9. A Platonic Interlude

You, being the father of written letters, have, out of goodwill, said the opposite of what they can do—for, in the souls of those who’ll learn this, it will produce oblivion, out of a lack of memory [mnēmē] practice: putting their trust in writing through outside marks, it is not from the inside that they will get remembrance [anamimnēskesthai]; you did not find an elixir [pharmakon] of memory but for reminding [hupomnēsis]. And to your students, you are providing wisdom seemingly, not truly: through a lot of hearsay, void of teaching, they will seem to know a lot, while being ignorant for the most part as well as difficult all together, and seemingly wise, rather than wise.1

In the recent brouhaha caused by the wider circulation of generative AI, a few commentators thought it useful to display a remnant of their humanistic education by bringing back Plato’s (or, at least, Socrates’s) critique of writing, expressed in the end of the Phaedrus. Needless to say, the voice of consensus was hostile to the Athenian philosopher, rapidly accused of ignorant technophobia. New credo: in the past, just like today, we should be grateful to progress and accept the way things will be, but ask for fairer, more humane, treatment. Or, in the words signed by an academic (who, clearly, was not trained in Plato’s Academia):

As Plato was wrong to fear the written word as the enemy, we would be wrong to think we should resist a process that allows us to gather information more easily. . . . The way forward is not to just lament supplanted skills, as Plato did, but also to recognize that as more complex skills become essential, our society must equitably educate people to develop them. And then it always goes back to the basics. Value people as people, not just as bundles of skills. And that isn’t something ChatGPT can tell us how to do.2

Nevertheless, these sentences are exactly something ChatGPT would tell us, including the touching appeal to “value people as people” in conclusion, plus, of course, the absolute nonreading of Plato they authoritatively exhibit. If the myth of Theuth in the Phaedrus were nothing more than a mixture of fear, dismay, and lament vis-à-vis a new technique, or a reaction tied to the systemic view of the privileged elite facing anything unsettling the old order, I would go further than my colleague does and say we do not even need to look at Plato, unless we crucially have to demonstrate that the old farts from the present merely echo those of antiquity. Now, if we are making analogies between Plato’s site of enunciation and ours, it is worth noticing that the Phaedrus was composed several centuries after the introduction of writing in Greece and long after it became widespread among educated citizens. Thereby, Plato does not fight a novelty the way generative transformers are for us. Thus, he does not stand firm on a position (orality being superior to the written word) that is doomed to fail because of the advance of progress, and, by the way, he writes against writing, a paradox that has always been obvious to his readers (at least, until the 2020s, apparently). The semantic inflection of pharmakon, both a remedy and a poison, as well as the play on its grammatical construction, inscribe, within the idiomatic, the ambivalence of writing. Moreover, the speech delegation is dazzling: in his written dialogue, Plato attributes to Socrates words on what Thamus and Theuth are “said” to have said to each other about writing.3 This might make us wonder who is speaking, and how, and through which course they are remembered, all issues that are otherwise at stake in that passage. It additionally matters that a different legend circulated in the Greek world about the origin of writing, with the Achaean Palamedes being its inventor, a topic treated in tragedies, before Plato.4 Socrates’s tale on Theuth is a deliberate, metafictitious reconstruction, and Phaedrus quips: “Socrates, you have a facility for creating stories, Egyptian or other.”5 There are also the many connections between this moment on Theuth and the rest of the dialogue, or perhaps even the rest of “philosophy,” as Jacques Derrida famously argues in the section of Dissemination entitled “Plato’s Pharmacy.”6 This is not the place to explain this constellation or to assess the validity of such interpretations, but let me just indicate that the density and richness of Plato’s writing is tersely obliterated when one turns it into an excerpted position against the introduction of new technologies. What kind of novel “complexity” are we speaking of if it begins with being unable to read and with crushing the textual “data?” From there, one could dissent with the organization of the workforce and of the education that the tech industry is pushing for, but the solidarity with the actual disaggregation of the demands of humanistic scholarship is unmistakable.

