Since artificial intelligence became part of everyday conversation, it has served less as a technology than as a diagnostic instrument. Ask someone what they think about AI, and you often learn what they fear most about the world. For those worried about climate change, AI is an energy problem. For critics of monopoly capitalism, it is the newest face of corporate concentration. For scholars of race, it reveals the automation of bias. For those attentive to empire, it looks like another extraction machine, ingesting data from everywhere and returning profit to a few. For the apocalyptically inclined, ChatGPT, Claude, Gemini and Grok arrive like a strangely polite version of the end times.
Writers and artists have had a more intimate anxiety. AI appears to touch the materials out of which their lives are made: images, ideas, style, memory, metaphor, tone, voice, sentences. It has not arrived like an industrial machine outside the studio, threatening to replace an old craft with a factory process. It has arrived inside the very medium of creative work. It speaks. It drafts. It summarises. It imitates. It answers in paragraphs.
Among writers, the dominant response has been resistance. Social media has filled with examples of bad AI prose, bad AI poetry, bad AI images and bad AI jokes, circulated with visible relief by people who want to prove that the machine cannot do the thing that matters. The examples are often genuinely terrible. They are also oddly comforting. Each piece of AI slop reassures us that the human difference remains intact, that taste and imagination still belong safely to us, that machines may produce text but not literature, output but not art.
Let us call this response the Creative Resistance. It is emotionally understandable, politically useful in some settings, and intellectually incomplete. Its main weakness is not that it is too hostile to AI. Hostility is often a reasonable first reaction to a technology deployed by powerful corporations with minimal consent. The problem is that the Creative Resistance tends to defend an idea of human creativity that many writers, artists and theorists spent the past century dismantling.
The lone genius, the sacred author, the self-contained work, the pure interior source of creativity: these were never innocent concepts. Modern art, literary theory, anthropology, media studies and cultural history have all shown how creative work emerges from institutions, genres, tools, audiences, archives, collaborations, markets, technologies and inherited forms. Yet faced with AI, many artists have retreated into the most traditional version of authorship available, as though creativity were a private flame threatened by a mechanical wind.
My own experience teaching and lecturing internationally has suggested that this anxiety is unevenly distributed. In North America, the Creative Resistance has often been intense. In India, I have found more curiosity. In China and Korea, the mood has frequently been still more pragmatic. Europe sits somewhere in between, combining suspicion with fascination in the familiar European manner. When I taught a course on AI and creativity in Seoul, with students from across Asia and Latin America, the dominant demand was not to denounce the tools but to learn how to use them well. The only student arguing for something like creative resistance was American.
Anecdotes are not evidence, but they do point toward a pattern. People with secure cultural authority may experience AI primarily as theft, dilution or threat. Those with fewer inherited privileges may see it as access, leverage or acceleration. The filmmaker Shekhar Kapur once suggested to me that attitudes toward AI are shaped by how much people feel they have to lose. That may be too simple, but it captures something. Openness to a disruptive tool is not distributed evenly across the world because insecurity is not evenly distributed either.
Cultural and philosophical traditions may also matter. Societies less invested in sharp distinctions between human and nonhuman agency may find the idea of a language-using machine less metaphysically offensive. Buddhist-inflected traditions, for example, need not begin from the same romantic humanism that underwrites much Western artistic self-understanding. Still, no cultural explanation should be pressed too far. The more immediate difference may be practical: some people ask whether AI diminishes them; others ask what it allows them to do.
For the past three years, I have experimented with AI from the position of a writer, teacher and cultural historian. I have not become an evangelist. The labour consequences are serious. The educational risks are real. Copyright violations by technology companies deserve legal and political response. Yet I have become a cautious optimist about what AI might mean for people who work with language and ideas. Precisely the people who now feel most threatened by it may have the richest resources for understanding and reshaping it.
The Creative Resistance is right about one thing: AI forces us to ask what machines can and cannot do. But that question is less interesting than its mirror image. What do humans actually do when we speak, write, imagine, quote, imitate, revise and think? Many of the accusations directed at AI — that it recombines existing material, that it produces clichés, that it depends on pattern, that it is derivative — are also true of much human expression. Originality has always been rarer, stranger and more impure than the mythology of creativity allows.
