the replacement of complex inner density with a new kind of self-evolving under the pressure of information overload and the technology of the “instantly available.” A new self that needs to contain less and less of an inner repertory of dense cultural inheritance—as we all become “pancake people”—spread wide and thin as we connect with that vast network of information accessed by the mere touch of a button.

My modest suggestion in conclusion is this: perhaps we do well to re-evaluate how we think about memory and what I have called the labor of articulation.

New technologies challenge us. If we are up to the challenge, they give us the opportunity to reconsider things we have taken for granted. They invite us to rethink and recalibrate our assumptions about what it means to be human, perhaps even to reclaim some goods we had lost sight of along the way. LLMs confront us with just such a challenge, and in the vital realm of language no less. If we have assented, in large measure, to the promise of outsourcing our memory and now consequently find ourselves tempted to surrender the labor of articulation. Perhaps the best way to respond to the challenge is to consider how we might deliberately re-source our minds so that we might take up the labor of articulation with confidence and enjoy its very human satisfactions and consolations.

---

The Convivial Society is made possible by readers who value the work and have the means to support it. If that is you, please consider becoming a paid subscriber.

1. I say that somewhat facetiously. Some might take issue with the claim. Maybe there’s another earlier text that better fits the bill.

2. Even if one grants that Plato was wrong about writing, this is a non-sequitur.

3. In Postman’s 1993 book, Technopoly.

4. “Criti-hype” is historian Lee Vinsel’s term for criticism of technology that takes the hype for granted and thus appears as an equally unhelpful inversion of the tech boosterism.

5. To the best of my knowledge, the term was coined by the late sociologist Peter Berger.

6. The relationship can be dialectical. I may adopt certain technologies and find that their use becomes the plausibility structure for the formation of tacit beliefs. In using the tool, I find that I come to believe something about the world or about the self that I would not have otherwise. So it is not simply a matter of what I had to believe to justify my use of a technology, it’s also a question of what I come to believe because of my use of the technology (in order to justify my use, for example).

7. Consider the Google Gemini ad that has run during the Olympics. It features a father using Gemini to help his daughter write a fan letter to an Olympic athlete.

Max Read

had a useful discussion of these ads in his latest installment.

8. I want to acknowledge that writing is a distinct use of language, one that is already informed by a technology, the alphabet. Writing and articulation are not necessarily co-terminous, and articulation in literate societies is already influenced by writing.

9. Some will rightly note echoes of Charles Taylor’s work here.

10. Along with invention, arrangement, style, and delivery.

11. St. Augustine, who was classically trained, wrote movingly of memory: “I come to fields and vast palaces of memory, where are the treasures of innumerable images of all kinds of objects brought in by sense-perception.”

12. These lines were cited by cited by Nicholas Carr near the end of his 2008 essay on some of these very themes of this installment.