“If ChatGPT can do it, it’s probably not worth doing by humans, and it very well may not be worth doing at all.”
From More than words: how to think about writing in the Age of AI
by John Warner
Writing, according to Oxford American Dictionary and Thesaurus is: 1) A group or sequence of letters or symbols; 2) Handwriting; 3) The art or profession of literary composition.
AI generated text concurs with the first definition, the second is irrelevant, and according to Mr. Warner, the author of More than words: how to think about writing in the Age of AI, it fails the third definition, because “large language models do not ‘write.’ They generate syntax.”
The book opens with a survey of leading views on AI’s intelligence, then discusses whether text-generating Large Language Models (LLMs) can eliminate the “rote work” of writing in the way that calculators did for math. The author’s verdict is an unequivocal “NO”. Only Humans Write.
Writing, according to Mr. Warner, is not the end product, but first and foremost a complex process that involves memories, thought, and insight “spontaneously” emerging from our subconsciousness. LLMs, on the other hand, are simulations trained on billions of texts written by humans to generate texts that mimic human output. As these texts are plausible and sometimes even engaging, they appear to people accustomed to anthropomorphize like “writing.” As AI’s response to prompts is almost instantaneous, it’s efficient and cost saving.
Efficiency and cost-saving may have little place in a book about writing, but unfortunately, they are the ugly elephant that crowds and blocks the room. LLMs were created using copyrighted text without owners’ consent, compensation, or even knowledge. OpenAI, Anthropic, and other tech companies that have been dedicating mind-boggling resources to create and run LLMs have vested interest to (mis)represent AI as the ultimate solution for every problem, and push it everywhere in total disregard of human and environmental cost. This elephant keeps ballooning as the companies race to create larger models and more data centers, and one can only hope that its uncontrolled expansion would be stopped before it will explode causing unimaginable suffering to people and to the planet. However, as the main thesis of the book is not decrying Sam Altman and his ilk, I’ll try to ignore the elephant. But first a few words about those who inadvertently nourish it: a chapter titled “Writing is feeling” relates how a university used ChatGPT to help draft a heartless statement of condolence following a mass shooting. The story shows that LLMs are not the only ones lacking in both the department of thinking and of feelings. Sometimes the problem is the users and not the tool.
The meat of the book is the usage of LLMs for writing, and since Mr. Warner wears two hats, that of an English instructor at a college and that of a writer, he addresses both the issue of students’ use of LLMs in lieu of doing actual writing and of the effects of the shenanigans of “AI slop” on professional writing.
Some people enjoy the process of writing. Many more relate with the quote: “I hate to write, but I love having written.” There is no argument that at school, students’ acquisition of writing skills is by far more important than the end product, and that their use of LLMs for writing (and other) assignments results in their learning nothing. The question is: What are teachers willing and able to do about it? There are two problems. The first, which existed long before LLMs, is that students are taught superfluous skills. The second is that, due to LLMs, students are not acquiring any writing skills.
The author admits: “I incentivized the production of bullshit by students for many years. I engaged in a highly prescriptive teaching process that had students jumping through hoops with me coaching them every step along the way in an effort to get them to write something for which I could give a good grade.” That sounds like better instruction than I remember from my school days or what I saw with my children, who were inundated with writing what we all agreed was “busywork.” No one argues that proficiency requires practice, but it appears that only after educators had to read copious amounts of LLMs’ generated “essays”, they started to question how much practice should be required from kids unlikely to write a single essay after their graduation. Mr. Warner’s own words: “If we’re going to confront what ChatGPT means for the kind of reading and writing students should be doing in school, we have to get down to the root values of what is important and meaningful about the writing we ask students to do.” Finally, the new problem caused educators to ask what is essential to teach about writing.
