The algorithm’s autograph: Can Artificial Intelligence ever become an ‘author’?

Writing is a form of artificial intelligence. That is pretty much what Socrates felt about the new technology of transcription, about which he was rather suspicious. Plato’s reconstructed teacher did not actually call it AI, of course. But he described writing in terms that mysteriously evoke the current reality and conversations around machine intelligence.

Writing was, for Plato, an “invention” that “will produce forgetfulness in the minds of those who learn to use it, because they will not practice their memory”. People’s “trust in writing, produced by external characters which are no part of themselves, will discourage the use of their own memory within them”. This is gradually how we come to outsource our own human functions to external technology – a process, in today’s language, called “metacognitive laziness”.

He almost sounds like one of today’s teachers, frustrated by the rampant use of AI, when he tells the writer that “you offer your pupils the appearance of wisdom, not true wisdom, for they will read many things without instruction and will therefore seem to know many things, when they are for the most part ignorant and hard to get along with, since they are not wise, but only appear wise”.

Most importantly, the silence of the written word precludes real-time interaction with a teacher physically present through active spoken dialogue. Writing, at best, can only be a feeble copy of the real thing, speech, where the true source of the words is present in real time. Hence, only an orally delivered speech or conversation comprises the original discourse and has a greater philosophical and pedagogical value.

Where is the author?

The privileging of speech over writing, Jacques Derrida has argued, marks Western philosophy as logocentric. It’s a profound argument that is difficult to grasp simply because it questions and seeks to uproot our fundamental assumptions about texts. All texts with linguistic content, Derrida seems to be saying, promise the presence of an author by conflating author with speaker and writing with the speaker’s message. Speech, according to this model, becomes synonymous with the original, with its implication that the speaker is actually present, and writing with the idea of the copy, since it cannot claim the real-time presence of the writer.

Derrida has bigger worries as he tracks logocentrism – the notion of absolute Truth and language’s claim to reflect a stable reality, for instance. But a key concern for Derrida is the metaphysics of authorial presence, which implies a real person is speaking to us through language.

But there is also another lineage of modern authorship where the Author is divine and hence can never be present. In this tradition, the text is the subject of hermeneutic interpretation; that is, an attempt to understand what God might have meant, regardless of whether God offered Their words to a man atop a mountain or on a battlefield. Since no one can ever meet God or even affirm Their real presence, hermeneutic interpretation is endlessly delayed and unfulfilled. It is an unceasingly frustrated attempt to reconstruct meaning from words that are available but inadequate to conceiving a whole.

The secularisation of Europe’s scriptural hermeneutics, Edward Said has reminded us, is the foundation of modern humanities – philology, literary criticism, language-based study of all kinds. The absent source of meaning is no longer God but the amorphous and mercurial character of a literary symbol, the playfulness and equivocation of non-representational writing – anything that wreaks aesthetic violence on language and destroys its transparency as a medium of expression. Meaning becomes elusive again, and interpretation must search for it, just as the reading of scriptures once searched for the absent God.

Stephen Dedalus wants to fly his flag for both camps, and, consequently, commit to neither. Somewhere in The Portrait of an Artist as a Young Man, James Joyce’s famous autobiographical character demands that the author be like God in relation to their creation – present everywhere but visible nowhere. Omnipresent but invisible. This is not about the formal ascription of authorship, which, as Michel Foucault and others have reminded us, is hard to escape in modern times.

Today it’s impossible not to have an authorial name on the title page, cover and spine, and all the legal and moral copyright claims that go with that name. But Dedalus has something more ethereal in mind, whose tangible dimension manifests itself through style, and the less tangible but deeper form takes shape in the authorial worldview.

Where exactly is the author in the work if they are present everywhere but visible nowhere? This is a good moment to return to the Socratic understanding of writing as an artificial and disembodied technology. It is rather deliciously ironic that all these arguments take shape through conversations: Socrates with Phaedrus, Stephen with his buddies Lynch and Cranley. But yet again, the Greek exchange is fictionally reconstructed, and the Irish, fictionally represented – very likely drawn from Joyce’s real-life walks with his Dublin friends. In this, Stephen’s exchanges are Socratic in form and spirit – the speakers are with each other in real time.