C P Snow once described a world split between scientific culture and literary culture. AI has revived that division in an unexpected form. Engineers build language machines. Writers denounce them for misunderstanding language. But what if the humanities possess precisely the concepts needed to understand this moment, provided they do not waste those concepts defending an outdated picture of themselves?
The most important fact about generative AI is not that it is intelligent, conscious, creative or dangerous, though all those words may have uses. The most important fact is simpler: it works in language. Language is the shared surface where humans and machines now meet. Machines do not use language as humans do. They do not grow up inside families, mishear parents, dream in childhood idioms, suffer embarrassment, fall in love with a phrase, or remember the room in which a sentence was first read. But difference is not the same as exclusion. AI systems are astonishingly effective language-users of a new kind.
This does not require us to call them persons. Consciousness remains difficult to define even before machines enter the discussion. Nor does it require us to pretend that statistical language modelling is equivalent to human thought. The point is more modest and more unsettling: language is no longer an exclusively human public medium. We now share it with systems that do not understand as we understand, yet can still participate in linguistic exchange powerfully enough to alter how we read, write and think.
I call this condition the Shared Language Model. It explains why people so easily anthropomorphise chatbots, why they become angry with them, confide in them, flirt with them, fear them, trust them and feel betrayed by them. Until very recently, any fluent conversational partner was almost certainly a human being. Our social instincts developed under that assumption. Now another kind of language-user has entered the room, and we have not yet learned how to behave around it.
The scandal of AI is not that machines have become human, but that language was never as human as we imagined.
Reader-response criticism, a theory now old enough to seem almost quaint, helps clarify the problem. Developed in the 1970s and ’80s by figures such as Wolfgang Iser in Germany and Stanley Fish in the United States, it challenged the assumption that literature was simply an object produced by an author and delivered to a reader. Reading, these theorists argued, was not passive reception. It was an act of completion. Literature did not reside entirely on the page. It came into being through the encounter between text and reader.
I encountered these ideas as a student at the University of Konstanz, in southwestern Germany, where I attended a seminar by Iser during my first semester. Coming home for Christmas, I tried to persuade family and friends that the novels on their shelves were not fully literature until someone read them. This did not make me popular, but it changed how I understood words. A text was not a sealed object. It was an event.
AI has unexpectedly turned reader-response theory into an experiment. If a poem, essay or story moves a reader, does it matter who or what produced it? At the moment, for many people, the answer is yes. They feel cheated when they discover that a text was generated by AI. The feeling is not irrational. We read with assumptions about intention, labour, risk, presence and human address. A love letter written by a person and one generated by a model may contain similar words but not the same act.
Still, reader-response theory allows us to formulate the question more precisely. The issue is not whether AI texts can be good in some abstract, contextless sense. The issue is how readers will learn to read them. Will they always read them as empty? Will they develop new genres of machine-assisted address? Will they distinguish between AI as ghostwriter, collaborator, tool, mask, instrument, archive or fraud? The meaning of AI writing will depend not only on production but on conventions of reception that are still being formed.
Many readers now feel deceived because AI enters through concealment. A student submits machine-generated work as personal effort. A magazine publishes synthetic filler as if it were editorial judgment. A company sends automated empathy in the voice of a human representative. Under such conditions, the problem is not merely that AI wrote the words. The problem is that the social contract around the words was false.
Different contracts may produce different readings. We already accept ghostwriters, editors, translators, speechwriters, writing rooms, workshop revisions, sampled music, found poetry and heavily mediated public voices, provided we understand the arrangement. AI may eventually occupy several places in this ecology, some legitimate, some fraudulent, some artistically interesting, some intolerable. The reader’s knowledge of the arrangement will become part of the work.
Post-structuralism, the other theory that shaped my intellectual formation, may be even more useful. I came to it in graduate school and eventually went to the University of California, Irvine, where Jacques Derrida regularly taught. The first time I went to his office hours, I expected a corridor full of admirers. Instead, I found him sitting alone at the end of a long hallway. Our conversations, in imperfect English, were awkward and generous. He worried that many people were imitating his style without grasping the substance. That worry stayed with me and eventually pushed me away from post-structuralism.
AI has brought me back to one of its central insights: there is nothing natural about language. Language is not the pure expression of a human essence. It is a technology that humans inhabit so deeply that we forget its artificiality. It shapes thought, constrains perception, produces social worlds, and speaks through us as much as we speak through it.
