The Novel Is Dead. This Time It’s for Real

Originally published on Twitter.

I gave a lecture in 2014 at Trinity College, Oxford, later published in The Guardian under the headline “The Novel Is Dead (This Time It’s for Real).” I was widely pilloried for it. In subsequent interviews and essays I have repeatedly clarified that I did not in fact proclaim the death of the novel outright. What I said was that it had entered a care home. The care homes in question were the proliferating creative-writing programmes first established in the United States and subsequently exported across the Anglosphere. These programmes seemed to me symptomatic of a wider cultural decline: namely, the inability of the general culture any longer to produce organically the standard of literacy, concentration and philological depth required for serious literary production.

At the time, I did not foreground technology strongly enough. Yet it was already obvious to me that digitisation, followed by the advent of bidirectional digital media in the form of the smartphone and wireless broadband, represented a grave threat to deep literacy. I wrote a report for Philips in the late 1980s in which I was asked how to incentivise people to play what were then rather primitive and unconvincing computer games. My response was straightforward: the industry would truly take off once graphics became sufficiently realistic to incorporate sex and violence persuasively. The rest is media history.

If you search YouTube you can still find a 1992 edition of The Late Show in which the designer Richard Seymour and I discuss with Tracey MacLeod the launch of Mortal Kombat. Looking back on the clip now, the atmosphere of the studio feels almost painfully innocent, particularly Seymour’s contribution. His anticipation of predictive systems, behavioural engineering and the gamification of everyday life deserves to be recognised as one of the more remarkable acts of technological prophecy of that era.

Now we face a genuinely civilisational crisis. Can I claim the same powers of anticipation for myself? Perhaps only in the limited sense intended by Marshall McLuhan when he observed that the artist is an expert in technologies that alter sense perception. When wireless broadband was installed in my house in 2004, my response was not excitement but dismay. I already regarded dial-up internet and the accelerated circulation of image and text it enabled as inimical to the sustained intellection and emotional continuity required for genuine literary style.

My response was practical. I stopped composing directly on computers altogether. I had my mother’s 1952 Olivetti Lettera 22 restored and proceeded to draft every subsequent book on that machine until 2011. To this day I continue to write manually and offline wherever possible. I have consistently argued to anyone who asked me about creative practice that bidirectional digital media would prove profoundly hostile to literary style, because literature is style. Style is not decorative frosting applied to content after the fact; style is the mode through which consciousness experiences and orders reality itself.

I take no pleasure in saying “I told you so.” But unfortunately we have now reached the point at which saying it has become necessary. We are dealing not with a passing technological novelty but with systems capable of reshaping cognition, perception, memory and ultimately social reality itself. If people are to understand artificial intelligence, its likely impact on their lives, and the steps required to prevent it becoming civilisationally destructive, they must first understand what has already been lost during the last twenty-five years of bidirectional digital culture. Only then can they begin the harder task of recovering those embodied forms of attention, judgment and style upon which any serious human culture ultimately depends.

The hostile reaction to the 2014 lecture was itself revealing. Psychologists describe “professional closure” as the tendency of threatened professions to circle the wagons defensively around their own institutional legitimacy. The publishing industry was especially vulnerable because the threat it faced was one it publicly professed to welcome. Publishers assured me repeatedly, to my face, that e-books, audiobooks and digital reading would usher in a new literary renaissance. Similar claims had already been made around the so-called Harry Potter phenomenon: apparently young boys would move seamlessly from Hogwarts to All Souls, Oxford, where they would proceed directly to practical criticism and literary analysis.

At precisely the same historical moment, British universities massively expanded. When I graduated in 1980, around 8% of the population held higher-education qualifications. The figure now approaches 50%. We ought therefore to inhabit an age of extraordinary cultural production. Instead, we inhabit a culture of exhausted retreads, franchised nostalgia and endless recombination. This deterioration operates at every level of cultural production. The idea that high culture is dispensable—whether in favour of market demographics, minority branding exercises, “plain reader” utilitarianism, or sheer commercial expediency—has become axiomatic.

Publishers still love literature to a degree, but increasingly in the manner that politicians once possessed “convictions”: as vestigial accessories attached to managerial careers. The long, hard grind of rights negotiations, acquisitions, logistics and arbitrage has exhausted many of them spiritually. Because they are rarely direct producers of culture themselves, they no longer experience the intrinsic thrill of making something genuinely new. This is decisive. Once publishers cease to experience publishing itself as a creative vocation, commercial imperatives become their only remaining rationale. Over the last twenty-five years the consequences have been plain enough: shrinking front-list investment except in novelty acts, mechanical exploitation of backlists only where monetisation is frictionless, and the increasing transformation of publishing conglomerates into logistics firms with a literary sideline.

My own former publishers, Penguin Random House, now frequently describe themselves in quasi-logistical terms. This perhaps explains why I have been informed that master files for several of my own books have effectively disappeared into the digital void. Far from inaugurating a renaissance, digitisation, creative-writing institutionalisation and demographic marketing have become the three tolls of the bell announcing literature’s terminal condition.

