Annotations
Whether the scribes saw beyond the unchanged content to the upheaval in its origin, who can say; but we, looking back, can see what they couldn’t: that the revolution was invisible in the output – it lived entirely in the means.
The question that might eventually have come to haunt the scribes of the 15th century – what happens to us when machines can do what we do? – has resurfaced with some vengeance. What happens to writing when the production of prose no longer guarantees the presence of a mind behind what is written?
The answer, if there is one, will possibly be found in what writing has always asked of the person who does it: a willingness to stand behind words, to mean them, and to accept the consequences of having claimed to have written them.
Writing has always been understood as a trace of human thought; when we read, we assume that behind the words lies a consciousness that selected them, a mind that deliberated over their arrangement, a person who stands accountable for their claims.
Writers like Henry James and Mark Twain, who were among the first to compose on typewriters, reported that the machine changed not just how their prose looked but how it felt to produce it. The clatter of keys imposed a different rhythm and a different relationship to revision. Something was lost; something else was gained.
But generative AI represents a rupture of a different order, because, where previous technologies changed how writing was produced or distributed, LLMs change what writing is, or, more precisely, what it can be assumed to be. When a reader encounters a text, they can no longer take for granted that a human being composed it – as long as LLMs exist, there will always be doubt as to whether a piece was entirely written by a human.
This contamination of doubt has spread quickly, most notably online, as the internet, once imagined as a vast library of human knowledge, is filling with synthetic text.
But the question of writing’s future cannot be answered by cataloguing losses. If writing is to survive as something more than a nostalgic practice, it must find a new basis for its value. When it can now be almost entirely simulated by machines, what remains?
Human writing is only partly concerned with the production of words; more essential to its essence is the assumption of responsibility for those words. When a person writes, they are committing themselves, something a language model cannot do. They are saying, in effect: ‘I stand behind this; I am willing to be held accountable for the attempt.’
These functions can now be performed by machines with considerable competence, but what machines cannot do is bear witness or stake a claim grounded in lived experience and personal judgment. Large language models cannot enter into the implicit contract that says: here is a mind engaging with a problem, here is a person who cares about getting it right.
In an environment saturated with synthetic text, this testimonial function becomes newly precious. Readers may stop asking whether a piece is well written and begin asking who wrote it, under what conditions, and why they should be trusted.
The future of writing may look less like the frictionless content economy of the recent past and more like the older, slower forms of correspondence and publication that preceded it. Letters, essays, criticism, investigative journalism, genres where the identity of the writer matters, where readers seek out particular voices and measure what is written against what has been written before.