The chapters about “Content vs. Writing” and “On the Future of Writing for Money” brought the elephant back to my mind. Over the last three years, the shenanigans of “AI slop” have taken central stage in many conversations, so the book didn’t add anything new to me. It still was interesting to read the author’s perspective and his own story. Also, something seemed to be missing. As I’ve read the physical book, I cannot check whether the word “talent” appears in the text. My impression is that it’s no longer a prerequisite, nor a key to success, in professional writing.
The last part of the book, titled “A Framework for Action” was disappointing, for it felt detached from reality, and even condescending. Mr. Warner might be well justified to “have very little fear we’re going to be left behind,” but I find it disingenuous to argue that “we’re in charge of what we allow AI to do to our world and to our lives” when so many people were made redundant and lost employment prospects in professions they have studied for many years. He cites Tolstoy’s, What Is Art?, then asks, “…but without any original feeling driving the impulse to create, and then without any intention driving the creation, why should we find any meaning to it?”
Well, people seem to find as much meaning looking at the lines and colors of an original painting as on a good forgery as long as they don’t know that one is fake. A novel written with LLM recently won a respected literary prize, and several other novels became best-sellers. These anecdotes don’t necessarily mean that Tolstoy was wrong, but they do suggest that the count’s deep thoughts and writings about Art while illiterate peasants tended his estate, may not align with modern public taste.

Left: a page from William Blake’s illuminated book The Marriage of Heaven and Hell. Image from William Blake Archive. Right: Jean Miélot was an author, translator, manuscript illuminator, scribe, and priest. He lived in the fifteen century, and was mainly employed in the production of de luxe illuminated manuscripts for dukes of Burgundy library. Image from Wikipedia.
My greatest disagreement with the book is the hypothesis that if something can be automated then it’s not worth for humans to be doing. An example may show what I mean. For centuries, scribes copied texts that kept Greek and other ancient knowledge from oblivion. Their work, time consuming and labor intensive, become economically nonviable after the invention and spread of printing. William Blake’s illuminated poems, however, show that this practically obsolete endeavor, is, in special occasions, not only worth doing by humans, but in a hand of a genius is a unique work of Art. My conclusion is that there is no general principle of what is worth for people to be doing, and that the matter is more nuanced than what the book posits. Forgers of artwork are prosecuted when caught not because their forgeries render art meaningless, but because art’s monetary value is determined by demand and supply (e.g. a thousand Mona Lisas could smile as enigmatically as the one in Louvre, but their monetary value would be completely different). AI nowadays does more effectively what scientists and artists have been doing for years. Does this render the outcome valueless? According to Mr. Warner, “Generative AI content is, by definition, a great averaging of what’s in the world. An embrace of this output is a kind of capitulation to the machine, rather than staying true to our nature as creatures.” Is this statement correct even when the outputs include proofs that eluded mathematicians for years? Or books that draw non-readers to read? Does it change when the output is a combination of human/AI contribution? Will a discovery that saves lives also be meaningless?
The book does not address these questions, which I consider relevant to any discussion on AI’s value to its users and society as a whole. Additionally, I find the assertion that work made partially or completely redundant by AI is not “worth doing at all” to be insensitive, and not what I’d have expected from the author. Mr. Warner does as he preaches when thought is involved: he eloquently summarized others’ work and views. His writing showed feelings when it was about his family, about teachers who influenced him. But great writers like Tolstoy wrote convincingly about fictional humans whose experiences they never shared, their imagination and humanity substituting for the lack of personal experience. Such writing is art, and possibly what differentiates us as humans from animals and from machines.
To sum up these musings, the book made valid points about the damage done by AI to individuals and communities, and was pretty clear that indiscriminate use of LLMs would result in students leaving classrooms without acquiring basic skills. I believe that the conjecture that “If ChatGPT can do it, it’s probably not worth doing by humans, and it very well may not be worth doing at all” should be more nuanced to be of any actual use to people and organizations when they weigh in the human cost and long term consequences of using AI vs. its benefits.
A school library. Image from Wikipedia.






A scene from Macbeth, Banquo’s Ghost by Théodore Chassériau 1855,