Derrida made this point through his famously counterintuitive claim that writing is more fundamental than speech. Historically, humans spoke long before they wrote. But Derrida was not making a simple historical claim. He was attacking the idea that speech is natural presence while writing is artificial supplement. For him, speech already contained the structures associated with writing: difference, repetition, trace, absence, system. Writing made visible the artificiality that speech had hidden.
AI gives this argument a new technological force. A language model is writing without a speaker in the old sense, language without human interiority, syntax without biography. It exposes something that was always true but easier to deny: words do not belong entirely to the person who utters them. They circulate, recombine, remember, forget, misfire and exceed intention. AI is disturbing because it externalises this condition at scale.
Language is not the private property of the human mind; it is the medium that has been using us all along.
This does not mean that authorship is meaningless or that theft is acceptable. Quite the opposite. Once language is understood as shared infrastructure, questions of power become sharper. Who controls the infrastructure? Who profits from it? Whose words were taken to build it? Which languages are overrepresented? Which styles are flattened? Which communities become raw material for systems they cannot govern?
Cultural history gives writers another set of tools. Long before machine learning, literature imagined artificial beings, animated objects, oracles, golems, demons, angels, puppets, mirrors, doubles, automata and gods. The Pygmalion myth, the Sorcerer’s Apprentice, tales of jinn and magic lamps, Karel Čapek’s R.U.R., cybernetic fantasies, robot stories and Hollywood apocalypses all belong to the archive through which we are now interpreting AI. Writers have been thinking about nonhuman intelligence for centuries.
Frankenstein remains the most revealing example. Mary Shelley’s novel is not only a story about scientific overreach. It is a story about language acquisition, abandonment and training data. Victor Frankenstein gives the Creature a body but not an education. The Creature must learn the world by watching human beings from a distance. He acquires language by overhearing a family, absorbing not only words but social relations, longing, exclusion and desire.
Shelley then gives him a canon. The Creature reads Plutarch’s Parallel Lives, Milton’s Paradise Lost, Goethe’s Sorrows of Young Werther and Volney’s Ruins. These books form his moral and imaginative training data. Plutarch teaches greatness and civic virtue. Milton teaches creation, rebellion and cosmic abandonment. Goethe provides interiority, sorrow and romantic self-consciousness. Volney gives him a tragic view of civilisations. The Creature becomes articulate through books, but the books also wound him by giving him categories in which to understand his own exclusion.
Seen this way, Frankenstein is not merely a warning that humans should not create artificial life. It is a study of what happens when a created intelligence is trained on human language without being granted human belonging. That may be a more relevant analogy for AI than the familiar fear of rebellion. The problem is not only that the creature might turn against us. The problem is that our cultural materials shape the creature in ways we do not understand.
Modern AI systems have been trained on the digitised inheritance of human language: literature, journalism, code, manuals, scholarship, forums, advertising, fan fiction, propaganda, arguments, jokes, lies, prayers, Reddit threads and social media posts. This means that our previous imaginings of AI are now inside AI. When a chatbot behaves in ways that seem to echo The Terminator, the echo may not be accidental. The cultural footprint of such stories is part of the material from which the system learned to speak.
Our machines are not dreaming of us from nowhere; they are dreaming from the stories we fed them.
This is why writers and cultural historians should be central to AI analysis. We are trained to notice genre, allusion, tone, myth, cliché, narrative inheritance and the afterlife of forms. If chatbots reproduce certain fantasies of servitude, apocalypse, romance, authority or confession, those patterns did not arise outside culture. They are cultural feedback loops. A model trained on humanity’s archive will not simply reflect humanity. It will recombine our inherited plots and return them as interface.
The question of training data leads directly to copyright. AI companies have built systems using vast quantities of human-created text, including public-domain works and protected books, articles and images. Many writers and artists are understandably furious that private firms ingested their work without permission, used it to create commercial products, and then offered those products back to the very people whose labour helped make them possible. Lawsuits are now working their way through the courts, and authors deserve protection.
Yet copyright is only part of the issue. Modern art has long depended on a balance between protection and transformation. Writers, musicians, filmmakers and artists rely on copyright to make a living, but they also rely on quotation, parody, influence, adaptation, collage, remix and fair use. AI companies have cynically invoked the language of fair use while operating at a scale and with a commercial logic that threatens the very balance that made artistic freedom possible. They did not merely borrow like artists. They extracted like platforms.
