The Right Words in the Right Order

Cultural elites and educators of the Global North are now in an advanced state of panic surrounding reading, literacy, and the humanities generally. University professors find to their dismay that students either cannot or will not read at the length that was once expected. Enrollment in humanities degrees in the US has been falling ever since peaking around 2010. The recently published OECD PISA scores (from an international test for 15-year-olds) included troubling evidence of a collapse in reading ability across the OECD, with scores peaking around 2012, and then falling off a cliff since—a drop equivalent to 1.5 years of learning.

Reading for pleasure is also cratering, especially among boys and men, with major consequences for the nonfiction book market. Commentators are now writing darkly about a new era of “post-literacy” or a return to one of “orality,” in which video replaces text as the central medium of knowledge-sharing, individual expression, and public argument.

Two technologies loom large in this cultural panic. The first is the smartphone, whose mass adoption in the early 2010s is often blamed, by Jonathan Haidt among others, for destroying attention spans and threatening young people’s mental health – a charge that intensified in the early 2020s as algorithmically curated short-form video became ubiquitous, following the rise of TikTok. Second, Large Language Models (LLMs) became an everyday technology following the launch of ChatGPT in November 2022. The impact on the humanities is twofold: Not only can text be generated almost automatically, but so too can its comprehension.

Every wave of industrialization provokes a countermovement: English Romanticism flourished at the dawn of the first industrial revolution, and the Arts and Crafts movement emerged with the second. It is unsurprising, and largely welcome, that literacy is now the subject of its own defensive movements which seek to celebrate the craft of reading and writing. Some of this has been forced on universities at speed: The rapid normalization of AI-generated coursework essays has driven the return to in-person, handwritten assessment methods. But some has been more pedagogically motivated, such as the recent announcement by the University of Chicago Law School that all digital devices would be banned from classrooms for the new academic year in order to avoid “stunting intellectual growth.” Columbia’s Literature Humanities and Contemporary Civilization sections followed suit.

Amid this countermovement, a new agonistic ethic of literacy has emerged, which not only celebrates the craft of reading and writing, but emphasizes its difficulty. Some literature professors have come out fighting, arguing that reading should be a struggle that students must confront. In The Atlantic earlier this year, Walt Hunter made the case for refusing to “meet students where they are.” In a New York Times op-ed, Matt Dinan suggested that professors perform their own struggle with great texts before their students. Every time an AI engineer or executive brags about the imminent bypassing of reading or writing, social media becomes awash with humanists arguing that the purpose of text lies precisely in the labor it requires. On this view, inefficiency is what gives literature its worth, which at its best is a Herculean translation of human experience from subjectivity into text, and then from text into another subjectivity.

LLMs have upended assumptions about reading and writing faster than any other technology, outstripping even the rise of short-form video. So intense is the disruption that there is a risk that the wider crisis of the humanities is seen entirely through a panicked post-2022 lens. A polarization sets in: on one side is the frictionless machinery, which makes reading and writing instantaneous, and on the other a romanticism about how the humanities were before the arrival of LLMs. We are in danger of misrepresenting the pre-2022 world. The question deserves to be asked: when exactly was the age of literacy that educators and humanists are seeking to defend or revive?

The OECD’s Programme for International Student Assessment, whose most recent scores have added to the literacy panic, was launched in 1997, following multiple attempts to produce a common measurement of literacy. Earlier efforts had been trialed by bodies such as UNESCO and the US Educational Testing Service, dating back to the 1950s. What gave the PISA initiative momentum was a wider ideological shift, derived from Third Way neoliberalism, that placed learning and literacy at the heart of social and economic progress.

As the education scholar Gert Biesta has argued, over the 1980s and ’90s the very idea of “education” was displaced in favor of an ideal of “learning.” Education implies a teacher, a subject, and a purpose; learning is something an individual does. What Biesta termed “learnification” meant that the transmission and application of knowledge became a priority across society and the economy in order to produce more active, independent, and prosperous citizens. Not only was “learning” (unlike “education”) something that could occur anytime and anywhere, but evidence of it would be found in practical applications, that is, the ability of individuals to achieve their goals. Pedagogy became anchored in cognitive science, so as to instrumentalize it in the service of economic ends.

One aspect of the turn towards learning was a fixation upon literacy as a measurable cognitive capacity. The literate individual in the “learning society” was one capable of using information to execute tasks successfully. The more complex the information and the task, the more literate the individual needed to be to complete it. On this basis, PISA compared literacy levels across dozens of nations. In 2000, the OECD offered a definition of “reading literacy” that few humanities professors would recognize: “Understanding, using and reflecting on written texts in order to achieve one’s goals, to develop one’s knowledge and potential, and to participate in society.” This is one of the measurable capabilities that are now reported to be in decline. Academics in literary studies would resist such a utilitarian vision of reading, and make the case for interpretation and textual analysis as worthwhile in themselves.

