Why ChatGPT Is a Balloon and the Book Is a Bicycle
If AI is going to transport literature to new heights, why can't anyone explain how?
[A quick note before the article to say that my second novel, Metallic Realms, is coming out in just three months. The novel was recently named one of the 20 most anticipated books of 2025 by Esquire, and there is currently a Goodreads giveaway. Otherwise, preorders are always welcome and deeply appreciated!]
Are Chatbots Cars, Scooters, or Hot-Air Balloons?
It has been over two years, seemingly millions of articles, and untold billions of investment dollars since ChatGPT came out. Yet the debate about GenAI has remained strangely stagnant. I’ve seen the same arguments copy and pasted for several years now. The core pro-AI argument remains a techno-optimist one: “It’s new and impressive technology, ergo it will inevitably change the world! People who don’t get on board will look like the fools who mocked the personal computer or electricity!” This prompts an easy techno-pessimist rebuttal: “The same people, and often same companies, said the same thing about NFTs, Web 3, AR/VR glasses, and the Metaverse! Stop falling for every tech hype cycle!”
There are other questions with AI, such as the ethnical and legal issues of the world’s richest companies training programs on copyrighted data for their own IP-protected products. But the debate how about AI will or won’t transform society remains “Is it the next worldwide web or the next web 3.0?” Perhaps I can shift to a transportation metaphor. Are GenAI programs the next cars, aka something that will fundamentally change our daily lives and culture (for better and worse)? Or the next Segways, aka overhyped products that are interesting technologically yet ultimately have minimal effect on society and daily life? Younger people might not realize the hype around the Segway was on par with recent NFT or ChatGPT mania. I have a vivid memory of the network news saying we would “have to redesign our cities” around it and no less than Steve Jobs declared it would be “as big a deal as the PC.” Yes, tech titans and media boosters really thought a goofy scooter would transform the world.
The Segway is a useful metaphor because it is an example of technology that worked but also didn’t. The Segway isn’t like Theranos, which can be dismissed as mere scam, or like self-driving cars, which perhaps just need more development time. The Segway worked. You can scoot around on it and its gyroscopic sensors perform as advertised. It’s just that, well, people didn’t care that much about scooting around at moderate speeds. It simply wasn’t useful enough to replace cars much less force us to reconstruct cities from scratch.
But new, radical technology isn’t always a car or a Segway. Sometimes it is a balloon.
I’m currently reading The Age of Wonder: The Romantic Generation and the Discovery of the Beauty and Terror of Science by Richard Holmes. It’s a bit dry, but there is a fantastic chapter on the “balloonomania” craze of 18th-century Europe when the discovery of hot-air balloons led to huge public interest and wild predictions that hot-air balloons would change everything in life. Travel, obviously, but also warfare—Napoleon briefly experimented with a balloon corps—science and everything else. It didn’t work out that way. Hot-air balloons were too unreliable and uncontrollable. Soon, they were surpassed by the adjacent technology of Zeppelins (rigged machines using hydrogen gas or helium) and ultimately the quite different technology of airplanes.
I say the hot-air balloon represents a third way technology goes because while it didn’t succeed in radically changing society in the way of the car it also didn’t merely flop like the Segway. The hot-air balloon signaled a fundamental shift to flight that would be realized, some time later, by other technologies like airplanes and rocket ships.
Increasingly, I wonder if current LLMs like ChatGPT and Claude are the new hot-air balloons. Endless money and labor has gone into these models yet they really haven’t improved all that much. Sure, they’ve taken hold in a few niche areas—students cheating on take-home papers, code writers, Kindle store spammers—but they have hardly changed society on a level comparable to the car, computer, smart phone, social media, or anything else they get compared to. I know people will respond “its only been a few years!” Yet it didn’t take that long for the potential of other recent technology like smart phones to become apparent. People realized quickly what wonders a “pocket computer” could produce (we were perhaps slower to realize the horrors). The only thing that slowed smart phone adoption was money. Early smart phones cost a lot. It took years for a mass of people to acquire them. However, ChatGPT and similar programs are already used by hundreds of millions of people. LLMs are even more popular than the NFTs and crypto coins that tech companies and their online boosters previously predicted would lead to global transformation. The adoption has already occurred. The results just haven’t met the lofty expectations.