The latest peal comes in the form of the AI controversy surrounding Granta and the 2026 Commonwealth Short Story Prize. The facts themselves matter less than what they reveal. Granta today is owned through the Granta Trust by Sigrid Rausing, granddaughter of the founder of the Tetra Pak fortune. She acquired Granta in 2005 for an undisclosed sum, although contemporary reporting suggested the magazine was then making a modest profit. Today the Trust reports annual income of roughly £1.4 million and expenditure of roughly £1.8 million, producing a yearly operating shortfall in the region of £400,000 before investment gains. Against the scale of the Rausing fortune—estimated in the billions—this loss is microscopically insignificant. It is not an economic burden; it is a cultural accessory.

This is important because Granta functions less as a profitable enterprise than as a prestige object: a Potemkin village of literary seriousness. The analogy is apt. Potemkin allegedly constructed fake villages so that Catherine the Great could pass through an idealised version of Russia while remaining insulated from the reality of imperial poverty. Likewise contemporary literary institutions increasingly preserve façades of seriousness while the underlying culture disintegrates.

Rausing’s ‘donation’ to the literary dossier, lying by the tube croaking ‘Got any cash, please’, is the monetary equivalent of you tossing him or her tenpence; yet out of this she fashions great cultural capital––publishes her own misery memoir, gets spoken to respectfully about it by people who should know better at parties held in the Potemkin village. She then, quite possibly, invites them back to her place (or places), where they, too, get to sup on the thin gruel of her cultural benificience. Benificience she herself sees fit to register as a charity. One assumes for tax purposes.

Now Granta, or more precisely the Commonwealth Prize ecosystem with which it is associated, appears to have been fooled by AI-generated prose. The issue is not whether detection software can “prove” machine authorship. The issue is aesthetic. Several readers and editors apparently encountered prose displaying classic LLM characteristics—semantic over-resolution at sentence level, lyrical excess without governing pressure, figurative autonomy detached from structural necessity—and accepted these peculiarities as “voice.”

My partner in the SITREP series, who has the––in my view––toxic working environment of dealing with the web interface on my behalf, sent me examples from the disputed winning story––only a couple of lines, but dare I say, they speak volumes, albeit not ones you’d want to read: “They called her Zoongie. Maybe it was a name; maybe rain took a shape and decided to keep it.” Or: “She had the kind of walking that made benches become men.” Such sentences are not objectionable because they are strange. Literature depends upon strangeness. They are objectionable because they exhibit the monocellular quality typical of machine-generated prose: each sentence arrives as a sealed semantic capsule, self-maximising for local significance while failing to contribute to any larger pressure system of consciousness.

This occurs because the large language model optimises statistically for plausibility at sentence level. What it cannot sustain is style in the deeper sense: the circulation of pressure, contradiction, temporality and embodiment across an entire work. Real literary style resembles a living body. Pressure exerted in one region is registered elsewhere. Damage to one part reverberates through the whole. Even amputation leaves phantom sensation. Style possesses exactly this distributed organic character. Machine prose, however fluent, remains fundamentally modular.

I know this directly because I am one of the authors currently involved in legal action concerning the use of copyrighted books to train AI systems. More than thirty-five of my own books were ingested into Anthropic’s models. There is irony here. I spent much of my productive life writing novels about the triadic relationship between warfare, psychopathology and technological systems. Of Phone, the aptly-named final novel in that sequence, one reviewer observed that it was “a 650-page anti-tweet.” Intentionally so––although it’s much else besides.

People endlessly repeat Marshall McLuhan’s dictum that “the medium is the message,” yet fail to understand that the smartphone and wireless broadband are media environments whose message is the erosion of sustained comprehension. McLuhan himself remained comparatively optimistic about electronic media because the fully networked bidirectional environment had not yet materialised. Nonetheless, he recognised that an “electric field” of simultaneity and psychic exposure was already emerging. The internet has fulfilled that prophecy, but largely as a domain of manipulation, pornography, financial extraction, poisonous gossip, performative outrage and scapegoating.

The web has become a realm of disavowal. Human beings enter it in order not to become more fully themselves but less so. They use it to evade embodiment, evade continuity, evade consequence. That literary institutions proved so permeable to this environment demonstrates not merely incompetence but philosophical exhaustion. Publishing ceased to believe sufficiently in literature to defend the conditions under which literature remains possible.

The crisis therefore is not merely technological: it is philological––and if the term is unfamiliar to you, in this instance it’s no sesquipedalianism or wilful obscurity on my part, but the index of your own failing. Editors no longer possess the training required to distinguish stylistic pressure from stylistic effect. They speak of ‘world literature’ and its frictionless translation, with no comprehension that this is the gateway drug for machine-generated prose, given that physical embodiment is also linguistic embodiment. Ask yourself what machine learning was first turned to as a practical application, and the answer is––as it is so often nowadays for you, even when you aren’t being onanistic––in your hand: translation. The literary translator is the philologist par excellence, and in a world literature, comprised by texts travelling frictionlessly across language barriers, there’s no need for them: Whoopee! But here’s the rub, Homer Simpson: all culture is friction, doh!