Anachronistically, however, Plato’s Phaedrus might address some of the challenges that are brought by generative AI. I am not stating that Plato anticipates our current moment, that history repeats itself, that the great minds of yesteryear unearthed eternal truths, or that the logic exhibited in the work would necessarily be paradigmatic for us. I am not suggesting that we should, or even could, map the Platonic reflection onto our contemporary. I advance that we may activate the Phaedrus differently by confronting it to what it could not debate though still responds to, in the moment of our reading. Building on the quotes from the Egyptian king, we could determine the worth of our twenty-first-century automated transformers along the following lines: the new technique might achieve the opposite of what its promoters allege it does; while AI could contribute to the factual extendedness of the human mind, this movement of externalization could have consequences on the way we think (on its way itself, not only on its content);7 “memory” is not one thing, and we might conceptually distinguish it, for instance, from “remembrance” and “reminding,” further locating in the operation that engineers name “memorization” an important difference between automated generation and human textual composition;8 there is a difference between thinking through (“wisdom”) and regurgitating (“providing wisdom seemingly”); where all this happens is “in our souls,” understood not as a mystical entity that would be immortal but, I once proposed, as the potential and collective expansion of our noetic capabilities through our ability to go beyond the limits of thought.9 I am not looking for making mine Plato’s, or even Socrates’s, or even Thamus’s, conceptions of writing. But by grafting generative AI on the Phaedrus, I see emerging arguments that deserve more pondering than the formulaic dismissal of passé technophobia.

Anthropological and psychological observations certainly confirm that literacy directly reshapes human cognitive abilities, even beyond the specific case of memory. If we do not subscribe to the Platonic vision of the autonomous noetic subject reminiscing the true world of Ideas, the change introduced by the technique of writing might look less dramatic. The phenomenon of reshaping does remain. Then, believing that all literacy innovations would participate in a zero-sum game or even in “progress” (goodbye, old skills; hello, new ones) is as puerile as it is unwarranted. The predictive autocorrect function, for instance, is an indication of the AI-mediated destruction of expression and poetics. The instant the machine is not only fixing misspelling but giving us the most probable words, standardization can only ensue. This could be represented as a democratic advance insofar as writing appropriately would now be accessible to all those who could select a turn of phrase—or opt for the best paragraph outputted by the automated generator. The other side of this reality is the concerted effort against unpredictability and the profound normalization of the sayable, according to a patterned description of the said.10 It leads to democracy without free speech. Besides, the downloading of mental activities to computers will univocally fragilize competences in their unassisted form. The bureaucratic operations of the mind could be outsourced with moderate harm, but as for the production of images, texts, sounds, knowledge, do we really wish to put our trust in the state, a private company, or an international consortium to guide us in such endeavors? Finally, the profusion of content automatically reshuffling the data from the past, our immersion in the standardized mind, unless it is actively counterbalanced by an emphasis on maximalist humanistic scholarship, will blunt the faculty to read, appreciate, evaluate what lies outside of the minuscule, computerized, Erwartungshorizont. In this regard, Plato’s form of warning—a complicated, equivocal one—is as evocative as it could be. Being “seemingly wise” could be read as “giving a simulation of wisdom.” With Turing’s standard of imitation games for automated computation, it is fair to consider that, so rooted, “artificial intelligence” is, at most, a simulator.

One of Plato’s errors, of course, was to consider the arts to be mimetic (and reality to be a reflection of Ideas), a thesis Aristotle already corrected. But, beyond Plato’s original dogma and the anachronistic avatar we crafted of him, his admonition should be heard in the arts as well, independently of his indictment of representation, conflated with imitation. It would be irrelevant to issue a blanket ban against the recourse to AI in the creation of oeuvres. First of all, automated entities have been dreamed all along by writers and painters, from Hephaistos’s moving tripod in Homer to George Grosz’s self-portrait as a “pedantic automaton” and beyond.11 Then, obviously, the integration of coding in poetry, the computerized software for editing and modifying images in film, or the structuring of songs through sequencers and samplers are no novelties. The mechanical, as anything else, could be turned into artistic material. This is not to say that any use of artificial intelligence is interesting, valid, or thoughtful, no more in the arts than in other domains. What are we seeking? Is this, for the artist, time efficiency, more easily earned money, a compensation for otherwise poor technical skills? Is it fun and being cool? Is it an attempt to negate in action the existence of poiēsis by exhibiting works that are no worse than others (as if it were showing anything but the existence of human mediocrity, which we may be sure of)? To reproduce more successfully reliable variations on the same images and more plot twists than do MFAs? Or to produce a vision that would otherwise be impossible to represent? To bypass the self-closure of the technological and concentrate on its bugs and impossibilities? To exceed automation by making it run its course up to its undoing by signification? None of these aims are identical. In keeping the topos of the pharmakon, I would say that if AI is seen as an artistic cure (for a lack of time, talent, inspiration), the potion will be bitter. Let us not forget it.