Other forms of “literacy” soon followed: media literacy, financial literacy, political literacy. As literacy rose in economic and political significance, it lost its association with literature. While progressive neoliberalism may have delivered better outcomes for disadvantaged children, it’s questionable whether the “learning society” represented progress for the humanities themselves once economic growth dried up after the financial crisis. In England, English was the most popular subject for A-levels (final high-school exams) in 2010, but has since dropped out of sight, falling from 83,000 students in 2013 to 54,000 a decade later.

Parents, teachers, and professors will have some idea as to why this is, and the reason is more than simply a matter of harsher economic imperatives. Over the past forty years, the shift from a world of “education” to one of “learning” and “literacy” has narrowed what it means to understand a text, whether that be Coleridge or a train timetable, and blurred the distinction between those two. Once everything is about successful task execution, questions of meaning and interpretation are approached via disaggregation, breaking texts down into ever smaller bits, where they can be processed with varying levels of executive skill. Text becomes a form of code to be “broken” and then converted into different code. This is a considerable distance from the agonistic ideal of literacy that we see in our post-2022 world, and yet it has been on the march since the 1980s.

A few months before the launch of ChatGPT, I wrote an essay for the London Review of Books reflecting on my experience homeschooling my then-nine-year-old daughter during lockdown while, at my university, dealing with the rising tide of plagiarized student essays (a problem that, in its conventional form of copying, has now evaporated). My daughter’s literacy exercises required her to read and rewrite the same short chunks of text, day after day, with the expectation that eventually this would be done correctly, but providing no facility or confidence to approach writing as a space of creativity or meaning. It struck me that the way “literacy” was being taught to children was connected to the difficulty students, taught in the same way, had in sustaining a commitment to their own original authorship, both of which were greatly exacerbated by the enforced pivot to online learning.

In both cases, language was being treated as a mere technology for completing a task or satisfying a tester, not as a route to self-expression or understanding. Once literacy is reduced to the correct ordering of words to produce sentences, it becomes unclear why one would care whether the order is one’s own. It is also unclear why one would choose to study human languages and literature, if they are taught and learned no differently from math or computer science, as a matter of right and wrong answers.

We have been inhabiting a world of post-human literacy for far longer than the present paeans to reading and literature recognize. The current crisis of the humanities has a history that is obscured by the panic surrounding LLMs, and the slightly longer-standing one about smartphones. Neoliberal “learnification” and literacy testing did considerable violence to the humanities by subjecting them to the rationality of measurable task execution. If young people in 2026 feel intimidated by language, and assume that a machine will use it “better” than they can, we need to ask how and why language became something to be evaluated statistically in the first place. The origins of the crisis lie even further in the past.

In the 1940s, the problem of communication was rethought in two parallel traditions. Both, in different ways, asked how one agent could transmit a message to another in a world of noise and chaos. Cybernetics, as pioneered by Norbert Wiener, provided a means of studying and modeling control systems, whether in humans, nonhuman organisms, machines, or combinations of the three. In this first tradition, language (such as that of a military command or a persuasive advertisement) came to be treated as a tool for influencing behavior. Meanwhile, information theory, as developed by Claude Shannon, offered a mathematical theory of communication. The information contained in a message could be calculated quantitatively, in terms of the improbability of a certain set of signs being placed in a given order, regardless of the meaning.

In the years that followed, these two traditions would provide the foundations for both the development of digital computing and the cognitive turn in psychology. These developments were closely related, as the French cognitive scientist Jean-Pierre Dupuy has detailed. Cognitivism modeled the human mind and brain as information-processing systems, along the lines that Shannon had laid out, in a deliberate analogy to digital computers. Cognitive processes of learning and communication could be studied in humans in the same terms as in machines, once psychology abandoned humanist problems of understanding, interpretation, and meaning. Evidence of learning was to be found in effective communication, that is, in putting words and other symbols together in the right order.

Communication of this nature requires no human subject capable of mining their subjectivity, interpreting meaning, or empathizing with an author or speaker. Instead, it is achieved through pattern recognition and mathematical calculation, precisely the techniques that machines can execute on a larger scale and at greater speed than human brains. Shannon’s probabilistic approach to information means that messages sit on a spectrum, between a probability of one (in which nothing new is communicated) and zero (in which there is just pure noise). It is on this premise that LLMs are able to complete sentences for us, predicting which words will likely appear next. The cognitive revolution—which would come to reshape pedagogy in the late twentieth century—used these post-human methods and theories to radically affect how human cognition and communication were conceived.

The temptation for humanists and cultural theorists is to recoil from this mechanization of mind, understandably given what has happened to the study of literature, history, and languages over the past two decades. But recent intellectual histories have troubled such comfortable romanticism. Bernard Geoghegan’s fascinating Code: From Information Theory to French Theory tells an overlooked story of how in the mid-twentieth century the project of reducing communication to probabilities overlapped with the structuralist ambition to ground cultural analysis in a general science of signification.

