When Plato tells the story of Theuth in the Phaedrus, the god offers his invention as a gift to humankind. King Thamus declines, with the warning that writing will “implant forgetfulness” and give only “the appearance of wisdom.” The common accusation against AI writing—that it weakens thought, produces imitation rather than understanding, and severs authorship from the living speaker—is the latest form of the same worry.
Derrida’s famous reading of the Phaedrus reframes Thamus’s fear. Writing is not simply a tool added to speech; it is a supplement, both addition and substitute. It appears to aid memory, but only because speech itself is already dependent on spacing, iteration, and deferral—the conditions Derrida names arche-writing. The supplement therefore exposes that the supposed origin (the speaking, remembering subject) was never self-sufficient. Writing does not corrupt presence; it reveals that presence is already trace.
From a neurological perspective, writing does of course literally re-wires the brain. It recruits visual and spatial circuits that oral culture used differently, redistributing the part of the labour of memory from the hippocampus to the page. In this sense, Plato’s complaint is empirically true: writing does change us. But the change is not necessarily degeneration—it can be seen as the exteriorization of the same operation that already structures memory internally. Derrida’s arche-writing here meets Clark and Chalmers’s “Extended Mind”: cognition and recollection extend into the environment through inscriptions that function as parts of the cognitive loop. The notebook, the screen, or the archive is not outside the mind but part of its system of traces.
What AI systems do is generalize this exteriorization. They no longer merely store traces; they process and generate them. The writing machine remembers, recombines, and returns language to us in new configurations. In functional terms it is another layer of the extended mind: a dynamic tertiary retention, in Stiegler’s phrase, that supplements human thought. As alphabetic writing once externalized static memory, AI writing externalizes and increases memory as process: it actively constructs what we call ideas. This extension into process suggests a greater difference than there may actually be. The same structure of the supplement recurs: the aid that threatens to replace, the prosthesis that transforms what it extends.
Each stage—speech, writing, AI—alters neural, social, and cultural patterns, yet none of these abolish the structure of arche-writing itself. The trace remains the constant; the embodiment of the trace shifts. The human, then, is not displaced by technology but continually re-inscribed by it. The history of media is the history of arche-writing writing itself through new substrates—from mouth, to hand, to code. The question is not whether AI will change us (it will) but how we will inhabit the new spacing it opens in the field of memory.
But this is too simple. The notion that the same phantasy or concern exists between speech to writing and writing to AI writing is valid, yet to reiterate Plato was empirically correct in a sense and likewise expressions of concern are likewise correct, because it will alter the human. The issue concerns what it is exactly we think a human is. From a materialist perspective there is little issue here; likewise from a Deleuzo-Guattarian perspective (which is not necessarily materialist) there is also a lack of problem here —humankind simply extends its becoming other possibilities.
This thinking more concerns the phenomenology of the human as it takes itself to be in an incoherent coherence as opposed to its deconstructed coherent incoherence. The incoherent coherence is that of a being of a certain autonomy, possessing its own thoughts and feelings. To place these outside of it have a sense that undermines its sovereign importance. This tension is what is felt (currently) and brings the AI anxiety; literally a threat to perceived human ontology.
There is one more issue, which arguably is more potent than the above. This is that Derrida actually misreads or at least flattens Plato. Derrida treats Plato’s notion of memory more as a cognitve function, but arguably Plato means by anamnesis something much more spiritual. If the Platonic memory is more akin to Bruno’s art of memory, then Plato warns against the loss of a channel further back into being in an unambiguously magickal form. Neural rewiring in this sense is ontologically more than simply a change of cognitive functioning. Likewise then, the more recent shift in which process itself becomes externalised, can be seen as yet more damaging still to this access. From that perspective, every exterior inscription—whether written or algorithmic—is a distraction from the inner act of remembering the Good. If Derrida and Clark show that thought is always already technical, Plato reminds us that it may also be more than technical: a form of recollection that no prosthesis can perform on our behalf.
Without an absolute moral register, we cannot privilege the inner motion or the outer motion. The problem is thus ethico-ontological: the choice concerns not only what we ought to do, but what we choose to be. Ethics comes into play here in the sense of a choice, where we must consider from various angles which one constitutes what we wish to be—the autonomous subject whose access to Being is internal and effortful, or the re-inscribed human whose becoming is always already mediated by the technical trace. The history of media is the history of this ongoing ethical negotiation over the very boundaries of the human self.
Centre For Experimental Ontology 