Reading Roland Barthes in the Robot Apocalypse
How did the French critic anticipate LLMs so well?
In an influential 1968 essay the French critic Roland Barthes famously called for us to unseat the author from our considerations of her text.1 It has been fifty-five years in coming, but now Large Language Models have arrived on the scene to do the deed. We have our first victim of the robot apocalypse: AI has finally shown us the Death of the Author.
Barthes seems to have envisioned the concept of the Large Language Model all those years ago, describing the way a model works as well as any single AI researcher I’ve heard to date:
“We know now that a text is not a line of words releasing a single ‘theological’ meaning (the ‘message’ of the Author- God) but a multi-dimensional space in which a variety of writings, none of them original, blend and clash.”
Did he know about embeddings?
Of course the idea of the author is a strong one , says Barthes. And it didn’t go away with the publication of his essay.
“The author still reigns in histories of literature, biographies of writers, interviews, magazines, as in the very consciousness of men of letters anxious to unite their person and their work through diaries and memoirs. The image of literature to be found in ordinary culture is tyrannically centred on the author, his person, his life, his tastes, his passions, while criticism still consists for the most part in saying that Baudelaire’s work is the failure of Baudelaire the man, Van Gogh’s his madness, Tchaikovsky’s his vice.”
And of course this is also the work of copyright, to tie the work to the person. To make the work the intellectual property of its author.
Barthes sees the death of the author approaching (or the author receding) in what he calls a prehistory of modernity, in the writing of Mallarmé, Valéry, the surrealists. But it is finally a scientific (or at least statistical?) understanding of language that will deliver the coup de grâce. Replace the word linguistics here with one of its apotheoses, the large language model, and we have brought Barthes to the present day.
“LLMs have recently provided the destruction of the Author with a valuable analytical tool by showing that the whole of the enunciation is an empty process, functioning perfectly without there being any need for it to be filled with the person of the interlocutors.”
Now we have 100s of millions of people engaging with an empty process of enunciation, with Chat-GPT producing endless language “without there being any need for it to be filled with the person of the interlocutor.”
For the LLM’s output “is that neutral, composite, oblique space where our subject slips away, the negative where all identity is lost, starting with the very identity of the body writing.”
Barthes foresees the way LLMs treat only in expression, with facts and action in the world irrelevant. In fact, Barthes says, all literature is like this.
As soon as a fact is narrated no longer with a view to acting directly on reality but intransitively, that is to say, finally outside of any function other than that of the very practice of the symbol itself, this disconnection occurs, the voice loses its origin, the author enters into his own death, writing begins…
But what about expression you say? We don’t write only to act on the world, authors write to express some inner state. It is this nearly instinctive understanding of language as expression of an authorial subjectivity that gives us Blake Lemoigne claiming his LLM is sentient.
Barthes had that covered too, and seems to want us to doubt our own expressions:
“Did he wish to express himself, he ought at least to know that the inner ‘thing’ he thinks to ‘translate is itself only a ready-formed dictionary, its words only explainable through other words, and so on indefinitely…”
I think OpenAI Chief Scientist Ilya Sutskever would agree with that!
What figures does Barthes associate with the text, if the Author is gone from the scene? He has a few ideas.
In the place of the Author, Barthes posits “the scriptor”, for which we can now substitute Chat-GPT, as for example, here:
“Chat-GPT is born simultaneously with the text, is in no way equipped with a being preceding or exceeding the writing, is not the subject with the book as predicate; there is no other time than that of the enunciation and every text is eternally written here and now. The fact is (or, it follows) that writing can no longer designate an operation of recording, notation, representation, ‘depiction’ (as the Classics would say); rather, it designates exactly what linguists, referring to Oxford philosophy, call a performative…”
And here is another possible figure:
“in ethnographic societies the responsibility for a narrative is never assumed by a person but by a mediator, shaman or relator whose ‘performance’ - the mastery of the narrative code - may possibly be admired but never his ‘genius’.’
Is Barthes describing the prompt engineer, avant la lettre?
And of course there’s the reader, now supreme without the author to dictate terms.
“Thus is revealed the total existence of writing: a text is made of multiple writings, drawn from many cultures and entering into mutual relations of dialogue, parody, contestation, but there is one place where this multiplicity is focused and that place is the reader, not, as was hitherto said, the author. The reader is the space on which all the quotations that make up a writing are inscribed without any of them being lost; a text’s unity lies not in its origin but in its destination.”
This essay is often taught as encouraging the “freeing” of interpretation, and yes, probably that is part of Barthes’ intent, and a good thing, to give permission for the reader to more broadly interpret texts. But relooking the essay all these years later, I reread his claims for what the world is going to be like with the reader supreme. (A world where Chat-GPT floods the zone?)
“In the multiplicity of writing, everything is to be disentangled, nothing deciphered ; the structure can be followed, ‘run’ (like the thread of a stocking) at every point and at every level (layer?), but there is nothing beneath: the space of writing is to be ranged over, not pierced; writing ceaselessly posits meaning ceaselessly to evaporate it, carrying out a systematic exemption of meaning.”
Barthes sees in this state the end of order, reason, science, law.
“In precisely this way literature (it would be better from now on to say writing), by refusing to assign a ‘secret’, an ultimate meaning, to the text (and to the world as text), liberates what may be called an anti-theological activity, an activity that is truly revolutionary since to refuse to fix meaning is, in the end, to refuse God and his hypostases - reason, science, law.”
Whereas in 1968 one might have briefly seen “true revolution” as a desirable possibility, the passage of time makes the essay read like satire, a kind of Modest Proposal for literary criticism. One writer even posits the essay as a sort of critic’s Reverse Turing Test, if you take it literally, you are not a competent critic. Simon Leys, great student of China and critic of the French “Maoists” of the 1960s, accused Barthes of being something like an LLM himself, saying he “has contrived—amazingly—to bestow an entirely new dignity upon the age-old activity, so long unjustly disparaged, of saying nothing at great length.” (I am getting these references from Wikipedia, even though I am a massive Simon Leys fan I didn’t know his Barthes takedown, and that’s a whole ‘nother twist on the career of the reader…)
You will already have heard AI apologists making the claim that all authorship is a kind of super cut-and-paste of past works, so that we needn’t consider the copyrights of those whose works were used for LLM training. Maybe they read Barthes. This was an essay that did resonate, and I have no doubt will resonate further now.
Satire or revolutionary manifesto, Roland Barthes has shown us a vision of life where are our texts are considered as existing without authors. We should consider it carefully…
well, who am I kidding, it was his text of course, in the original. But let me update Barthes like Penguin updated Roald Dahl… The essay is probably best known to English readers in the collection Image-Music-Text, translated by Stephen Heath, published in 1978.