Still, there is something these chatbots point to: the dream (or nightmare) of AGI. Actual artificial intelligence. The kind from science fiction. Or at least something far more useful than a generic text regurgitator that, at best, saves you a few seconds of Googling. Yet it seems likely to me that AGI will only be reached by some other technology just as the airplane realized the dreams of the 18th-century balloonomaniac. It won’t be LLMs, or at least not only LLM tech. It will take something else. The fact that DeepSeek was able to recreate OpenAI-level models with minimal money and input points to the fact that simply tossing endless billions of dollars and endless billions of pieces of training data isn’t the path toward science fiction AIs. Making the balloon bigger and the air hotter won’t turn it into an airplane…
“AI Will Revolutionize Writing!” Okay. How?
The above are some thoughts about GenAI in general, but this is a newsletter about writing and publishing. I’m more interested in the narrow question of how these programs affect writing.
There were a couple interesting articles here on Substack about this question that got me thinking about it again. First was Henry Oliver offering the techno-optimist argument. Writers and literary culture have to embrace AI, Oliver says, because not doing so puts you in the camp of people who dismissed “television, radio” and even “the novel and the indoor theatre.” The problem with modern literature “is that it too often refuses to have anything to do with the new sensibility” and “the next new sensibility will draw profusely on AI.” Ross Barkan responded by asking, essentially, what do you actually mean? “Oliver refers to a ‘new’ sensibility. What is it? What literary revolution has A.I. unleashed? What’s the A.I. aesthetic, the point of view, the lingual innovation? What wondrous piece of literature or criticism have these programs produced?” I enjoy both Barkan and Oliver’s Substacks, so recommend reading both articles and checking out their newsletters. Barkan mentions that Oliver will be debating the AI in literature question with Barkan’s co-editor at The Metropolitan Review—a new book review Substack that is also worth checking out—and I’ll be interested to see the argument fleshed out.
At this stage, several years into the GenAI revolution, I find questions like Barkan’s more compelling than more vague gestures toward “the new.” We’ve been hearing the latter repeated verbatim for years now. It is exactly what we heard about NFTs, web 3.0, and the Metaverse. Everyone can agree writers should be aware of what is going on in the world. Plenty of writers, fiction and nonfiction, are already writing about AI. But what about these constant claims that writers must “collaborate” with AI or else be left behind? It has been several years of “just wait a year and everything will change!” Yet we’ve hardly seen AI-collaborating authors rocket past the rest of us as claimed. Not in sales, acclaim, or aesthetic innovation. So what are we really talking about?
GenAI companies brag about having hundreds of millions of regular users. They’ve spent more money and consumed more data than perhaps anything in history. It is fair to ask for results and specifics. AI boosters are still light on those. When I ask AI boosters to spell out what they are actually recommending—what specific collaborations with current GenAI are so vital for writers—the visionary pronouncements rapidly deflate. Grand claims that AI is a “revolutionary technology that will change literature and leave non-AI-using writers in the dust” shrink to small suggestions that GenAI, if prompted correctly, is somewhat more useful for research. Or for spellchecking. Or as a beta reader. (Most of the comments in the above articles are long these lines.) This leaves me saying: Okay. But so what?
Putting all ethnical questions aside, let’s grant these claims. Better spell and grammar checks will be welcome. Quicker research seems useful. And so on. Yet none of this is pointing toward any kind of radical change in literary production or consumption, much less artistic innovation. It doesn’t point to a new type of literature. How is any of this—in the field of literature—akin to the radio, the computer, or the car? I stress the “field of literature” because at this point in the conversation the goalpost typically shifts and suddenly discussions of the artistic use of LLMs for creative writers turns into “well, some other AI programs helped make vaccines! Are you against vaccines?!” or the like.
Others will say, “Okay, yes, LLM collaborations haven’t currently transformed literature. But the programs will only get better. In the future, they will revolutionize literature in some way we can’t predict.” Quite possibly. One cannot argue against such an undefined future. We can only really discuss what exists or is on the near horizon. And the argument that writers (or really anyone) simply must collaborate with AI now is an argument about the present. If something greater is coming along in the future, then the future is when society will have to grapple with it.
