A few days ago, I asked myself this question: “What happens when AI can say what you want to say better than you?” That question only scratches the surface of the malaise. For more than a year now, I have struggled with what to do, as someone who likes to write, when writing becomes ‘cheap’. Or should I say, to be more precise, that it is actually generating written artifacts that has become easier. There is a feeling that, now that one can write enormous amounts of blogs, essays, and reports, each one becomes less valuable. Text is no longer rare and precious like in ancient times. Just as synthetic diamonds are killing the natural diamond market, there is a sense that the synthetic writings we currently have wide access to are degrading the value of traditional human writing.
Even if you don’t believe writing fits the laws of market economics, the abundance of text is, by itself, disconcerting. How do you choose what to read when there are so many things to read, and literally anyone can produce hundreds of text artifacts per day? In the past, the fact that something was written often meant somebody took the time and poured some effort into putting together these words and sentences. Effort was a proxy for whether something was worth investing time in reading. Sure, that was already partially abused for the purposes of SEO, with blogs written following a very specific script, clickbait titles and articles, and so on. I would say it was still easy back then to separate the signal from the noise. You could feel the difference in care by reading a few paragraphs. You could sense the marketing copy trap, with the short sentences and the overall fake feel-good atmosphere. Now the writing is artificial, but also convincing, somewhat factual, somewhat pleasing to read (because it is optimized to be pleasing for humans to read). The confusion is more powerful and seductive. So what do you do? Why spend hours, days, reading what could well be AI writing? Why not summarize it using AI and just read the quintessential parts as presented by AI? Reading is no longer something you do to enjoy the hatching of someone’s thinking, one line after another. No, just get the TL;DR. There are tons of other things to read in order to “catch up”.
There is something upsetting about realizing you can vaguely explain an idea you have to an AI chatbot using voice mode and get a sharp articulation of that idea back. An articulation that is even sharper than your raw idea, identifies weaknesses you didn’t even think about, and refers you to existing literature you barely knew existed. That’s good and bad. Good because it means you can quickly move from vague ideas to something solid to present. Bad because when the AI seems to know more than you about a topic, there is a temptation to stop thinking critically and just be submerged by AI’s apparent knowledge, tricks, and fluency. So you end up with an idea that sounds strong but might not be beneath the surface.
Writing itself is a painful process, painful in the sense that it requires silence in a noisy world, focus in a world of divided attention, and effort when passive consumption would be easier. Good writing even requires you to take some distance from your own idea as you put it into words. You need a certain degree of self-criticism for your arguments to be sharp, self-awareness for your emotional narratives to be complete. But like everything else, the more you practice it, the less painful it becomes. For some people, it even becomes a very enjoyable experience, even if there is still effort to be poured into the craft, into the organization of thoughts.
I think writing forces you to polish your thinking. Having a vague idea in one’s mind is easy. It does not have to be precise; you do not have to confront it with objective standards. But the process of putting that vague idea, that thought, into words forces you to commit, forces you to see how imprecise it probably was, how ugly, how self-contradictory in some parts. I do not know why writing is so good at making us see these gaps. I don’t think writing is the only way to see these gaps. Dialogues or debates seem to have similar properties. Which makes me believe the key is the confrontation of one’s thought with an alternative thought, one emotion with an alternative emotion, one narrative with even a modestly different one. In that sense, writing forces you to confront your half-baked thoughts by externalizing them and making you look at them from a distance. That’s when you see the gaps, the awkwardness, the parts that need strengthening.
Writing is also a search process. A lot of times, you start writing without knowing exactly where a specific line of thinking will lead you. You know more or less vaguely what you want to say, but don’t know how. So writing actually helps you clarify what you really think, what you are willing to think, or what you are willing to let other people think you think (since there are a lot of people who write stuff they don’t believe in, either for ideological, commercial, or other practical reasons).
As much as I like to idealize the art of writing, it is important to admit that not all writing is the same. There is good-faith writing, and there is bad-faith writing. There is honest writing, and there is dishonest writing. Because at the end of the day, writing is just an incarnation of thinking, and if the seeds of thought are tainted, so will the outcome be. Text is ultimately a tool, and as such, neither pure nor impure per se, but usable to achieve all sorts of aims. That’s why I often feel conflicted about subcultures that put such a huge premium on writing. In those, being a writer, regardless of how precise, just, fair, or even sane what you write is, automatically puts you on some kind of pedestal. Mastery of the craft gets confused with the depth of the soul or the purity of intentions. If something can be written so beautifully, it has to be right, right? No.
