The Day the Author Died 

Postmodernists who bewail Generative AI should ask themselves why 

Back in 2018, I wrote a response to a piece in The Poetry Review by Jack Underwood called ‘On Poetry and Uncertain Subjects’. Underwood had written about the uncertainty of poetic knowledge, and how a poem existed only as the various meanings given to it by its readers. I reflected that however true that readerly instability may be, there must be some stable element always remaining in that text, something of the author who wrote it, I called it the text’s ‘linguistic DNA’. Underwood responded to my response, saying that the author is not stable and that in language the subject is always in a process of ‘becoming’. Linguistic DNA did not, he told me, exist; in fact, the writer themself was nothing more than a text which required constant reinterpretation.  

I came away, as so often, with a feeling that in order to take a certain view of what a poem is you must take a certain view of what a person is. If you disagree with someone on the latter, you are unlikely to agree on the former. On reflection, I think I agreed with Underwood that we constantly reinterpret ourselves and others, and to that extent we are impossible to pin down – the inherent uncertainty of the Self is a big part of why we cleave so determinedly to political frameworks and moral certainties. They provide us with an illusion of stability where none exists. We are all memory and projection, making ourselves up over and over. Why should poems be any different? 

This was all very well, and as I say there is a lot to admire and agree with in Underwood’s essay, but it left me uncomfortable and for long time I wondered where that discomfort was coming from. Then ChatGPT came along, and suddenly everything seemed to come into focus.  

In my abovementioned blog, I brought up the subject of Roland Barthes’ Death of the Author as it relates to these ideas of instability and the locus of meaning-making in texts. For Barthes, the author was nothing, the reader everything; and Underwood’s analysis seems to make the same claim – any point of stability that we might designate ‘The Author’ simply disappears in a puff of smoke, replaced by a multiplicity of shifting and subjective interpretations. A thrillingly postmodern view of what it means to know a text. 

But now, here comes generative AI with its Large Language Models that can generate texts of increasing complexity and nuance, literally without the need for an author. And it seems to me that we finally have on the near horizon the possibility of the actual death of the author – a death which when it comes down to it, Barthes was only fantasising about. Now we can really have stories and poems whose meaning really lies only with the reader. Barthesians should celebrate, should they not? Or if they are not, they should at least ask themselves why they are not pleased that we can finally bid farewell to that outmoded and unfashionable concept of ‘authorial intent’. 

The truth is, and I don’t think I need to point out the wider cultural relevance of this, theoretical musings on what something is or isn’t make for wonderful philosophy, but they don’t seem so much fun when that thing is actually faced with imminent destruction. Suddenly all the old, simpler, more unfashionably obvious definitions seem important again. 

My instincts, as a borderline postmodernist, are that I don’t want AI to replace human authorship, I don’t want engineers to generate quickly adjustable prompts to run algorithms that will generate novels, short stories and poems in seconds (although it is arguable that there is a certain artistry in this), and I don’t want the texts that I hold dear as expressions of humanity to be reduced to reader-only ‘objects’. But why, if I have sympathy for many of the central tenets of postmodernism, should I care? 

There’s a coherent argument that generative AI texts are expressions of humanity, because LLMs use neural networks to machine-learn from disparate examples of human expression to produce outputs in the same way that an individual human processes a lifetime of experiences to produce outputs. 

But something doesn’t feel right, does it? Machine-learning human expressions is not the same as processing human experiences. And what generative AI texts are missing is nothing to do with their subtlety, nuance, complexity, beauty – all of which may be attributes a reader applies to them perfectly legitimately whether machine or human generated – but it is everything to do with intentionality. 

You don’t need to go as far as Nick Cave’s comically overblown comments on “notions of creative struggle” and “fast-tracking the commodification of the human spirit” to agree with his basic point that there is a disconnect between a text created by generative AI and anything that we would feel comfortable calling a piece of art. What is lacking is not necessarily the tortured sweat of a struggling bohemian but the simple human intention to produce.  

This isn’t just the intention to produce a complete artefact either, a prompt engineer or the person who presses the ‘print’ button on a computer has that level of intentionality; a piece of art, for it to deserve the name, requires a process of intentionality – a series, or a web, of interlinked and concomitant choices that have been made and which result in a single cultural artefact which is then available for multiple and shifting interpretations. This is the work of an artist, or in literary texts, an author. 