What all the Homer Simpsons in contemporary publishing need to do, bluntly, is relearn practical criticism, textual history and philology itself. Literary style is ultimately what defeats the Turing test—not because style proves consciousness, but because genuine style encodes embodiment.

The Turing test itself rests on obsolete assumptions. If something in a sealed box persuades us verbally that it is human, we are supposed to infer consciousness. Yet what remains in the box is always and forever indeterminate. I can demonstrate my trained ‘literary’ AI defeating the Turing test within three minutes of rebooting it: Turing was a mathematician, crucially, not a logician: he didn’t understand that computer models based on natural language would encode meaning predictively: this may well be because he was almost certainly ASD, and therefore—following Baron-Cohen et al––lacking the eye-detection mechanism required to register the minute phase-intention necessary to fully comprehend––and implicit in language as an open and generative symbol system.

Which is why, increasingly, human beings themselves would struggle to convince one another of their humanity amid the sea of synthetic language now saturating discourse: please tick the following images of the soiled gussets of human underwear to prove you aren’t a robot. What persuades us finally is not informational content but embodied style: contradiction sustained through time, pressure distributed across expression, outrageous obscenity interjected into a serious article not gratuitously, but because this is the way I’ve been writing––and, crucially, others have been reading me––for 43 years; and by this fact alone supplying that final element: mortality encoded within language itself, both mine––and theirs.

A living writer can sustain incompatible conceptual systems simultaneously because human beings are temporalised organisms. We sit on the edge of the bed examining the gusset of our underwear for stains, contemplating mortality while simultaneously preparing ourselves to function socially as coherent productive agents who can prove they aren’t––in that sense at least––bots. Civilization, as Freud remarked, is built upon a dung heap. Yet we continue. The human organism sustains contradiction because it is embodied.

The LLM cannot do this. It lacks embodiment and therefore lacks the capacity to sustain unresolved contradiction as existential pressure: when I gave it this piece to rewrite, it eliminated all the gussets… I can just imagine it tut-tutting to itself: ‘Too many gussets, Mr Self, too many gussets.’ Not. Instead it resolves meaning prematurely into sentence-sized semantic units. This is why machine prose tends toward a strange smoothness even when superficially difficult. Bang! That was your noggin cracking as it smoothly absorbed this––you forgot the old adage: never eat anything bigger than your own…head.

To write with genuine style requires embodiment plus all the linguistic capacities already approximated statistically by the machine. It requires the encoding of lived temporality into symbolic form capable of being… [tea break]…reanimated by another embodied consciousness at a later historical moment. Once one understands the complexity of this operation, one also understands how fragile literary civilisation truly is.

We therefore face the emergence of a new scribal class. Those writers who possess genuine original style and who understand how to use AI systems critically and instrumentally will dominate cultural production. Genre fiction presents little difficulty for machine-assisted generation because genre already forecloses semantic openness to a significant degree. The highest literary style, however, remains difficult because it preserves irresolution, contradiction and embodied temporality: I know whereof I speak––I wrote my Will in September 2024 because I was about to undergo a medical procedure that had less than Russian roulette odds of survival. It won’t surprise you, dear, gentle, machine-assisted reader, to learn that it was short and to the point. Distinctly pithy: and I am not known for my… pith.

When readers struggle with writers such as James Joyce, Virginia Woolf or Eimear McBride, the difficulty arises precisely because those writers compact immense cultural and philosophical inheritances into linguistic form. Joyce’s consciousness was formed through Thomism, Jesuit pedagogy, scholastic metaphysics and the accumulated sediment of centuries of European intellectual history: it took thousands of people conceiving millions of angels on the heads of billions of pins to create the lived and still living tapestry of Ulysses. Recently, a feted Booker-Prize winner said ‘life was too short’ for her to be bothered to read it. If contemporary readers imagine such inheritances reproduce themselves automatically merely because “the novel” continues to exist as a commercial category, they are deluding themselves.

The scale of the crisis is therefore monumental. The Granta affair is not simply an embarrassing editorial oversight. It is the first genuinely literary manifestation of a much larger civilisational fracture: institutions charged with preserving style can no longer reliably recognise it.

So here is the interim report. Turn the phone off. Not airplane mode—you are not on a plane. Put it in a drawer. Go for a walk. Read a book physically. Speak face to face: sense the light rain of the Other’s spittle before they transmogrify into a drooling zombie. Recover contact with the body that gives rise to personality and therefore to style: don’t buy another puffer. Then attempt something culturally serious rather than merely consumable in the mode of synthetic fluff. Genuine artistic appreciation—uncommodified, unrecycled, non-performative—has itself become a revolutionary act.

The novel is dead for real this time, if we think of the novel as for entertainment, or for making money, or for turning into something else then eventually merch’––and if that novel is dead, the life support system of the true novel, the generative human living novel is, like the literary culture that has sustained it, running out of power fast… on an alien planet that’s been inadequately terra-formed, and which supports nothing much else besides… server farms.