But all hell might have broken loose if Socrates and Stephen were to talk to each other. Stephen, I daresay, would rather frustrate Socrates’ view of writing as a sad mimesis of speech.

It is not just that Stephen is a creature of the European Enlightenment and modernity, with his sensibilities shaped by print, and hence naturally immersed in a culture of writing. It’s really about how he leaves God the Writer hanging. Omnipresent but invisible. What would Socrates do with that? In speech, the speaker is very much visible, or at least certainly audible. Writing holds no material trace of the writer. It is an open-and-shut case of physical absence. Style and worldview play no part in this. Stephen is muddying the waters.

Perhaps Stephen and Socrates would have no quarrel at all. They are, in the end and in the beginning, speaking/writing about different things. But then, does Stephen really believe in the invisible omnipresence of the author? In asking this, I, too, play a presence-absence game: the real question is not about what Stephen believes but what his creator James Joyce does. What, in Joyce’s oeuvre, is the poetics of invisible omnipresence?

The AI ‘author’

Artificial Intelligence would be very interested in this question. Okay, not “interested” – they have no sentience, only sapience, an increasing amount of it as we speak – so let’s say they would be deeply “implicated” in this matter. AI, it is clear, involves no authorial persona at all, being a simulation of the author-effect sustained by the multitudinous potentials of training data shaping Large Language Models (LLMs). In that sense, AI busts the cult of logocentrism. But that author-effect grows more and more convincing in its simulation with each passing day, as AI sapience grows more and more powerful. Sentience plays no role in this trajectory of increasing sapience – that is the nature of the beast called AI – but that’s a whole other discussion.

At present, AI-generated writing, particularly writing that seeks to be literary, is almost the exact reversal of the Dedalus dictum. It is visible absence, as opposed to invisible omnipresence. There is no real author (so much for Godlike presence), but plenty of signatures that make this absent author visible. Socrates has finally won the debate. See what happens when you move utterance from the body? It becomes fake, and worse, you can’t tell that it is.

Neither can, it seems, editors and publishers and jury-members of prestigious literary awards. Just about every week we hear of a book being withdrawn from the market, an author advance cancelled, upon the discovery of AI use by the writer, substantial and significant enough to discredit the author of “writing” it. Enough awards have now been shamed for conferring glory to writing subsequently revealed as AI-generated.

It is clear that certain kinds of stylistic, thematic, narrative signatures are popular with the literary zeitgeist that runs through the book industry and its citizens who have awarded these awards. These signatures are now visible enough to become algorithms that LLMs, trained on endless data drawn from what now seems to be just about every word ever written and spoken in the history of humanity, can identify and master quickly.

The publishable, successful, and prize-winning book (most commonly a novel) is in danger of becoming an algorithmic formula. In the past, it appeared that only human editors, publishers, and certain writers were able to crack the formula. Now AI has done it too – when it comes to patterns, formulae, and algorithms, we already know AI’s capacity for mastering them is way higher than that of puny humans than it’s even worth trying to enumerate.

Can a successful work of art or articulation of thought be identified through a formula or a pattern? The conventional wisdom in literary criticism is that popular genres tend to use the scaffolding of certain narrative patterns. The detective novel, the romance, the action thriller – they need to follow certain structures of lack, mystery, thrill, fulfilment, puzzle, and solution. But even the genre of high tragedy, according to Aristotle, had to abide by the structures of hubris, hamartia, the great tragic fall, leading to the revelation and purgation of anagnorisis and catharsis.

Perhaps the greatness of the tragedy hinged on the unique ways a tragic play embodied these structures. Even modern genres such as crime and spy thrillers have transcended the limitations of their structural scaffolding by bending language and narrative in powerfully idiosyncratic ways, shaping novels such as those written by John le Carré and Graham Greene.