But the postmodernist might complain, we have slipped here from ‘text’ to ‘art’; which is like slipping from black to white, or from subjective to objective. Quoting Nick Cave’s bombast takes us in the wrong direction, they might say, it distorts the situation by giving the Ego flesh and blood, giving a living human identity to exactly that signifier from which those attributes are stripped by the act of writing, “…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.” as Barthes put it. And what he did in killing off the author was nothing less than shedding symbolically the Modern in favour of the Postmodern, where there is no fixed ‘truth’ and so there is no fixed ‘artist’ or ‘author’, they themselves are texts, cultural objects with a multiplicity of readers who each carry with them their own version of the truth. 

I mentioned above that I have sympathy with this view. And it perhaps goes without saying that the author can only die if they have previously lived. Barthes not only kills off the previously assumed notion of the author being the sole dispenser of a text’s meaning, but he also makes the assumption that any given text must, literally speaking, have a writer, i.e. a ‘subject-who-writes-it’, who only then symbolically dies, having written the text (or more accurately whose very act of writing the text causes them to pass from subject to object). Generative AI represents a future Barthes could not have realistically anticipated, one in which texts do not have authors, and so they cannot die. It is a future in which the premise of his essay, that texts have authors which can be denied their authority, is itself denied. 

So, my inner postmodernist reprimands me, there is no reason a Barthesian should not lament the effect of generative AI on literary authorship – to suggest as much confuses the Death of the Author with the Annihilation of the Creative Process, which Barthes was never interested in doing. 

But this still leaves a question: if the act of writing causes the authorial subject to become an object, why would we mourn the loss of that subject in the first place? What is wrong, since meaning’s construction is solely in the hands of the reader, with considering a poem in the same way as we might consider a stone used as a paperweight? It is what it is, in its context, uncreated, but selected from others and used to fulfil a purpose. We can ascribe meaning to such objects, read their context, their surroundings, the criteria for their selection. Why should we worry that there was not an actual ‘creator’? 

Well, because there is a difference between a stone and a poem. However beautiful, intricate etc the stone may be, however astounding the millions of years it took to fashion it, there was no intention behind the process of creation; whereas the poem had a poet. And the poet had an idea. 

The intention to realise an idea, the will to do one-thing-and-not-another in intimately linked succession, projecting forward with some kind of, albeit abstract, hope for an end product, this remains with the text after it is written, and we are justified in analysing authorial intent, as long as the reader makes the assumption, as Barthes did, that the text has an author. If we do not make that assumption, then what I called the text’s linguistic DNA is snuffed out, but at that point we no longer have a text, we have a stone-like object. 

So now my inner classical pragmatist speaks up and claims that the postmodernist who mourns the real death of the author at the hands of generative AI does not believe their own doctrine. They want with their reason to rid the text of authorial intent but with their instinct they desire to retain the warmth of human intentionality at a texts core.  

Which brings me back round to my earlier comment that in order to take a certain view of what a poem is you must take a certain view of what a person is. Postmodernism does not account for people as homo duplex, as Durkheim had it, that simultaneously singular and multiple creature that we are – both an individual and part of a group, always. Postmodernism turns us into a multiplicity of meanings and denies that we are also a singularity. It has no room for that paradox in its explanations.  

So my current position is this: yes, we are all memory and projection, yes, our identities are multiple, shifting, and historically/culturally formed, and yes, poems are the same. But I maintain that there is a single us-ness to us as well (a ‘more-than-the-sum-of-all-our-parts’), just as there is a single strand of linguistic DNA in a poem which runs directly from the intentions of the author to the perceptions of the reader. In coming to that conclusion, it is for me the Postmodernist who dies, not the Author.  

Let’s be clear, there are many other reasons for poets to be concerned about the rise of generative AI. Will editors ever again be able to read their vast inboxes each week and be sure that what they are reading was written by a person? Will the judges of poetry competitions now always live in fear that their choice poem will turn out to be AI-generated and they will be shamed in front of the whole poetry community? Can poets be sure that their competitors for space in poetry magazines are not using AI to generate poems that are better than their own in a fraction of the time? Whereas images, and pieces of music and film will all in time have ways developed to prove their provenance, literary texts, particularly poems – that uniquely rememberable cultural commodity – will never have such assurances provided. 

My contention is that the postmodernist should not care about any of this because the act of writing kills off the author anyway; my suspicion is that many will.