I wrote a novel using AI. Writers must accept artificial intelligence – but we are as valuable as ever | Stephen Marche

I recently heard an exchange at a playground that should worry the executives at AI companies more than any analyst’s prediction of a bubble. A boy and a girl, maybe 10 years old, were fighting. “That’s AI! That’s AI!” the girl was shouting. What she meant was that the boy was indulging a new and particular breed of nonsense: language that sounds meaningful but has no connection to reality. The children have figured the new world out quickly, as they do.

Artificial intelligence is here to stay, neither as an apocalypse nor as the solution to all life’s problems, but as a disruptive tool. The recent scandal over Shy Girl, the novel by Mia Ballard, was doubly revealing. Hachette cancelled its publication amid claims it was reliant on AI generation (Ballard has said that an acquaintance who edited the self-published version used AI, not her). But the book was originally self-published. Apparently readers and editors didn’t mind until the use of AI was pointed out to them.

The fact that machines can generate meaning in the first place is an existential curiosity. But for writers, and for young writers in particular, AI has a more practical significance. A recent survey found that 86% of college students use AI regularly, which means that 14% are lying to survey-takers. The ordinary business of quotidian language – writing student essays, emails, memos, all the granular sentence-by-sentence work that once trained writers in their craft – is dissolving. Mastery of style, the laborious gift of the skilled writer, is being automated.

How are writers to live with meaning-generators? How should writers use AI? My perspective is slightly different from others mainly because I began using AI before ChatGPT. My first algorithmically generated story appeared in Wired in 2017. I published the first AI-generated novel reviewed in the New York Times, Death of an Author, in 2023. Currently, a generative text box I designed, “An Infinite Prayer for Peace”, is showing at the Bildmuseet Gallery in Sweden. It uses AI to articulate a different prayer every minute. It is a new kind of linguistic act, possible only through transformer-based artificial intelligence.

There seem to be two options facing writers. The first is not to use AI at all, or to pretend not to use it. The other is to automate their writing practice. The first is retrograde and fearful. The second forgets that art is a human practice, made by people for people. As becomes obvious when you actually try to use AI to make art, this is a false binary. Already a few paths through the slop are emerging.

Do not underestimate your value

The inventors of the Transformer, the T in ChatGPT, and the architecture by which all generative AI works, believed, against the grain of research at the time, that language was the key to abstraction. Language, rather than images or mathematics. They were more right than they ever could have imagined. At the core of the new magic is language. Language now is power. The revenge of the humanities is now fully on. The new cliche among tech lords is the need for taste in the artificially intelligent future. How do you think you develop taste? By reading. By writing. By being trained in reading and writing.

Researchers in Italy discovered they could use poetry to jailbreak the large language models into giving them instruction in how to build a nuclear bomb. This is more than a metaphor. Revel in it.

All killer, no filler

What LLMs do well, particularly ones shaped by human learning feedback, is generate convincing expressions of dead language. The more formulaic the task, the better they are. Software coding is their primary ability. But literary students, asked to write formulaic essays, asked to produce answers in a code, naturally use AI to compose them.

Generative models are fundamentally cliche machines. If you ask AI to write a film script, it will produce an average film script masterfully. If you ask it to write an essay, it will produce an average essay masterfully.

Once upon a time, mastery of the banal was adequate for writers. It was enough to prove that you were capable of writing. But that skill has no purpose any more – it can be automated. Skill will be found in the purpose of the work. What can you alone make happen with language?

Be the pusher, not the button

Chess is a good model for anyone trying to figure out how to use artificial intelligence. AI transformed chess well before it changed any other field. It has completely altered the nature of the game, the nature of training and analysis, the entire conditions under which the game is played. Every grandmaster alive today has trained with AI.

But Gukesh Dommaraju, the current world champion, took an unusual path: his coach, the grandmaster Vishnu Prasanna, encouraged him to avoid AI until he was fully formed as a player. He sharpened his skills, he grew his creativity, he steered his talent away from bad habits, and then, and only then, did he turn to the engines. It worked.

The key is to control the machine, rather than having the machine controlling you.

Dance in the dark

Why should fear be the only reaction to artificial intelligence? This technology, it is becoming clear on the ground, is not good at replacing people. I wrote Death of an Author through prompts, but the process was no easier than simply writing. Controlled language requires control over language, whether you’re using AI or a fountain pen. It took me years to create a truly regenerative work, what no human could do, a poem that reproduces itself every minute.

The surrealists, on encountering the linguistic destabilization of various theories of the unconscious in the early 20th century, no longer asked if they were masters of language. They did not compare themselves with Shakespeare or Keats. They invented games. They splurged on expression. They decided to use art to alter the nature of human experience. They made art for life.

There is real joy here. There is green grass.

For artists and writers of the past 50 years, constant adaptation to new modes of generation and dissemination has been the cost of freedom. It’s exhausting. On the other hand, the struggle faced at this moment is much the same as it has always been. Cliche is the historical norm. The world has always been full of letters that start “Dear Occupant” or “You are one of our most valued customers”. Machines may as well have been doing most of the writing. If you feel like being a writer is like swimming upstream, you’re right. Everybody has been swimming upstream forever.

As TS Eliot wrote nearly a century ago: “There is only the fight to recover what has been lost and found, and lost and found, again and again, and now, under conditions that seem unpropitious.” Language naturally deadens. Either grow new language on the rot, or shock the old language back to life. The task of writing in the age of AI is the same as it has always been: to see through the various manias and cults, to unpack the mechanisms that make the world turn, to ring the bells that still can ring.

Common sense might also help. “Render unto man the things which are man’s and unto the computer the things which are the computer’s,” Norbert Wiener, the father of cybernetics, wrote. “This would seem the intelligent policy to adopt when we employ men and computers together in common undertakings. It is a policy as far removed from that of the gadget worshiper as it is from the man who sees only blasphemy and degradation of man in the use of any mechanical adjuvants whatever to thoughts.”

It is worth remembering that most art has always been slop. When I did my PhD on early modern drama, my supervisor made me read every tragedy written in English until 1640. The process was grueling but instructive. Almost all of them, the fruits of the definitive art form of the greatest period in English literature, were garbage. Go and watch what was on TV in the 80s, if you want a more contemporary example.

There is good news for humanists in the arrival of AI. According to the New York Fed, unemployment for computer science graduates sits at 6.1%, while for art history majors, the number is 3%. Merely practical education will be of limited value in the AI future. To make yourself merely technically useful to a company or an industry is to make yourself vulnerable to replacement by automation at any moment. Thinking, creating, understanding – these cannot be replaced, certainly not by artificial intelligence. Trust me. I’ve tried.

The kids I overheard on the playground knew the difference between language that sounds meaningful and language that is meaningful. Do you? Does the literary community? Two roads diverge into a sloppy wood: one goes through what machines can do, the other goes through what only people can do. To write now is to wage war against cliche as usual, just this time with the AI and against it.

• Stephen Marche lives in Toronto and is the author of The Next Civil War and On Writing and Failure