AI Is a Technology to Avoid Reading and Writing
The most obvious contribution GenAI has offered in the fields of reading and writing is, well, a way to skip writing and reading. A program to write your boring work emails or summarize the boring emails you receive. A program to generate the text of your take-home school assignments or to give you a quick answer when the professor calls on you in class. This is even how AI companies market their products. The problems GenAI is prompting in academia are real. It is increasingly easy for students to skip doing any reading or writing at all, especially given the number of classes that have been moved online or else allow laptops in classrooms. This is not an English-specific problem. It applies to all subjects. But, the fact that GenAI makes it easier to skip reading doesn’t tell us much about what publishers or creative writers should do.
I will admit that I’m skeptical a technology whose primary utility (so far) is minimizing one’s time spent thoughtfully reading and writing is going to benefit writers artistically. Perhaps GenAI will help with social media publicity posts or the like. But the less time we spend reading with care and writing with intention, the worse our writing will be. Sometimes, you have to put in the work. If you get a machine to hit tennis balls for you, you are never going to develop your forehand. If you using tracing paper to draw everything, you will never learn to draw yourself. Etc. Technological shortcuts may take you to a destination, but they don’t give you the skills to take the journey yourself. [Editing to add: after publishing, I saw this article in 404 about a Microsoft study showing AI use leads to “atrophied and unprepared” human cognition.]
That said, there are certainly artistic uses for GenAI. Several books have been published that use GenAI to generate the dialogue of AI or robot characters, for example. I’m sure someone can write an interesting novel using Claude for research or DeepSeek for character name suggestions. Nothing wrong with any of that. But also nothing revolutionary. None of this points towards any radical change, nor anything that writers writ large must embrace. The main argument here is quantitative. GenAI programs could allow you to “write 100x books in the same amount of time!” This has never seemed terribly thought out. Are LLMs going to produce 100x as many readers? The world already has infinitely more manuscripts than there are readers to support them. If you can’t sell your one novel you poured your heart and soul into, are you going to sell 100 AI-imitations you don’t care about? Especially when countless others are flooding the market with their own slop?
If AI is to become vital or revolutionary for literature, it needs to help produce something different.
Despite my AI skepticism, this seems possible to me. Back when ChatGPT debuted, my first thought was that the use of GenAI for text would be in customization. GenAI has, theoretically, the ability to customize text in ways humans really can’t. Maybe that means authors design digital books that are intentionally rewritten to each reader’s specifications. A simple version of this would be a children’s book where the protagonist is rewritten to have the name and appearance of each child. Or a haunted house novel that rewrites the house’s description to resemble the house the reader lives in. Though more complex and potentially interesting customizations could be dreamed up.
Another version of this would be interactive literature, where the reader interfaces with the text via a chatbot of sorts. Talking with a chatbot is a different interaction than reading a traditional novel. (If AI is transforming how younger people interact with text, I think it is through the interactive chatbot aspects more than anything else.) Perhaps this kind of interface could be harnessed for literature. A crime story where you interrogate the potential murderers, trying to find clues? A Choose Your Own Adventure type story where the possibilities branch of endlessly? “Open-world” texts?
I’m not saying these are necessarily interesting ideas to me. The interactive chatbot novel idea harkens back to early text-based video games, and there is probably a reason why those are less popular today than graphics-based video games. Perhaps you, or someone else, have better ideas about how GenAI could produce new types of literature. But whatever they are, they should be a type of literature that simply couldn’t or wouldn’t exist without GenAI. Better spell check and the like are marginal improvements, not revolutions.
Why the Book Is a Bicycle
Even if such ideas did take off, it does not necessarily follow that it will affect most of literature or be something all writers need to jump on. Even transformational technology does not necessarily touch every area of life. Often, new tech simply produces different pleasures and distractions. The fact that TikTok made short amateur videos popular doesn’t mean that traditional film and television had to shift to cellphone shorts. Just ask Quibi (RIP).
I would expect a GenAI-powered literature, if it emerges, to be its own niche of the market. Not something that replaces traditional books. The reason for this is the book is a remarkably durable technology. It may not be the most popular technology, these days. But it really does what it is supposed to do and it is hard to find many ways to improve it. The book, to continue the transportation metaphor, is a bicycle. It has stuck around.