Structuralism, as advanced in the 1950s and ’60s by Claude Lévi-Strauss and Roland Barthes, shares various family resemblances with information theory. Both sought to grasp communication as such, through studying the internal relationships in symbolic systems. Both opened doors to post-humanism, in which the question of subjectivity and interpretation could be simply set aside in favor of attention to networks of signs. Barthes’s famous 1967 essay “The Death of the Author” liberated a generation of humanities students to abandon the question of intended meaning and interpretive depth in favor of intertextual and intratextual analysis. This appeared after Michel Foucault’s The Order of Things, which had concluded by dreaming that “man” might soon be erased, “like a face drawn in sand at the edge of the sea.”

In different ways, information theory and structuralism both extend a project of decoding in all directions, reading society at large as a system of signs and texts, and dissolving the distinction between literature and society, or between computation and society. While structuralists adopted a veneer of Parisian radicalism, promising to disrupt power by decoding it, Geoghegan’s argument is that the endpoint was conservative, neutering political resistance by restricting it to a form of literary studies. The dual prongs of information theory and cultural theory that appeared in the 1950s to be advancing separately were increasingly conjoined from the late twentieth century onwards, first as “literacy” became something to be measured on the basis of task execution (a form of communication, as Shannon would see it), and then in the age of platform capitalism, as vast cultural archives became technically computable as information. The “digital humanities” represent the consummation of a flirtation that could be dated back to the postwar period. “Culture,” “society,” and “information” are now all available to the same disciplinary analysis: the one performed by algorithms.

This insistence that the humanities are no longer distinguishable from mathematics is the premise of Leif Weatherby’s audacious and daunting Language Machines: Cultural AI and the End of Remainder Humanism, a post-2022 theory if ever there was one. For Weatherby, LLMs bring us face to face with a brutal reality that we have been inhabiting since at least the 1940s, but which we prefer to duck by clinging to some cultural ideal of humanist exceptionalism (“remainder humanism”). LLMs demonstrate the core truth expressed by both information theory and structuralism: Language and communication exist independently of what we think of as human minds. Our difficulty in even speaking of LLMs in non-anthropomorphized terms signals how caught up we still are in a pre-structuralist worldview, in which meaning and signification are things that only humans are capable of. The LLM, much as Geoghegan anticipates, represents a fusion of probabilistic and linguistic technique that is no less mathematical for being cultural, and no less cultural for being mathematical.

Placed in this longer intellectual history, the present crisis appears to be a failure to grapple with much slower and deeper transformations in the status of text in society. The current panic surrounding screens and AI may be rooted in the tangible harms these pose to the intellectual development of children and students. But where this panic generates romantic ideals of the humanities as a form of “craft work,” it obscures what literacy, reading, and communication have actually meant for the vast majority of people. Whether they intend to or not, the most impassioned defenders of the humanities often sound like they want to return not only to a pre-2022, or even pre-2010, world, but to one before the mid-twentieth century, before the drift to post-humanism began.

What might an alternative stance on literacy look like? It is a worthwhile goal to separate culture from computation within the school or the university. Such an agenda would need to also take aim at the Edtech industry, which has elevated “learnification” to new heights, profiting from the development of feedback systems that close down the space for experimentation, error, and creative nonsense. A depressing aspect of the plagiarism cases I discussed in my 2022 essay (but also students more generally) was that many students seemed to view writing as a series of traps, which they were certain to fall into sooner or later. In order to eliminate risk, they wanted to know in advance how an essay should be structured, what should be in it, and how it should be expressed. The threat posed by LLMs is not so much that they remove the difficulty of reading and writing (as the agonistic humanists would have it), as that they remove the playfulness, the uncertainty, and the intriguing mistakes.

What needs reviving is not some classical ideal of “Man,” but, as Weatherby argues, poetry. Textual generation, he writes, “puts poetry on one end of a spectrum bookended on the other by ideology.” Where we feel compelled to use the correct words in the right order, for fear of consequences, we are victims of ideology. LLMs allow us to witness our own ideological constraints, showing us how our culture has been streamlined, disciplined, and constrained into certain forms of expression. Poetry, by contrast, is the exploitation of the improbable, based on the assumed freedom to put words in an order that wasn’t expected and hasn’t been done before. The point is not that only humans can write an improbable sentence; machines can too. It is that a student should be free to try one, and to get it wrong.

Literacy, in its current guise, is deeply ideological. Reading has become a cognitive processing habit, necessary to support the kinds of “symbolic analysts” (elite knowledge workers, able to work back and forth between numbers and narratives) once deemed core to post-industrial capitalism. Given that many of those same jobs now appear most at risk from the AI revolution, there is an opportunity to liberate symbolic analysis from the demands of the knowledge economy, and turn English classes into rare places of freedom.

William Davies is a writer and researcher based in London. He is an Open Society Fellow for 2026-27. His writing is available at Williamdavies.blog.