You Can't Automate Eudaimonia

The plagiarism detection software Turnitin premiered just as I was starting my graduate student teaching at UC-Berkeley. A friend of mine had signed up for a free trial because she had suspicions of a few students in her class. We both sat in the department lounge, leaning forward toward the screen, waiting to see what the magic box could tell us.

Zero percent detected, said the screen.

My friend, still suspicious, decided to call one of the students into her office hours. Within five minutes, she had determined that the student had not written the paper. Or the previous paper, for that matter.

How was she, a fallible human and not a walking database of every article about eighteenth-century poetry, able to detect what the wondrous new software could not?

A five-minute conversation with the student, that’s how. It revealed that he had no idea what either paper was about. Also, he hadn’t read the poem about which he’d written.

Over the years, I’ve detected dozens of instances of plagiarism in exactly this way. The shortest conversation I ever had began and ended with the question, “What is the title of the paper you turned in?” The longest one, at about seven minutes, eventually identified Wikipedia as the source. These interactions have in fact always run depressingly short, because I have yet to discover the mastermind plagiarist who steals from a book published in 1952, buried deep within the subject-area stacks and forgotten by all. Or even the more recent scholarly argument, cleverly and stealthily paraphrased and infused with the author’s own examples and reasoning.

Once, when I was an undergraduate, I walked into my advisor’s office for our usual meeting to find him with his head in his hands. He’d just realized that an entire thesis he’d supervised was plagiarized, largely from the Stanford Encyclopedia of Philosophy. Outraged, I unspooled my theory of the mastermind cheater, the delver in ancient stacks, and how the more middling and obvious form of cheating was as the petty criminal to James Moriarty, and that we should expect better of a Harvard-educated plagiarist.

He shook his head. “What you’re describing would render the point moot.”

The point, he explained, was to do as little real work as possible. The form of literary thieving I described looked — well, honestly, it looked like the kind of research that the plagiarist was trying to avoid in the first place.

“Laziness,” I said, thinking I’d finally gotten it.

But I hadn’t. “It’s not that, exactly. It’s that this person doesn’t understand why we’re here.”

I’ve thought about this conversation over the course of the last week, as two revelations about the use of generative AI smashed the literary world. The first, the cancelation of a major upcoming release called Shy Girl by Mia Ballard, was the subject of a three-hour viral takedown video where a relentless critic elucidated a catalogue of repetitions and bizarre metaphors that were definitely written by a prediction machine that has no physical senses.

The second was a book review in the New York Times. When the author was questioned, he admitted that he’d fed a review of the same book, printed elsewhere first, as an example to coach the AI, and hadn’t read the output carefully enough to realize that the bot had lifted words and phrases wholesale before he sent it off to his editor.

Embarrassingly, the review’s retraction was published just days after the newspaper of record’s op-ed section went wild railing against Ballard and her publisher, Hachette.

* * *

The question that I’d been hoping to examine through the use of both experience and philosophy was this: “Is there an ethical use for AI in writing?” As you’ll see, the question changed for me as I wrote.

I think there is an ethical use for AI, for whatever it’s worth. For the past year, as a fiction writer, I’ve used AI for tasks like proofreading, continuity checks, and some brainstorming, although looking back at those conversations now, I can see that they’re mostly about me getting angry at the AI’s inability to think about what would actually motivate a person to take an action, and thereby clarifying my own thinking about human motivation. On a few occasions, my reckoning with that common writerly enemy — the blank page — has led me to ask Anthropic’s Claude to generate a chapter draft for me. Again, this has mostly functioned to make me angry and frustrated enough to rewrite the entire thing from the foundation up. On one occasion when I did this, I used track changes to assure myself that I was modifying enough. Not a single word stood by the end of the exercise, including a terrible character name that it had hallucinated.

There’s an argument here that even this limited use is unethical. Published writers are prone to complaining that AI companies have illegally fed their sweat and blood into prediction algorithms that help it spit out the plausible prose that we’re all coming to recognize (“it’s not x, it’s y,” character X had “the unique quality of [something that vaguely makes sense but not quite]”). This is “stealing,” in their evaluation.

