Josh Abbey writes: Essays about close reading often begin with a cliche: IA Richards and his experiment. Soon the formula will include a disquisition on AI: can a large-language model close-read? I asked ChatGPT and it said ‘sort of’. So I asked CoPilot, and it said ‘Absolutely!’ It was a little shy and reticent to do any at that moment but when the AI criticism proliferates, we will know whose data centre needs to be smashed first.
Sadly, it seems impossible to issue an absolute metaphysical prohibition on an AI close reading or an AI producing a half- decent close reading. On Twitter, ‘the most lugubrious predictions are sent forth’ while forgetting that most critical factor, ‘the feuds and jealousies of literary [people] among one another’. AI recipes, AI horoscopes, AI gossip columns (they’re made up anyway), AI promotional reviews for sofas, these all seem likely. Less so anything literary, and literary outlets that do publish AI articles will want to have ironclad legends for their robo-writers. Some magazine will probably sneak a 400-word review of a fair to middling novel into its back pages and none but the most fabulously bored will google the byline and discover there is no one who goes by that name but ChatGPT.
But The Paris Review? Less so. Writers, and most especially, critics are jealous bastards; they like to know who the lucky one was. The word will get out. Why would anyone buy a literary magazine written by an AI? The reader of the literary review is conscientious, as unlikely as any to stand for it; they have as little patience for low-grade AI crap as they do for low-grade human crap. (The only reason they continue to buy low-grade human crap, if they do, is envy, morbid curiosity.) The consumer has an absolute power to make unviable AI-generated art and media: don’t read it, don’t watch it, don’t listen to it, don’t buy it. How will the Warner Brothers survive without revenue? That most AI generated art is indigestibly bad means this dynamic is already at play.
If criticism written by AI finds its way into the reviews of record, it will probably the writer who snuck it in there. A prospect that is less the omen of AI-induced mass redundancy, and more the exacerbation of a trend begun by crib sheets like No Fear Shakespeare, formulaic pedagogies (Topic, Explanation, Example, Link), rewriting a friend’s essay, searching on the internet for a near match or paying someone to write it for you. The causes have been biblical indolence and the belief that there is nothing to be gained from learning to close-read. ChatGPT has made it even easier for the student who wants something to do their thinking for them. The student need no longer rely on someone to do the thinking first. And the ease of ChatGPT entices those who found rewriting a friend’s essay too bothersome, paying someone too expensive, and their question unanswered by a previous sufferer of education. Of course, the idea of a teacher who wants to make education fun is that of a frumpy teetotaller, who is always in bed by nine and says ‘shoot’ instead of ‘fuck’. Unfortunately, there is no encouragement to laziness like a task deprived of fun.
Unlike learning to play the guitar, which begins with technique and exercises and ends in a heaving basement, verbal analysis or literary interpretation is seen in the classroom to have no creative end. The guitar teacher can point to the riffs, the jamming, the band, the friendship, the expression, an as yet unspoken kind of attention; what can the English teacher point to when explaining the subjunctive? Part of this, of course, is the insignificance of literary criticism in modern society, though it seems unlikely that teachers did ever say ‘you too could be like William Empson’ or Matthew Arnold or Samuel Johnson and expect much change in their student’s enthusiasm for relative clauses. Even ‘you too could be like Susan Sontag’ would seem to have little to do with the international phonetic alphabet.
Writing is a craft in which there is seen to be no immediate connection between the basic elements and creativity. Indeed, using words and telling stories is such an everyday thing, so inseparable from simply being alive, that writing is often not seen as a craft like music that develops from the most basic exercises. Some writers, mistaking Twitter for the id-receiving analyst, say that writers needn’t read novels to become a better writer. At best, in this dismal approach, close reading is thought to improve one’s creative practice. Criticism is known to admit of creativity – Geoff Dyer’s Out of Sheer Rage, Derrida’s Glas, Susan Howe’s My Emily Dickinson – but the idea that close reading can be creative in itself, ingenious or mischievous, is not entertained. And so one approaches the analysis of a passage from Romeo and Juliet with all the liberty and excitement of one filling out a licence application; we read only to wonder who should Elizabeth Bennet marry. The consequence has been to reduce criticism, in the minds of the SparkNotes-skimming, Rotten-Tomatoes-scanning public, to the simplest reflex: gushing superlatives, solipsistic complaints, or irrelevant ponderings (Is Harry right for Sally?); to alienate ourselves of our immediate impressions; and to curb the impulse to be ingenious and ingeniously human.