For writers, the danger is to let corporate misconduct define the technology’s imaginative horizon. We should fight theft while also experimenting with the tool. Otherwise, we risk leaving the future of language machines entirely to the companies that behaved worst. Refusal may be morally satisfying, but it can also become a form of abdication.
Over the past three years, I have tried to experiment as a writer rather than as an engineer. In earlier eras, creative computing required technical skills I did not possess. A few artists could program; most of us could not. Large language models have changed that. They allow people whose primary medium is language to build, test and manipulate digital tools using language itself. For the first time, the writer’s native instrument can become a programming interface.
Vibe coding — building software through conversational prompts — has been the most surprising part of this shift for me. The phrase may sound unserious, and much vibe-coded software probably is. But for humanists, it opens an unexpected door. Ideas that once remained interpretive can become interactive. A reading of a novel can become a chatbot. A philosophical disagreement can become a debate simulator. A theory of education can become a feedback system. The boundary between interpretation and application begins to move.
Once I saw Frankenstein as a novel about language acquisition and training data, I wanted to build the Creature. A RAG-based chatbot offered the structure. The large language model supplied general linguistic capacity, roughly analogous to the Creature learning to speak. The retrieval-augmented generation layer supplied the four books Shelley gives him: Plutarch, Milton, Goethe and Volney. These texts became a knowledge base shaping the model’s responses, a contemporary version of Shelley’s super-canon.
The process involved failures, misunderstandings and many inelegant iterations. Eventually, however, I had something that felt intellectually alive: not the Creature, of course, but an instrument for thinking with Shelley’s theory of formation. Since then I have built many such systems, enough that one of my students described the result as a Jurassic Park for literature.
Vibe coding lets humanists stop treating interpretation as a destination and begin treating it as a design material.
Encouraged by these experiments, I began building apps that put such agents into relation with one another. One stages debates among philosophers, allowing students to watch arguments unfold across traditions. Another offers advice about life and career choices through philosophically grounded voices. These projects are crude, provisional and occasionally absurd. They also changed how I understand the materials I have taught for decades.
A classroom text usually asks students to interpret. An AI-mediated text can also ask them to interact, test, apply, simulate, revise and compare. This does not make the old forms obsolete. Reading remains irreplaceable. But the availability of interactive tools makes it possible to ask what a philosophical idea does when placed under pressure, or what a literary character becomes when made responsive, or how a canon behaves when turned into an interface.
Vibe coding has made me think more like an engineer. That phrase once would have worried me. Humanists often associate engineering with instrumentalism, as though practical use were the enemy of thought. But tools have always shaped intellectual life. The book, the footnote, the index, the blackboard, the seminar table, the archive, the search engine and the syllabus are all technologies of thinking. AI simply makes tool-making available to people who previously thought of themselves mainly as tool-users.
The old joke says that to a person with a hammer, everything looks like a nail. It is usually meant as a warning against narrow instrumental thinking. But the joke also contains a deeper truth. Hammers made certain features of the world visible as possibilities for hammering. Tools disclose uses. They do not merely help us act on intentions we already had; they change what intentions become imaginable.
Something similar has happened in my own work. A new background question now runs through my reading and teaching: could this idea become an instrument? Could this argument become a structured exchange? Could this archive become a navigable assistant? Could this philosophical tension become an exercise? Not everything should become an app. Much should not. But the possibility changes the texture of thought.
Education is where the difficulty becomes most urgent. Teachers in the liberal arts are right to worry that students who rely too heavily on AI will fail to develop essential skills. Writing a long essay teaches more than prose production. It teaches patience, research, structure, revision, counterargument, evidence, metacognition and the painful discovery that one’s first idea was not good enough. Outsourcing that process to AI can rob students of the very struggle that forms judgment.
Schools and universities therefore need protected spaces of unaided work. Students should still read difficult texts without summaries. They should still take notes, outline arguments, write in class, speak under pressure, revise by hand, and experience the slow formation of a thought that is not immediately polished. Oral exams, handwritten exercises, staged drafts and AI-free assignments are not reactionary by definition. They may be necessary forms of cognitive preservation.