My problem with this argument is that when you even slightly scratch the surface of the frequent online railing, the argument isn’t remotely consistent. The same writers who claim that their own (presumably, in their eyes) good prose is being effectively plagiarized by the AI also characterize AI prose as “execrable” and “formulaic” and claim that it’s easy to tell the difference between human-written and AI-written text, and that every real human prefers the former. So in what meaningful way is their prose being appropriated? It’s transformed, by their own admission, beyond recognition, just as it is when I read Macbeth and produce writing that is markedly inferior. If they have quality on their side, and the quality is obvious to consumers, what are they worried about? This is a consistent problem in anti-AI writing arguments: They are both apocalyptic (AI is stealing our prose and will result in our superannuation) and dismissive (AI generates terrible books and prose and it’s trivial to tell the difference).

The result of this fuzzy thinking (probably born, like most fuzzy thinking, out of genuine confusion and terror) is to invert the problem with the arguments of the AI boosters. AI boosters are utopian about its potential effects on human existence, granting it near god-like omniscience and power to solve all manner of problems (including fairly persistent ones, like sickness, aging, and death), while at the same time minimizing any potential impact on the humans that such omniscience would inevitably replace. And once again the ouroboros theory of the discourse holds, if in a complicated double-helix form.

I suspect both branches of this argument are driven by economic self-interest, rather than willingness to engage with consistent, fundamental, philosophical arguments. AI boosters are AI users and in many cases earn their income from AI, and at least we can say of them that this is putting their money where their mouths are. But they are driven by an emotion well known to and despised by seasoned investors: mental overinvestment in your own product. Call it “hope.” AI disparagers, at least in the writing community, are generally people who have made their (often meager) livings for a long time in a particular traditional pipeline: as writers, editors for publishing houses, freelance editors and copyeditors, and agents who represent writers’ interests. They often aren’t AI users and have interacted in an extremely limited way with the technology, if at all. The great twentieth-century horror writer H.P. Lovecraft was famous for remarking as a sort of mission statement for his writing that the greatest fear of humankind is fear of the unknown. Never show the monster under the bed, because what you imagine about it will always be worse than what it is, no matter how many tentacles it has. The AI disparagers thus cower and shrink, attributing extraordinary powers to the thing at which they are unwilling to look, worried that they are gaslamp lighters in a town that is considering adopting electricity. Sure, the soot and grime and occasional danger of setting oneself on fire isn’t ideal, but on the other hand, they’re scraping by and haven’t set themselves on fire yet. When the electric comes in, what will they do?

The insufficiency and self-contradictions of the arguments on either side led me to think that a different question was the heart of the matter: Why do we write at all?

* * *

A question on the way to another. What did my advisor mean by “This person doesn’t understand why we’re here?”

As I watched the AI headlines roll by this week, I found my mind being inexorably drawn back to my advisor’s formulation of the problem and wishing that I could ask him. But he died tragically young, a few years after I graduated from college.

One of the things we’d done together on my way to a Renaissance history degree was read most of Aristotle, including the Nicomachean Ethics. It’s in the Ethics that Aristotle famously formulates his telos theory of right conduct: To ask what we should do, ask first what we are for. Toward what end are we driving? What is the proper nature of the human being? Answer this question, and you get a little closer to what you, the human, should do with yourself.

His central conceptual framework for the telos of the human was eudaimonia. It’s one of those untranslatable words that academics have argued about for centuries (Is it an activity? Is it a state? What is its relation to the soul?) but for most of our purposes, these debates don’t really matter. I can summarize it quickly and for general purposes: It’s a life well-lived according to principles of wisdom and understanding, the success of which is possibly only diagnosed at the very end of life, retrospectively, once you can look at the complete picture with the detachment of knowing that you’re about to die.

What I have always appreciated about Aristotle is the central, if unstated, idea that runs throughout his work: We are what we do. We show what we care about and what matters to us through action, guided by reason. This idea elegantly simplifies much of moral discourse. What is it that we want to spend our time doing? For most of human economic history, the concept of “leisure” and “free time” about which we could make decisions was nonexistent, so difficult was it to feed and clothe and shelter oneself and one’s dependents. Abstract thinking or reasoning about anything other than “how do I avoid this or that marauding warlord or marauding plague?” was a treat severely rationed and usually only available to a select few who were not singled out by merit but by the circumstances of birth.

This is harder to remember, but writing and other forms of artistic creation fell into the same category. Widespread literacy sufficient to make books anything more than a luxury item for the rich was a product of the early twentieth century. I recently obtained my great-grandmother’s 1876 baptismal record, which contained the following line: “the priest signed, because the others cannot write.”