One thing I’ve noticed in theses debates is that no one mentions the failed history of technologies that were supposed to “disrupt” the book. In the 80s and 90s, “hyperfiction” was a big trend and many great writers thought it would radically transform literature. Here’s Robert Coover in the New York Times rhapsodizing about the “new narrative art form, readable only on the computer, and made possible by the developing technology of hypertext and hyper media.” More recently, the introduction of ebooks and ereaders had many predicting the rise of a whole new genre of literature: the “enhanced ebook.” Basically, these were going to be books in which audio, video, and image files were embedded. Imagine reading a novel about a pianist where you could listen to the music being described. Or a historical novel where you could click on a link and see pop-up videos of Hitler or Churchill speaking. This was the basic idea, although more complex possibilities were batted around.
I interviewed at an enhanced ebook company around 2011. They were convinced the technology could take over publishing. I was more skeptical—and perhaps unable to hide my skepticism, as I failed to get the job. My feeling was these “enhanced ebooks” sounded less like a revolutionary new type of book and more like, well, a crappy movie or a boring video game.
So far, my skepticism has proved right. Ereaders still exist, but they try to mimic the feeling of reading as possible and the text is identical to what is printed on paper. The “enhancements” never took off. It turns out readers actually want to read. They don’t want videos, interactivity, or sound files. Or rather, they want those things in other artforms: music, movies, video games, etc. Readers enjoy getting lost in a text and avoiding the distractions of pop-ups, ads, and notifications. The enhanced ebooks ran counter to exactly what people enjoy about books.
One could argue that some of the promise of hyperfiction and enhanced ebooks were realized by some other artforms. By “Visual Novel” video games, for example, or fictional “weird twitter” personas. But you’ll notice these things didn’t replace or displace books. They are simply other things.
Although my example of customizable and interfaceable stories might point toward something unique GenAI could offer, I’m not sure it is what most readers want. A central pleasure of literature is being in on the conversation and discussing stories with friends. If everyone is reading different versions of a story, that gets lost. (I also suspect that fundamentally people do not want to read computer program text. They want a human consciousness behind the work. This is why the writers who do use AI refuse to advertise that fact 99.999% of the time.) Sometimes you just want to ride a bicycle and don’t need a hot-air balloon attached.
Of course, one might think that books as bicycles is a depressing metaphor. Who wants to be a bicycle in a world of cars and airplanes and jumbo jets? It’s certainly true that in the current culture, film and TV and social media dominate. But people still read books. And people still ride bicycles. I took a break writing this and walked in the park by my house. I strolled past countless cyclists, and the even more of the “outdated” walkers using bizarrely old-fashioned human legs to get around. There wasn’t a single Segway or hot-air balloon in sight…
If you enjoy this newsletter, consider subscribing or checking out my recent science fiction novel The Body Scout—which The New York Times called “Timeless and original…a wild ride, sad and funny, surreal and intelligent”—or preorder my forthcoming weird-satirical-science-autofiction novel Metallic Realms.
Another example of the inexorable failure is 3D films. Sound was so revolutionary well acted silent movies were replaced by stiffly voiced talkies, with people clustered round the microphone. Colour was so revolutionary, audiences ate it up. 3D has been pushed since the 50s. It didn't transform the experience, I barely notice if any film is 3D. I am one of the 25% of people who can't benefit from 3D.
I think you are wrong but in an interesting way! Speaking as a mathematician who is not a genAI expert but does love a good vector embedding.....
Posited: This is a technology with impacts that won't come with better models, but with better human understanding of how to use it in non-trivial ways. And that takes time. (And, totally, there's a lot of triviality coming.) Like, we have not gotten our heads around what it means to be able to write so as to specify a point of view, and then interact with a writer grounded in just that point of view, on demand and all night if we want. It's a supremely weird power but unless you believe that co-writing is always worse than solo writing, there's great potential here when it's understood. Don't you find that feedback, even if it's only partially baked, is highly stimulating for going new places and discovering what you didn't know you believed?
What I'm inclined to try is developing a kind of commonplace book vector database, into which go the ideas that make me vibrate, and then with the points of view that can be interpolated from the space supported by those vectors, enter into conversation. My bet is that you're correct it won't be the people using LLMs to write by balloon, covering great aimless swaths of ground, to do something innovative, but the ones who figure out how to turn their obsessions into little characters that argue back, a menagerie of steelmanning debaters and authorial collaborators slash opponents providing food for reconsideration. Your own Age of Wonders-style bot café, inspiring & infuriating us to go further than the parts we contain within ourselves could go alone.