I knew something was wrong with my writing about AI before I knew what it was. The essays weren't dishonest; I believed what I was writing. The arguments even followed. Moreover, I wrestled with contradictions rather than editing them out simply because they made me uncomfortable. By most standards, I was doing exactly what I believed writing was supposed to help me do: think. And, yet, every time I finished an essay, I was left with the strange feeling that someone had gone missing.
The missing person was often right there on the page: the non-native speaker was there; so was the person with ADHD or dyslexia, or the person who knew they could think, but had always struggled to produce the written evidence required to prove it. While I could acknowledge that some people might experience AI as the assistive technology they had always needed, rather than an intrusion into their thinking, I increasingly realized that I rarely allowed these people to determine the terms of the conversation.
Once I noticed that I didn't allow the pro-AI people as much room in my essays as the anti-AI side, it became harder to ignore: it became harder to ignore that the pro-AI side was relegated to my essays' margins (a paragraph or a parenthesis), while the anti-AI concerns earned most of the essay's vocabulary, emotional seriousness, and intellectual real estate. I had produced nuanced essays, perhaps, but nuance had allowed me to miss something important: two perspectives can both appear on the page without possessing equal power over its architecture. This was particularly uncomfortable because I believe that writing is thinking, and, yet, my writing consistently failed to fully encompass what I thought.
Writing is thinking, that much can be true. But, what happens when your thinking carries another person's finished musings, not your own? Consider a sentence from one of my own essays: "I write to think and I think as I write, which means that polish should not be as important to me as are insight and original thought". "Which means". I barely noticed those words when I first wrote them, because the inference felt natural. If writing was a process of discovering what I thought, surely authenticity and discovery mattered more than polish. However, I later realized that this idea didn't describe my relationship with writing nearly as well as I had imagined. Polish matters enormously to me; it isn't decorative residue applied after the real intellectual work is done, but rather the container that makes thought receivable in the first place. An insight that can't survive transmission has a different social life from one that can. More importantly, nothing about "writing is thinking" logically required the conclusion I had attached to it. Somehow, I had internalized a set of oppositions that made the conclusion feel inevitable: authenticity versus polish, thinking versus assistance, human struggle versus machine convenience. Once the concepts had been arranged that way, "writing is thinking" seemed to tell me which side I belonged on. The vocabulary had arrived with relationships already installed.
This is what I find more unsettling than simply being persuaded by a bad argument. When someone tells me what to believe, I at least have a proposition to resist. Frames, on the other hand, can operate earlier; they can tell me what the problem is, establish whose experience represents the baseline and whose requires explanation, and even provide the questions through which disagreement becomes possible (does AI destroy authenticity? Does AI make us worse thinkers? Does AI outsource cognition? Can AI-generated writing have soul?). There are thoughtful answers to all of these questions, including answers favorable to AI. However, I have started wondering how much intellectual power is exercised before anyone answers them. What I mean is this: if I spend all my energy proving that AI doesn't destroy authenticity, I may never ask whose definition of authenticity established the trial. More, if I spend an essay demonstrating that AI users still think, it may never occur to me that I have already accepted that their cognition is the thing requiring proof. It is possible to disagree with the dominant answers while remaining captive to the dominant questions.
That realization keeps returning me to one deceptively simple word: whose? Whose authenticity? Whose humanity? Whose experience of writing? Whose definition of thinking? Whose vulnerability becoming the ethical concern? And whose finished thinking becoming the material from which everyone else is expected to begin?
When I tried to stop merely representing the pro-AI perspective and actually place it at the center of my writing, I expected, perhaps naively, to feel liberated. Instead, I froze. I suddenly felt as though I had lost my epistemic armor. Within the familiar AI discourse, even disagreement came with infrastructure, such as established concepts to respond to, recognizable ethical concerns, and familiar language through which seriousness could be performed, among other things. I could push against conclusions while remaining intelligible inside the conversation. Centering the perspective I had repeatedly relegated to the margins felt different. Instead of simply answering the established questions differently, I was beginning to question why those questions had acquired priority at all, which left me paralyzed.
That freeze complicated the story I had begun telling myself. It would have been comforting to imagine that beneath the inherited frame was my true, fully formed position, waiting to be liberated, that removing the intellectual constraints would unveil the essay I had always wanted to write. There was no such essay. Instead, there were intuitions and thoughts without any alternative intellectual architecture capable of holding them all. It became clear, then, that frames, even inherited, don't merely restrict movement: they also hold things together. Inherited frameworks, while potentially captive, give us the language through which we reason about the world. They also make some complicated ideas thinkable without requiring every person to rediscover them from scratch.