Remembering that text is first and foremost a tool for communication can help us better understand how to deal with synthetic text. There is a lot of communication theater going on in the world, depending on the place and its function. So, some text that was costly to produce in the past was often too costly given the function it served. For that kind of text, reducing the cost of producing it seems like a great thing to me. For example, let’s say I send you an email asking about the what, when, and how of something. I don’t care if your answer back to me is AI-generated or partially AI-edited, as long as you answer the what, when, and how I asked about precisely, and those answers fit the objective reality of whatever we are talking about. So, to me, it is great that a lot of professional emails will often be AI-generated from some human input (voice notes or bullet points). The only problem I see in that case is that when writing is cheap, people tend to write a lot… And laziness often means the text generated is not fact-checked, and hallucinations propagate. As long as veracity and conciseness constraints are maintained, I think a lot of routine professional emails benefit from being AI-generated or AI-edited. You spend less time overthinking the form of your email; your productivity goes up. Just don’t send two pages that could’ve been two sentences.
We tend to forget that writing was first invented for record-keeping and later heavily used for administration. Using it to tell stories and express personal emotions or opinions only came later. I think for record-keeping and administrative use cases, it doesn’t matter if the text is AI-generated or AI-edited, as long as it is accurate. It becomes complicated when the writing is personal. Would you be happy if your lover sent you an AI-generated love letter? Probably not, for the simple reason that the words would mean nothing if you knew they weren’t written by the sender. It would feel like a lie, a deception, a manipulation. Because what you expect from a love letter is not that it be grammatically perfect, but rather that it be the raw expression of your lover’s feelings. The moment the words are no longer chosen and the sentences carefully crafted by that person, the purpose of the love letter is completely defeated. People seldom write and send love letters these days. And I suspect some were deceptive in the past, letting somebody else write the letter for them, then sending it. A lot of famous writers we know, and some we don’t, have used ghostwriters or assistants without crediting them. Is that a form of deception similar to using AI without revealing it?
I think a lot of people are developing a firm resistance, almost a gag-reflex-like rejection of AI-generated text or content. And given the context, the abuses, the insufferable slop, I get where it is coming from, and I am sympathetic. I also feel like it can be dangerous. If the email perfectly expresses the need its author wanted to express and is faithful in its characterization of the author’s intent, why do I care if it is AI-edited, AI-generated after some refinements, or not? This, I think, is the new normal, but we will need time to adjust. Some people will try to abuse this by sending careless slop emails (fully AI-generated emails, longer than necessary, generated without care, often inaccurate or somehow misleading), so at some point, we are going to need to filter not only spam but also slop, or at least find a way to handle it.
Just as a lot of AI videos are already getting traction on social media, I think there is space for AI-generated or AI-assisted poetry, short stories, and novels. Like always, the bottleneck will be the distribution of the content and its quality. If you think people are going to boycott AI-assisted novels, for example, and go read a poorly written human one, then you are going to be shocked by what is likely to happen in the future.
I do agree with the pope. Fully AI-generated content will always be ontologically different from human creative outputs. But the thing is, almost no high-quality synthetic text is fully AI-generated without some significant amount of human input, direction, or editing. So the ontological difference is between purely human-created texts and those produced at the end of a human–AI collaboration process. I am conflicted about this issue because I sure want more human creative outputs in the future. I want humans to continue expressing themselves and getting appreciation from fellow humans. It is part of our collective experience. But I also think, as always throughout history, that will take new forms as technology changes. First some drawings in caves, then some writing, some sculpture, maybe some AI-generated artifacts in the future, at least partially. Also, there is a continuum from fully human-generated to fully AI-generated. There is a difference between something fully AI-generated from a loose initial prompt and something AI-generated through a lot of back-and-forth between a human and AI. It’s a bit like photography didn’t kill painting or 3D printers didn’t kill the art of human sculpture. Or synthetic music tools didn’t kill classical music, but liberated a new generation of artists and new genres. I see AI as a tool in that tradition, a tool that will liberate a new generation of artists and possibly lead to the emergence of new genres, new ways of human expression.
Wow, writing is wonderful. When I started this essay, I had a vague idea about what I wanted to say. And after several days of writing, pausing, doing something else, coming back, re-reading, iterating, here I am. I discovered my own thinking, or maybe I just thought during the whole time, and it led me here. A weird place where I stay hungry for human expression in all its forms, traditional or new. My gut tells me it doesn’t have to be a zero-sum game.
Maybe the real distinction, when it comes to text, will be between low-effort text and high-effort text. Any kind of writing, whether human or AI-assisted, can be either low-effort or high-effort. AI slop is low-effort AI-generated text. But there can be completely human-crafted slop too. In both cases, you can often sense it, and it has nothing to do with em dashes or certain sentence structures. It has everything to do with the sloppiness of the output, the lack of accuracy, the hallucinations and lies, the imprecision, the parti pris that pretends not to be a parti pris, the sheer ugliness of it all.
On the other hand, some AI-assisted writing can be the result of polishing a draft that took a lot of human effort and care. You can feel it when you read it. A lot of that type of writing is useful, functional, effective, albeit written with AI assistance. Hell, AI assistance can even be a great tool to help non-native writers write better, with more clarity, more beauty. Should we fight that? I personally don’t think so.
Ultimately, maybe the only thing that will matter is whether the writing does the work it is supposed to do, beautifully and with care, and whether enough effort was poured into it.