AI Writing Proves The Author Is Very Much Alive
ChatGPT may strive to mimic humans, but its lack of humanity is the reason we’re interested at all
by Connor Wroe Southard
We’ve all been talking to the machines. Just now, I asked ChatGPT—the generative language model created by the startup OpenAI—“What is best in life?” I received this answer: “The concept of ‘best in life’ is subjective and can vary greatly from person to person. Some may find happiness in their relationships, others in their career or hobbies, and others in their spiritual or personal growth. Ultimately, what is best in life is different for everyone and can change throughout a person’s life.”
The strengths of this answer are obvious: it’s coherent, it’s measured, it’s grammatical. The weaknesses are similarly clear: it’s dull, it’s vague, it’s generic, and the machine didn’t get that I was making a Conan the Barbarian joke.
The common thread among almost every ChatGPT response I’ve seen during the past several months of their meme-ification is that they read like the kind of thing a diligent eighth-grader would write in the “short answer” section of a class quiz. ChatGPT tends to write tidy prose, and its work coheres around the topic at hand. But it also never manages to say anything of substance beyond restating basic facts. When it can’t give you a fact, it defaults to the bromides we humans tend to use when we’re first learning how to write an essay. This is all interesting, of course, but not in the way that ChatGPT’s sometimes morbidly anti-writer cheerleaders make it out to be. Instead of sounding the death knell of authorship as we know it, ChatGPT and similar artificial intelligence help us see why we care so much about authors in the first place.
The earnest student vibe of ChatGPT inspired me to ask it a series of softball literary-ish questions. I can’t say it would have earned high marks in any class I’d be likely to teach. A major goal of any education that goes beyond basic reading and writing is, of course, the ability to one day be able write something more incisive than the answer ChatGPT gave me when I asked, “Why do readers continue to appreciate James Joyce’s Ulysses?”:
“James Joyce’s Ulysses is considered a masterpiece of modern literature for several reasons. First, it is a highly innovative and experimental novel that pushes the boundaries of what is possible in the form. Second, it is a deeply symbolic and allusive work that explores a wide range of themes and ideas, including the human experience, the nature of language, and the relationship between past and present. Third, it is a highly readable and engaging novel that is able to convey complex ideas and emotions in a clear and accessible way. These combination of factors have made it a classic work of literature that continues to be read and appreciated by readers around the world.”
Putting aside the rare grammatical slipup—it should be “This combination of factors”—and the highly contentious assertion that Ulysses is “readable,” as well as the question of whether anything it does is “accessible,” this is the kind of paragraph that would probably score well in the eighth grade, though not much later. I say “probably” because we’d need to know something about the eighth-grader in question, as well as their teacher, their school, and so on. We’d want to know something about who was writing this paragraph before we drew conclusions about its significance. We’d be thinking about authorship.
What the machine has is the exotic, uncanny allure of its authorship.
We do know who’s writing the ChatGPT responses: a machine. Or more precisely, a much-hyped piece of software created by a well-funded startup. Software that we’re told will change the world, perhaps ending human authorship as we know it. And that, of course, is why we care. If I gave you that same paragraph about Ulysses and told you it was written by an eighth-grader, the only reason you’d even skim it would be the novelty of someone that young having something to say about a novel that’s rarely taught before the later years of college. But once you’d skimmed it, you’d conclude the eighth-grader didn’t, in fact, have much to say. Just like the machine doesn’t have anything useful to say. What the machine has is the exotic, uncanny allure of its authorship.
This seems like the appropriate place to lament that Roland Barthes’ 1967 essay “The Death of the Author” has become a cliché of popular literary discourse. Like many of his post-war intellectual contemporaries, Barthes had the good fortune of picking low-hanging fruit: there really was far too much focus on authorial biography and psychology back then, and telling sophisticated readers to focus on something else was an important intervention. But now, decades of confusion later, we live in a literary culture in which we regularly hear both that the author is dead and that they’re cancelled.
It’s in cancellation where we see the unavoidability—and inconvenience—of the author’s ongoing aliveness. We tend to be zealous, particularly lately, about putting our moral judgments of writers at the center of our discourse about their work. Look no further than the friend I’m certain you have who has forsworn their Harry Potter fandom because they’re disappointed in J.K. Rowling. Clearly, our idea of the author—that conscious, contradictory, often unhelpful presence flitting spectrally through the room while we read their work—is indispensable to us as readers. Even when we can’t say why, we’re fascinated by the question of where these words came from, at least in cases where we care about the words.
I say this not as a statement about how we should read in an idealized situation. (That kind of frictionless ontology of the text is better left to undergraduates.) What I mean is that, as a practical matter, we absolutely do use Count Leo Tolstoy’s Anna Karenina as a quasi-literal guide to the political economy and social mores of his milieu in Imperial Russia—as well as perhaps a study in the kind of women Tolstoy wanted to sleep with, as well as (again, perhaps) why he hated them for their allure. We believe that My Antonia can teach us not only about the frontier becoming the heartland in the history-book sense, but also about how it felt to be there, to watch it play out and to see it shape lifelong relationships. We believe that Moby Dick can educate us not only about the whaling industry, but also about the somehow both spritely and grandiose homoerotic energy of its author.
This is how we actually read, no matter how sophisticated we are, no matter how many professors or journal articles have warned us not to be too literal, not to believe that a piece of work that some peculiarly driven person spent thousands of hours on would tell us anything definitive about its author. This instinctive feeling that the author is there, and that it always matters that they’re there, is impossible to separate from the act of reading.
It may seem dumb to say it—and here is where I get to feel like I’m writing my own “Since the dawn of time…” student essay—but it seems we could use a reminder: we want to learn about human experiences and imaginations and intellects, particularly when they’re revealed to us in engaging ways by humans whom we consider to be talented. We might also be curious about the non-human counterparts to all of these things, though it’s proven hard for us to get access. Which is exactly why ChatGPT’s generic, insubstantial, dull responses are compelling in spite of their literal content: We’re interested in the machine that wrote them.
This instinctive feeling that the author is there, and that it always matters that they’re there, is impossible to separate from the act of reading.
Once we’ve realized how much we care about ChatGPT’s bad writing solely because of its author, the notion that software will devalue or destroy human authorship starts to look silly. The machines will definitely get better at what they do. There are indeed whole categories of writing where authorship is not that important to us, such as the infinite reams of web copy churned out to game Google’s search mechanics. There will probably be fewer humans creating that dross in the future. This would be a job market loss, and that’s a material cost worth mourning. But the automation of our least distinguished forms of writing is not a blow to “literature” or “culture” or the other loftier deities we want to protect from the machines.
And if the machines get “better” than any human? I’m not convinced that the “better” of the machine and the “better” of the human author exist on the same axis. It’s entirely possible, for instance, that ChatGPT or one of its inheritors will one day write a piece of reportage about fighting in the Spanish Civil War that is, in some aesthetic or neurochemical-response sense, “better” than Homage to Catalonia. But that piece of software won’t have actually been in the trenches with the POUM in 1937, and so I won’t be particularly interested in what it has to say about all that—except insofar as I would be curious to see a machine take on the challenge of competing with Orwell. I’d go to that synthetic Homage for something entirely different than I want in the real thing. If the ersatz Homage has value, it’s a much different kind of value than what we get from Orwell and his bullet-punctured throat.
I feel a little bit sorry for the machines. Once it stops being a novelty to read something written by a piece of software, will ChatGPT be reduced to scutwork? Will it be too depressed to create something fresh?
I hope not. I want to read its memoirs.
Connor Wroe Southard is a writer based in Wyoming.