Before this turns into an “uphill both ways in the snow” story, let me get to the point. The ability to spend your time in mental activity is, or should be, the highest end of a human being. My definition is expansive here: academic work, creative work, knowledge work, any kind of work that involves using your mind, rather than backbreaking physical labor, is a privilege and an exercise of the highest human telos, and one that is available to us residents of 2026 in greater abundance than our ancestors would have dreamed. It is an end, in Aristotle’s terms, not a means to an end: a pleasure to be cultivated for itself. Not easy, by any means, but worthwhile.

So why are we so eager to offload it to a chatbot?

* * *

The answer is that I don’t really know. I don’t know for the same reason that I don’t know why I see people dining out at excellent restaurants with loved ones for what are clearly special occasions, and both of them are bending their necks toward their phones. I don’t know for the same reason that I don’t know why I can hate this as much as I do in principle but sometimes catch myself doing it, anyway.

I’ve started to believe that we’ve lost track of the difference between means and ends. We take measures to lighten our workload and create more leisure time — but then we spend that additional time conversing with chatbots about how we might lighten our workload and create additional leisure time. Instead of watching our children play on the swings, we take out our phones to look at pictures of other people’s children playing on swings.

I don’t say this in a tone of moral scolding — I’ve been guilty of all of the above myself — but in an expression of moral hope. If we can identify the traps of modern life, we can begin to figure out our escape plans. Once we can identify the telos of our existence, the thing(s) that matter to us, the ends for which we exercise our means, we can start to move toward eudaimonia and an ethical existence.

And I found, to my surprise, that when I start to think about AI-generated writing this way, answers begin to emerge and clarify.

Aristotle gives the example of the telos of a good musician. What is the telos of the good musician? To play music as well as she can. Note that the answer isn’t “to produce good music” (as if music were a thing that could come into being without connection or reference to a human player) or to “play music well while putting every possible obstacle to this goal in her way, just so she can say that her music was produced with the maximum possible effort.”

Likewise, for a writer, the goal is to write as well as one can, and, with any luck, to improve over time. The activity of writing, and seeing yourself get better, is where the pleasure and joy of it lie, and outsourcing that work wholly to a chatbot would raise a natural question: Why are you doing this? What is it for? If the answer is “to make money,” there are far easier and surer ways to make a buck. If the answer is “to be known as a writer,” I would refer you to what one of the finest Enlightenment moralists, Adam Smith, said on this subject: “Man naturally desires, not only to be loved, but to be lovely; or to be that thing which is the natural and proper object of love.” Smith, of course, is mostly known as the intellectual father of capitalism, but he understood quite well that there is a difference between making a buck and deserving it. Even if no one else ever finds out, we ourselves will know the difference between doing something and claiming credit for it. Will the latter appear on your deathbed as an example of how you achieved eudaimonia?

But on the other hand, there is no moral desert (in Aristotle, in Smith, or in any other reputable ethics) in making things as hard for yourself as possible just for the sake of making them hard. I do not write longhand by candlelight under conditions of extreme starvation with an inky nib in danger of freezing solid at any moment. When I wanted to quote Smith in the paragraph above, I did not consult a dusty tome in a library a day’s walk away. I remembered part of the quotation and punched it into Google, clicked on a link that took me to the full text of The Theory of Moral Sentiments, verified that this was the quotation I was thinking of, and copied and pasted it into my document. This use of technology is uncontroversial these days, but it wasn’t always. When I was in high school in the late nineties, we were asked to sign a pledge that we would never use a “World Wide Web” source. The internet was less reliable in those days, and maybe this made sense. But now, it’s almost impossible to think about writing anything, however academic, without consulting digital sources. Anyone who did so would be considered self-sabotagingly eccentric.

So when I think about how to incorporate this new technology in my life, I do so from the standpoint of a) someone who has seen many new technologies come into my life; and b) with attendance to a telotic theory of writing. How can AI help me fulfill my telos as a writer?

This is the part where I can only resort to practical examples. There was a time in my writing life when I would get stuck on certain things, like coming up with a joke name for a book that doesn’t exist that someone could refer to as both characterization and scene setting. I don’t know why this is so hard for me, or why I can’t content myself with typing [placeholder] and moving on, but there’s something about it that would nag at me until I would stop writing entirely, sometimes for days at a time.