The problem, then, can't be that I was thinking with other people's ideas. The problem becomes harder to name: what happens when some people's finished thinking becomes so structurally available that it begins to feel like the neutral terrain on which thinking itself occurs? And what happens to thoughts that never make it into view? Perhaps the easy answer is that they lose. If an idea can't be articulated clearly, there simply isn't much there to take seriously. However, a thought rarely arrives in the world with the language required to defend itself. It might begin as discomfort or intuition that something about the available explanation doesn't quite fit. Given enough room, that discomfort might become a question; the question might become a hypothesis; the hypothesis might encounter resistance and become more precise. Eventually, perhaps, it develops enough structure to enter the language in which reckoning is allowed. This is easy to overlook once an idea has acquired authority because we encounter the finished product. We see the book, the theory, the vocabulary, the citations, the carefully worked-out distinctions. Rarely do we see the years of conversation, disagreement, failed formulations, institutional attention, or intellectual labor through which the thought became capable of appearing authoritative in the first place. Under this light, authority becomes something a thought is allowed to develop, rather than something it possesses.
That raises the uncomfortable possibility that not all thoughts receive equal opportunity to undergo that development. While it is true that every thought doesn't deserve attention, some thoughts are unfortunately required to arrive already possessing the very qualities that sustained engagement might have helped them develop. They can be expected to be more clear or precise than would warrant the stage at which the thoughts are currently being experienced. In other words, the thoughts can be asked to become developed before being granted access to some of the conditions through which development happens.
I began to wonder whether this was partly what I had been observing in the pro-AI perspectives I kept relegating to the margins. Often, they didn't arrive as polished theories of cognition or authorship, but as something much humbler: confessions about being helped by the tool or being finally able to do things once unfathomable. While it is true that confessions don't settle the larger questions surrounding AI, their lack of theoretical sophistication doesn't mean that there is nothing underneath worth developing, either. This asymmetry between pro-AI thoughts and anti-AI thoughts can become self-reinforcing. On one side of the debate, we have books, established vocabulary, respected intellectual figures, theories, citations, and years of accumulated argument; on the other side, we have scattered experiences and people struggling to explain why the available account of what is happening doesn't describe them. We compare the two and observe, correctly, that one is better articulated. What we often miss is that articulation itself has a history.
If part of what an unfinished thought needs is sustained attention, then, AI can enter the picture in an unexpected way: a person can bring an unfinished thought to an AI system without first making the case that the thought deserves anyone's afternoon. They can spend twenty exchanges on one distinction, reject various interpretation, watch the thought expand an implication they hadn't noticed, and discover whether the problem lies in the AI's reasoning or theirs. The thought can become stronger or collapse altogether. This gives the person's perspective the center it requires to discover what there is to be right or wrong about. That has been one of the stranger discoveries of writing these essays. Some of the pro-AI thoughts that had previously been unable to be centered, due to unsophisticated language, suddenly become available for inspection. Some survive that inspection and others change. AI helps transform intuition into something that can finally be reckoned with in the language in which reckoning is allowed.
But if I am going to worry about inherited frames, I can't conveniently exempt the machine helping the pro-AI person from potentially providing them a frame through which they reason about their intuition. An AI system can name someone's experience before they have found their own name for it. It can even offer a distinction so elegant that one adopts it before asking what the distinction excludes. Still, it can organize one's intuitions into an argument whose coherence makes them forget that coherence and truth are different things. And, because the language appears to meet the person exactly where they are, the borrowed architecture may sometimes be harder, not easier, to recognize. That may be the most uncomfortable conclusion of all: there is no pristine thinker underneath the discourse, waiting for every external frame to be peeled away. Language itself is inherited; so are many of the distinctions through which thought becomes possible. Even rebellion borrows vocabulary from somewhere. So, perhaps intellectual independence is the wrong ideal. What I should strive for, instead, is greater awareness of the architecture while I am thinking inside it. One way I have learned to notice that architecture is to pay attention to what repeatedly fails to fit: the perspectives that keep ending up in parentheses or the conclusions that follow perfectly from my premises but somehow don't sound like me, or even the question have I answered so many times that I no longer remember agreeing it was the question. While these mismatches don't prove that the frame is wrong, they are invitations to inspect it.
Instead of rescuing one from the frame problem, AI simply enters it. Perhaps it can give an unfinished thought enough scaffolding to become visible. Perhaps it can also build new walls faster than one would have built them themselves. I am less interested now in asking whether I am thinking entirely for myself, and more interested in asking whose thinking I have been given to think with, which assumptions became invisible by becoming default, and which thoughts in me learned to remain small because there was nowhere respectable for them to grow. Writing is thinking, that much is true, but writing doesn't happen on empty terrain. In the end, the task might be to keep noticing who furnished the room and to pay particular attention to whatever could never fit through the door.