Such formulae and patterns do not belong to any particular author but to collective myths society erects from time to time. The signature of a particular author is something different, something more abstract but tangible all the same – you know it when you see it. It could have the sensory immediacy of a “style” or the ineffable but pervasive reality of an “ethos”. Their presence tells us that writing is not such a poor copy of speech after all – it can be its own thing. That writing is more than a Socratic system of artificial intelligence.

But can AI today emulate this authorial signature? The answer to that, I feel, holds the hint of whether the Socratic nightmare of all writing becoming an artificial system – of, in terms of our current reality, all forms of literature being produced by machine intelligence – will come true.

An experiment

Can AI strike the Dedalusian sweet spot between authorial omnipresence and invisibility? What does it even look like? How does Joyce practice it himself? In Ulysses, authorial invisibility takes the shape of radically different streams of consciousness attached to the three main characters: the flowing, elevated, literary style of Stephen Dedalus, the choppy, staccato viscerality of Leopold Bloom, and finally, the breathlessly spontaneous sensual style of Molly Bloom. The omnipresent author organizes it all but dissolves his own visibility between the distinctive voices of these three. Style and ethos come closer together in the paralysed worldview of Dubliners that holds the boy in the early stories in the same trajectory of failed bildung as it does Gabriel Conroy of ‘The Dead’, stringing together depressed middle-aged men and women in between – Little Chandler, Farrington, Maria. If you ask, AI will brave all of it: on my prompt to write a passage in the ‘style of James Joyce’, Google Gemini offered a choice between what it called ‘the structured prose of Dubliners’, the ‘stream-of-consciousness of Ulysses’, and the ‘wordplay’ of Finnegans Wake.

It gave me all three when I asked for it. To me, the riverrun word-soundplay of the Finnegans Wake-styled passage and the sensory, staccato, short sentence of the Bloomian sensibility in the Ulysses simulation looked more clearly identifiable than the paragraph that evoked the gloomy evening walk along the pavements of Great Brunswick Street which, admittedly, would also be familiar to any serious reader of Dubliners.

What would these passages mean to an imaginary reader who enjoys literature but has not read any Joyce? The imitation of Ulysses and Finnegans Wake would come across as instances of unusual and striking styles. The style of Dubliners, I feel, would be a greater mystery. Less distinctive, but with the hint of a mood and a worldview that is ominous and enigmatic. Omnipresent but invisible? Does Dubliners do it better than the later works?

I try to touch the ineffable. I ask the chatbot to give me passages that follow the ethos of these three books, not their styles. The titles of its products are predictable enough: for Dubliners, it is “The Cage of Paralysis”; for Ulysses, “the Democratic Vastness of a Single Day”; and for Finnegans Wake, “the Eternal Return of the Sleeper”. With Ulysses and the Wake, I get passages that approximate summary and analysis: with the former, a man carrying submerged grief past sandwich-board men, etc., and with the latter, the cyclical dream nature of history and so on and so forth. For the early collection of short stories, it gave me a kitchen smelling of cabbage and grease. Just a pile of sensory details. No analytic detail beyond the title, ‘the cage of paralysis’.

Curiously, it feels that the chatbot has a better conceptual grasp of the two later books, Ulysses and Finnegans Wake. It seems sharper in capturing their styles too. Dubliners, for it, seems more quietly immersive. Somehow, it doesn’t attempt to summarise the book – perhaps other chatbots elsewhere may do that, but what it does for me squares with my sense of Dubliners. It is also understandable that Gemini’s ventriloquism of the styles of the later, more experimental books yields more distinctive stylistic signatures; when it tries to capture the smelly kitchens and twilit horizons of Dubliners, we just get a submerged mood, nothing more.

