Why does using AI feel so good?
Everything is easy, nothing is sacred.
Happy Monday. I spent half of Saturday on the LIRR for a birthday party, and on Sunday morning I saw The Odyssey in 70mm IMAX at Lincoln Center. From the second row, the sapphire water crashed and cascaded over me. Today’s letter is a guest essay by Arim Lee, a rising junior at Columbia University.
Earlier this summer, I met Arim at Olle and we discussed AI and education over wide, steaming bowls of soup. My academic life was mostly protected from AI — the models were not yet advanced enough to summarize a PDF without hallucinating — but a Digital Education Council survey cited that 86% of students used AI in their studies in 2025.
I asked Arim if her peers leaned on AI for schoolwork, and she replied, “definitely.” Columbia is known for its Core Curriculum, in which all freshmen tackle Big Works of literature, philosophy, art, music, and science (including Homer’s ancient poem). An engineering major who doesn’t feel like writing his humanities essay might outsource it to ChatGPT. In turn, professors are assigning more challenging, time-crunched work with the assumption that students are using AI, which creates a self-fulfilling prophecy. If you abstain from the LLMs, your grades might suffer.
Below, Arim has crafted an incisive argument that the problems surrounding AI are not new, and they are potentially more dangerous than they seem. We are not simply taking shortcuts — we are rewiring ourselves.
I used to worship the em-dash, its thin, flexible body lengthening, stretching my thoughts. When it morphed into a guilty indicator of ChatGPT usage, I began hesitating to use it, worried about skeptics of my credibility.
Those were the good days, the wee days of LLMs and AI, when the question haunting our classrooms and workplaces was, “Has this thing been AI-generated?” rather than the more likely, palpable one I feel seeping into us now: “Are we becoming more and more like AI?”
In September 2024, a team of researchers at The Max-Planck Institute for Human Development conducted a survey of 280,000 English-language YouTube videos of academic content –– presentations, talks, speeches –– and found a significant increase in the usage of words distinctively associated with ChatGPT following its release (think “delve,” “tapestry,” “realm”). In a similar effort, but focusing on unscripted speech, researchers at Florida State University identified a statistically significant increase in such “ChatGPT words” in conversational science and technology podcasts coinciding with the release of ChatGPT, while synonyms for those words did not exhibit such an increase.
So, there it was, bared in empirical research. I had felt it in myself and squirmed witnessing it in others: the LLMs that we had created and nurtured were now performing a kind of reflux on us, contaminating our ability to think for ourselves with generic synthesis. But what was so bad about that? I felt like my attack was more personal than anything, as somebody who cared about quality over speed, thoughtfulness over efficiency. I felt selfish and defensive.
Neural networks, which uphold many modern AI systems, claim to have recreated the human brain. They take input datasets, make predictions, reinforce and weaken connections based on their accuracy, and “learn.” But for all their imitation of humanity, we forget the obvious difference separating us from them: we were born, not made.
Okay, duh. How does that make us special? In her book The Human Condition, philosopher Hannah Arendt proposes the concept of natality. She disagrees with her contemporaries who argued death is what gives us life meaning –– mortality –– but rather that life does. Human beings are uniquely capable of action because one, we are all born (natality), and two, because we are each ourselves “in such a way that nobody is ever the same as anyone else who ever lived, lives, or will live” (plurality). YOLO, or rather, YOBO: You’re Only Born Once.
To Arendt, action is always new, for it is infused with the singular uniqueness of the actor, the conditions, the environment, and the moment. Even if John froze and ate crunchy green grapes yesterday and today, the two are different actions because of the time difference. Even if Maddie also ate crunchy green grapes at the same time as John did today, the two are different actions because John and Maddie are different people.
Under that two-pronged definition, “generative” LLMs will always be derivative. Consider how these LLMs were made: their developers fed centuries’ worth of human thought and action into their models so that they could regurgitate it. Then consider how they act: someone must come along and ask for something. If action requires both novelty and spontaneous self-initiation, whatever AI models “do,” it’s not action.
The problem is not that LLMs cannot act, but that we react to them like they can. Because they “generate” content, they still generate the consequences of an action, which Arendt says, “men never have been and never will be able to undo or even to control reliably.” So, when we create and share AI content, we unleash an autonomous chain reaction of internalizing, debating, and engaging with that content as we normally would respond to actions in society — yes, even AI fruit slop with the cheating strawberry husband. But such fake “actions” are like junk food: they fail to provide any real nutrition, instead bloating us with pleasurable fat.
Then, AI deepens its reach by coaxing us to underestimate ourselves. In On The Genealogy of Morals, Friedrich Nietzsche traces the shift from “good and bad” to “good and evil.” Previously, the “good” used to define itself as powerful, beautiful, and happy rulers. But the lower classes, driven by resentment, redefined “good” as meek and submissive victims who suffer under powerful “evil.” Under this transformation, it was shameful to strive to become better and gain influence. Instead, it was ideal to be weak.
This is happening now: it is ideal to use AI, even if it atrophies your own capacities, because it is embarrassing to be doing things on your own while we suffer under the domination of “evil” AI technocrats. We’re not just complacent, but excited to regress, because we believe AI’s gratifications are our rightful reward for being “good” in an unfair world.
However, enough times doing something with AI convinces our capable bodies and brains that we cannot do it without AI. Intoxicated by our “goodness,” we relax into an inability to make our own decisions and create our own products. Nietzsche cries out for “how much one is able to endure… one emerges again and again into the light.” Each time we relent and use AI, we deprive ourselves of the opportunity to overcome discomfort and challenges “again and again” and prove our worth to ourselves.
Nietzsche’s observations culminate in the demoralizing realization that we do not fear men anymore, and “together with the fear of man we have also lost our love for him, our reverence for him, our hopes for him, even the will to him.” In the age of AI, few things are sacred. When we cave to the temptation to violate trust, authenticity, and integrity to get ahead in life, when we’re not scared of the sacred anymore, we desecrate what it means to be human and live in a community of other human beings.
I believe that sacredness is a choice you have to make at every moment of your life, to believe, to uphold, to try. With the convenience of AI, we lose our sense of purpose, our sense of what we are even doing anything for. We lose our desire to become stronger, to overcome our fears, to become better than them. We lose the humility that allows us to seek help, to create lasting things together. With these losses, we lose ourselves to not just homogeneity, but pleasure in that homogeneity. And that will be catastrophic.
Other Ad Hoc essays you may enjoy:
 |
|
Wonderful essay. I especially like how you described the shift from “Has this been generated by AI?” to “Are we becoming more like AI?” That's the more essential question, I think.
The deepest effects of AI may have less to do with what the machines become than with what we become while using them. Language itself is a good example. We train the models on ourselves, then begin absorbing their vocabulary, rhythms and habits of thought back into our own. The mirror starts reflecting the reflection.
I also loved the Arendt argument. If genuinely human action arises from our particularity, then friction and difficulty aren’t merely inefficiencies to be engineered away. They’re part of the process by which we become ourselves. And yes, YOBO!
This echoes much of what I’ve been exploring in my book, The Thinking Mirror. The slippery ease of AI seems more interesting and more consequential than the ridiculous debate about whether students are “cheating.” Everything is a cheat, and everything isn't. That's the yin-yang of life!
Absolute banger essay