‘Close reading’ has long been a vaguely understood term, and part of this is derived from the fact that neither the New Critics nor Empson, Leavis pr Richards ever used the phrase as we do. We use at least three senses of ‘close reading’: there is a close reading, what manifests on the page; close reading as a method, a set of techniques; and close reading as an experience. There is a fixed and fast relation between the first sense, the artifacts of close reading, and the second methodical sense. The methods and techniques of close reading are often prescribed by or derived from the example of these most vaunted practitioners. Recently, the critic John Guillory argued that close reading is a cultural technique and offered a definition that covers both senses: ‘showing the work of reading’. (We are left to wonder what this reading should be and how exactly we are meant to show to it. ‘All cultural techniques resist definition of the sort that specifies the sequence and components of methodical action … Techniques are transmitted rather by demonstration and imitation.’ We must learn, it seems, from example.) Sometimes this showing is simply an act of noticing: when, for example, in Wordsworth’s line
Even as a shepherd on a promontory,
Christopher Ricks notices that ‘the line itself functions as a promontory, with the self-referring word concluding it and with the punctuation circumscribing it’. And other times a pleasing little turn: ‘how charmingly the word environ may environ the word iron’. It is a work of noticing and connecting, the scope of which is not always done justice by our time in the classroom. Of Shakespeare’s Sonnet 7, Helen Vendler said ‘the central image of the sun’s car generates anagrammatically scrambled cars elsewhere: in gracious, sacred, and … tract’. Or in the opening line of Isaac Rosenberg’s ‘Break of Day in the Trenches’
The Darkness crumbles away.
Tom Paulin sees ‘that “crumble” has inside it the word “rumbles”, and when Rosenberg uses “crumbles” in the first line of “Break of Day in the Trenches” he wants us to hear the rumble of an imminent barrage. It is an attention to what the words are doing, and what this doing does to the ‘meaning’.’
For obvious reasons, writing about close reading has a tendency to focus on the most exemplary instances. Take in hand a forthcoming essay collection on the practice, Close Reading for the Twenty-First Century. There are no dunces under consideration. The contents page is an honour roll: Auerbach, Vendler, Empson. Because brilliance is supposed to come easy, this focus on the best has obscured the experience of close reading, an experience that is often arduous. Similarly obscuring is the interface of close reading: words in a particular order. (Though interpretative dance is yet to succeed the sentence as the interface its modes would best represent the experience of close reading: talking to yourself, pacing numbly through the library, hunching over a book head in hands, pulling out hair, scalp scarifying, forehead kneading.) Because of this anything that can put words in a particular order can fluke a sentence worthy of Vendler or Ricks and pretend to an act of close reading. But the sentence of close reading is not a proposition, even if it yields a proposition – it is a transformation of experience. To show the work of reading we must put that reading into words and until that reading has been set in words it remains an ongoing experience. Close reading is not made in an instant.
One does not simply take up a text and close-read. They read the poem on the internet, decide that it is no good and trudge to Waterstones for a Penguin Classics; they fish a tattered copy from their bag, and get through two stanzas before the train arrives at their stop; they leave off reading because the lead of their pencil snapped or write a small essay in the margin that is only partly decipherable when they come to write up their reading. An AI has no such luck. An AI doesn’t refuse to read an author for twenty-five years only to discover they’re pretty darn good; it doesn’t grow impatient as they feel the ‘tell-tale compression of the pages’; pause with a sense of achievement on arrival at the first inset of glossy plates; get hungry or fall asleep mid-sentence; take a three-month leave and come back to resonances previously inaccessible as before you hadn’t read all of Samuel Johnson; an AI doesn’t read it and have no ideas, read it again and still have no ideas, read it a third time and desperately contrive something outlandish only to become so sure of the reading that it seems too banal to be worth remarking; an AI doesn’t feel sick as it tries to work out what words mean; an AI cannot re-read, nor return to a book after a long interval of life has intervened; it can only recalculate the appropriate word order based off an update to its dataset. An AI cannot read; it cannot close-read. It cannot bring life to bear on a text.
What we notice, what we connect this to, and how we put this noticing into words is determined by ‘numberless feelings, phrases, images’, ideas and impressions obtained over a long run of time. To hear in a poem, we must have heard; what we hear derives from what we have heard and have not. Literature is not something separate to life; it is part of life; it is made up of life’s bit and pieces: things heard and seen, felt, thought, dreamed, imagined and encountered. ‘Close reading’ has long been thought to have done away with impressionism, rightly so, but we cannot do away with the responses of life. It is because we have heard words and reacted to them that the lines ‘I had a mother, but she died, and left me’ in Charles Lamb’s weepy refrain ‘The Old Familiar Faces’ strike us as odd. (She died and left you? Doesn’t all dying involve leaving?) We spend our days responding to language and stories: at funerals and weddings, waiting at the bus stop and waiting for a takeaway, with our friends at the pub and a friend seeking advice about their relationship, dealing with a cantankerous customer and a nice one who has lost their receipt. To be offered a fragment of life, a poem, novel, or essay, and seek the counsel of some probability machine is not simply lazy, it is tragic. It is to deny that our own experiences can be of any value in understanding those of another, in appreciating their art.
4/12/2025