I don’t want to make the provocative and highly debatable argument that Dubliners is closer to the invisible-omnipotent author than Ulysses or the Wake. A century of Joyce criticism and scholarship has proved Joyce much too complex for such easy claims. Nevertheless, I’m intrigued by the chatbot’s response to my prompts about these three books. Just one chatbot, just one time, I know. Maybe others can try this their way?

AI chatbots know what parody artists have known for a long time: the more obviously distinctive and unique your style of expression, the riper flesh you are for parody. AI chatbots have already become far more sophisticated and resourceful than human caricaturists can ever become. They attempt ineffable signatures as much as they try to emulate styles more distinctive and striking. But the results, I feel, are still markedly different.

The truly omnipresent and invisible author is still delightfully muddling to the chatbot, try as they may. The worship of the truly omnipresent and invisible happens in the truly human act of reading. Literature, as a range of thinkers from Derek Attridge to Michel Chaouli have reminded us, is not a noun but a verb – a singular event, an encounter between text and reader rather than something preset and contained within the text itself.

AI-generated works, or works significantly AI-shaped and inspired, getting celebrated through hyped publication and literary awards, have been the single largest mis-encounter or mis-event in our brave new literary world, a colossal oops! What is most interesting here is not what or how AI can write, but what it says about our education in how to read literature. Have readers of literature worldwide grown more adept at celebrating certain obvious markers of style than enjoying more subtle dialectics of presence and invisibility that make great literature? Is only that which is visible and striking to AI chatbots visible and striking to us? How are we failing to tell the difference so easily?

As a play of the omnipresent and invisible, literature remains beyond the emulation of LLMs, which have a nose for patterns and algorithms that would put the sharpest bloodhounds out of business. They would, for instance, smell very quickly the chutney of hybridised language, a phenomenon to which Amit Chaudhuri drew our attention in an early 21st-century discussion of the Indian novel in English.

Quoting passages from Nirad C Chaudhuri’s biography of Max Müller, Scholar Extraordinary, and VS Naipaul’s novel The Enigma of Arrival, Chaudhuri drew our attention to the evocation of a 19th-century German town in the former and of 20th-century Wiltshire in the latter. Neither of them has any hint that they are written by writers who arrive at these environments in complex and unexpected ways. “Both the passages could have been written by an English writer, in which case they would have been elegant and unremarkable”.

But the fact that one is written by an Indian and the other by an Indo-Caribbean writer, neither of them native to the landscape they describe, makes for a hybridity that is both omnipresent and invisible: “their peculiar meaning, however, operates in the fact that they are the products of hybrid sensibilities and histories”.

What is striking is that “these sensibilities and histories are not present in the writing in any obvious way, but are immanent in it”. The immanent is both omnipresent and invisible. Both (Nirad C) Chaudhuri and Naipaul “create a new language by seemingly reproducing the old, a language of altered meanings in which hybridity and post-coloniality reside like the colour of the snow and the rabbits in Naipaul’s passage, on the border of absence and recognition”.

A reader who understands any element of writing, be it style, ethos or the politics of hybridity, as something tangible through regular patterns is easily fooled by artificial intelligence that gets patterns before you or I can sense them. But literature, and forms of discourse we’ve valued before its modern worlding, have always played this delicate hide-and-seek of presence and visibility. Hence the discomfort of Socrates, and Derrida’s discomfort with that discomfort. Hence Dedalus’s aspiration of the omnipresent but invisible – what Chaudhuri calls “immanent” – as the reality behind authorial signature in great literature. I think chatbots that get this “immanence” are still years away, if they ever do.

The greater and more immediate danger is that our aesthetic faculties become altered so irreparably that we lose the worth of this immanence altogether. The fact that we seem to be already on that sad path is not the fault of AI. It rather speaks to large patterns – dare I say algorithms? – of aesthetic education worldwide.

This essay was first published in Literary Activism.

A novelist and critic, Saikat Majumdar is the author, most recently, of Open Intelligence: Education Between Art and Artificial.

We welcome your comments at letters@scroll.in.

• AI
• authorship
• philosophy