Last week, a writer at the Dartmouth Alumni Magazine reached out to ask if I would speak to her about artificial intelligence. She had read my columns in The Dartmouth and concluded I was a critic, while also asking if I was a member of some kind of anti-AI organization on campus. I was surprised on both counts. At any given moment, you can find me running three different large language models simultaneously, and here I was being accused of being an AI non-believer and a potential member of some technophobe cult. I spent 20 minutes digging through my own archives before understanding the confusion. I had never written against AI as a whole. I had written against AI in specific contexts, specifically art and mental health. Convinced I needed to rehabilitate myself, I wrote a piece arguing that AI is a tool and a college education must teach us how to use it well. Eli Moyse ’27 read that piece and wrote a rebuttal. This is my rebuttal to his rebuttal, but more importantly, a feeble attempt at clarifying that I am neither an AI-denialist nor an AI-romanticist, but a mere realist who dares to hope.
Moyse, in his recent column, worries about model collapse, which is the phenomenon where an AI trained on synthetic data begins producing degraded outputs, each generation slightly worse than the last, until the whole thing unravels. He extends the metaphor to Dartmouth itself. If we outsource our thinking, we will, like a model trained on AI-generated garbage, lose the capacity to think at all. It is a good metaphor. It is also, I think, misapplied.
This term I am taking a class called the HIST 40.02: “Intellectual History of Capitalism.” One of our assigned readings was Marx’s Paris Manuscripts, in which he describes what it means to accumulate at the expense of living. He writes, “The less you eat, drink, buy books, go to the theatre or to balls, or to the pub, and the less you think, love, theorize, sing, paint, fence, etc., the more you will be able to save and the greater will become your treasure which neither moth nor rust will corrupt — your capital. The less you are, the less you express your life, the more you have.”
This is alienated labor. Not the labor of the premodern craftsman who made the whole chair, but the labor of the assembly-line worker who tightens the same bolt for fifty years. Specialization extracted efficiency from human beings by making them less than human — less whole, less expressive, more fungible. We have been living inside this structure for 200 years and have come to mistake it for nature.
AI, I want to argue, will not deepen this condition. It will, if we are not cowards about it, end it. Moyse is right that Evergreen AI’s student dialogue writers, sitting down to fabricate a hundred fake conversations between a hypothetical distressed undergraduate and a chatbot, are performing a strange and diminished kind of labor. He is right that some of them appear to have used AI to complete the task, producing synthetic training data from a machine that was itself trained on synthetic training data — model collapse in miniature, absurdist and sad. But the lesson here is not that AI poisons everything it touches. The lesson is that bad uses of AI are bad. Using AI to industrialize the production of emotional support simulations is bad. Using it to automate social media copy is, I would argue, sad but morally fine, not because social media content is worthless, but because no one was ennobled by writing it.
This is the distinction that keeps getting lost. Moyse describes his first real encounter with intellectual life, which was trudging through “Capital, Vol. I,” and emerging a changed man. I believe him. That kind of experience cannot be outsourced, and in no world should it be outsourced. A student who uses AI to summarize Marx has not read Marx. They have obtained a description of Marx, which is an entirely different and considerably less useful thing. I said in my earlier column that judgment is the skill worth building now. I should have been clearer: Judgment trained on nothing is not judgment. It is a pose.
However, here is where I part ways with Moyse’s anxiety. The threat to deep learning at Dartmouth is not AI. It is the same thing it has always been, which is the conversion of education into credentialing, the treatment of a Dartmouth degree as a signal to be optimized rather than a formation to be undergone. Students were using SparkNotes before there were language models. They were writing papers they did not believe before ChatGPT. The problem precedes the technology. What AI actually does, what it does when used honestly, is redistribute the cost of the tedious and reallocate it toward the meaningful. Consider a researcher who once spent 40 hours wading through archives to determine whether a question was worth pursuing can now determine that in four. The 40 hours are not lost to history; they are returned. What the researcher does with them is the question, and has always been the question. Automation has been returning hours to human beings for centuries, and human beings have mostly spent those hours watching television. Or, well, scrolling Reels. This is a human failure, not a technological one.
The deeper hope — and I admit it is a hope, not a certainty — is that AI’s arrival is apocalyptic in that it is a revelation and an unveiling. What it reveals is that an enormous fraction of what we have called "work" was never really work at all. It was the performance of tasks that machines could have done and would have done, if the machines had existed. The assembly line bolt-tightener was always replaceable and it is only now becoming obvious. What remains, what is irreducibly human, is not efficiency. It is expression. It is the novel written because the writer wanted to write it and the reader reads it because they wanted to read it, not because either of them lacked alternatives.
Moyse asks whether he is irrational to resist the vision of humans as prompt engineers. No, he is not irrational. He is responding correctly to a real danger. The danger is that we automate everything and then forget to do anything with the freedom. That we become more efficient at producing things we do not care about, and never return to producing things we do. That Dartmouth trains students to evaluate AI outputs without first giving them anything worth evaluating against. The solution, as I see it, is not to slow the automation. The solution is to take the hours back and insist on doing something with them. To read “Capital, Vol. I,” all of it, with difficulty, because you want to understand the world, not because there was no other option. The monk who copies manuscripts because the printing press has not been invented yet is not more virtuous than the monk who copies manuscripts after it has. He is just less free.
AI will not save us from ourselves. Nothing will do that. But it could, if we let it, give us enough time to try.
Opinion articles represent the views of their author(s), which are not necessarily those of The Dartmouth.