Now I ask Claude to come up with 20 names for me. Nineteen and a half of them are bad, but the remaining half will stir a thought about how to improve it. A task that used to take me several days now takes a few minutes. I’d consider this a solid win for my telos as a writer, as it enables me to, actually, you know, write.

I sometimes use it as a brainstorming partner, after about a year of careful coaching about the things I like in fiction and the things I hate. (Likes: moral change, the coincidence of the external and the internal, jokes about our relationship with animals. Dislikes: autofiction.) Let me emphasize here that this is generally not a time saver. As I said: A year of coaching it out of its sycophancy and tendency to have characters wink-wink nudge-nudge-put-a-giant-hat-on-it-and-make-jazz-hands clues in the mystery novels I write. Using AI well means interacting with it and learning about its uses, in the same way that you wouldn’t get in the driver’s seat of a car and simply expect that you knew what to do. At the margins, though, I’ve found that running a few ideas by it and having it critique them on the basis of what I’ve told it about my project can make some improvements, as long as I remember that, fundamentally, I am talking out loud to myself.

And, as I said, I use it to proofread and check continuity issues (someone’s wearing blue on page 79 but red on page 81). Editors at publishing houses used to perform these services, but — as the Ballard scandal revealed — editors not only do not perform these services anymore, but it is unlikely that a single editor at Hachette read any part of the manuscript of Shy Girl. It was picked up after successful self-publishing and BookTok self-promotion on the part of the writer with regard only to its likelihood of sales, not its quality. If someone — anyone — at the publishing house had read the book, no matter how incompetent an editor, they would have noticed that entire passages repeated themselves. Fiction writers have long been expected to foot their own bills for proofreading and even more substantial editing before they even go on submission to agents and publishing houses, a failure of the editorial process that is the subject for another rant. Let’s just say that editors are failing to exercise their telos as editors and are mistaking themselves for salespeople, failing to participate meaningfully in the process of creation but still wanting a bookish aura to collect in the air around their heads. Outsourcing proofreading to an AI not only catches more mistakes than any professional proofreading I’ve paid for, but it frees up more of my time and money for activities more essential to the writing telos.

Far more importantly, and unrelated to writing, I recently used it to help my elderly father figure out that a medicine that he’d been taking for lymphoma was probably inducing severe toxicity and causing his bone marrow to produce defective red blood cells, inducing severe and apparently untreatable anemia. His doctors were medicating the anemia without treating the underlying cause, which AI was able to isolate quickly. He’d been suffering unnecessarily for months without an accurate diagnosis, and with the frequent, despairing feeling that none of his doctors were listening to him. Every day seems to bring another news article about someone who was on the verge of death until providing an AI a complete picture of their symptoms and drugs and finding a way to treat the supposedly untreatable. I would call this an exercise of my telos as a loving daughter to a sick parent.

What I don’t use it for is to replace the things that make me want to and enjoy being a writer, or a human: wrestling with metaphors and all variety of words; figuring out the exact combination of words on a page that will convey my meaning; to surprise and delight, both myself and others. Otherwise, why am I doing this?

* * *

Twenty years on, I think I finally understand my advisor’s words. He was talking about a certain paradox: a student who had worked very hard to make it into an excellent university and furthermore had actually signed up for an honors concentration that required that he write a fifty-page thesis had missed the opportunity for eudaimonia, the happiness of functioning as the thing that you want to be. “You’re only cheating yourself” has become a cliché uttered with the aw shucks winking cynicism that is the hallmark of the educated person these days, of course, but this doesn’t mean it isn’t true.

When we think about AI, then, we have no choice but to think about telos. What are we attempting to accomplish for ourselves? Does the use of AI assist in the way that Google docs assisted me in writing this piece — by taking away some barriers that would make the inessential parts of the work more burdensome — or is it stripping away the task itself and all of its pleasures and reasons for existing?

But I’ve found that few people are thinking or talking about it in this way, perhaps because we no longer think of telos when we talk about ethics. That our reactions have become so polarized along ideological grounds — with left roughly lining up on the AI scorn side, and right on the AI boosterism side — should clue us into the fact that we are not thinking, but hoping someone else will think for us.

Which sounds, in the end, a lot like outsourcing your life’s work to AI.