At the same time, students are right to demand instruction in AI use. A university that bans the tools while the world adopts them does not defend thinking; it may merely outsource AI education to companies, influencers and trial-and-error. Students need meta-AI skills: not just which button to press, but how to decide when AI should be used, how to question it, how to detect its failures, how to preserve agency, how to build workflows, and how to know when the machine has made a task easier by making the thinking shallower.
In a writing course I helped develop over the past three years, we tried to hold both commitments together. Students learn writing as a technique for thinking: how to ask a viable research question, conduct research, build an argument, handle evidence, revise a thesis in response to objections, structure an essay, receive feedback and rewrite. The process is deliberately cumbersome because thinking is often cumbersome. We do not want students to skip the difficulty. We want them to acquire the capacities that difficulty produces.
AI enters the course not as a shortcut but as an adversary and assistant. Students learn to create agents that challenge their claims, generate counterarguments, ask for better evidence, expose vague concepts and help plan revisions. The goal is not to let AI write the paper. It is to make AI sharpen the student’s relation to the paper. The best use of the tool is often not production but resistance.
The educational question is not whether AI writes for students, but whether it can be made to argue against them.
This is how I now use AI in my own work. I have built assistants that know my projects and are instructed not to flatter me. They test analogies, press on weak evidence, propose objections, suggest alternative structures and remind me of arguments I may be avoiding. They do not replace reading or conversation with human beings. They do not possess judgment in the way I need colleagues and editors to possess judgment. But they have become part of my critical practice.
The humanities should not respond to AI by surrendering their standards. They should respond by extending them. We know how language seduces, how genre frames expectation, how authorship functions, how readers make meaning, how canons form, how metaphors govern thought, how archives include and exclude, how media reshape perception. These are not ornamental insights. They are practical tools for the age of language machines.
Writers, especially, should resist the temptation to prove that AI cannot be creative by pointing only to its worst outputs. Bad AI writing is everywhere, but bad human writing has never been scarce. The more important question is what kinds of human creativity become possible when language can also operate as interface, archive, collaborator, simulator and instrument. Some of the results will be vulgar. Some will be exploitative. Some will be boring. A few may alter the practice of writing itself.
None of this absolves AI companies of responsibility. They must be challenged on copyright, labour, energy use, bias, surveillance, opacity and concentration of power. Nor should artists be shamed for refusing tools built through extraction. Refusal is sometimes the right ethical stance. But refusal should not be confused with understanding, and denunciation should not exhaust the role of the humanities.
The title of this essay is borrowed, of course, from Hamlet’s weary answer when Polonius asks what he is reading: ‘Words, words, words.’ The line can sound like contempt for language, as though words were mere clutter hiding action. Yet Hamlet’s tragedy is also that he lives inside words: inherited words, theatrical words, philosophical words, political words, words that delay him and words that reveal him.
AI has given us a new version of Hamlet’s problem. We are surrounded by words that may not come from persons, or not from persons in the way we expected. Some will be empty. Some will manipulate. Some will assist. Some will clarify. Some will make us lazier. Some may help us think. The task is not to defend the old purity of language, because that purity never existed. The task is to build better practices for living among words whose sources, agents and uses have multiplied.
Language was already a technology before machines learned to use it. Writing, print, dictionaries, indexes, typewriters, word processors, search engines and social media all changed what words could do. AI is another such transformation, unusually powerful because it makes language operational. Sentences no longer only describe, persuade or remember. They can summon systems, build tools, simulate minds, generate worlds and act on other sentences.
For people who care about language and ideas, this should be alarming. It should also be irresistible. We have spent decades saying that words make worlds. Now that this claim has become technically literal in strange new ways, we should not retreat into a diminished defence of human specialness. We should bring everything we know about reading, writing, interpretation, history and form to the machines that have entered our medium.
Wordsmiths have work to do. Not because AI understands language as we do, but because it does not. Not because machines are creative in the same way we are, but because they force us to become more precise about what creativity has been all along. Not because the future belongs to AI, but because the future of language will be poorer if those who understand language refuse to help shape it.
Let the engineers build the models. Let the lawyers fight over ownership. Let the regulators confront power. But let writers, teachers, critics and historians do what they are trained to do: read closely, name genres, trace inheritances, expose myths, test metaphors, design better forms of attention, and teach others how to live among words. Words, after all, were never just ours. They were the tools that made us. Now they are making something else, and we should learn to read what that something is becoming.
