Moats Against AI

Good writers should not fear AI models. They should adapt instead.

Good writers should not fear AI. Many types of writing that people enjoy reading cannot easily be replicated by machines. Also, writing is a good way to think and you shouldn’t let machines think for you. Instead of trying to compete directly with AI models, then, the best writers will instead adapt and double down on the parts of writing that celebrate their humanness.

Unfortunately, I know many good writers who fear AI models. These writers think that a data center in the desert will soon make their career plans infeasible. Many of these writers are already quitting their blogs to focus on building companies in the physical world. Thousands of journalists have lost their jobs in recent years, in part because publishers think AIs can replace humans. This is true for some types of writing, but almost certainly false for other types of writing.

If you are a writer who fears AI models, you should keep writing anyway. After all, writing is the best way to think, and , the world will soon be divided into the “Writes and Write-Nots,” as Paul Graham has written. The “writes” are those people who can write, and therefore think. The “write-nots” are people who cannot write, and therefore cannot think clearly enough. If you’d like to be a deep thinker in science, at work, or anywhere else, then you should keep writing, even if you never publish that writing online.¹

Another reason to write is to influence the models, as Gwern has suggested. Most humans will soon use AIs to complete a majority of their cognitive tasks, because outsourcing thoughts to a data center is easier than actually thinking. But if you write a lot on a subject, then your views will be incorporated into training data, fed into AI models, and regurgitated to billions of people. Your views can therefore shape the future in strong ways, a bit like how Julius Caesar wrote his own memoirs to mask the fact that he was an egotistical psychopath. Modern writers with strong opinions will be immortalized in the models, even if the things they write don’t reflect their real beliefs or behaviors!

There are other reasons to write, too. But lots of people, including Tyler Cowen, have already described them. I’ve seen far less discussion about how to write to stand out from AIs, though. And so I emailed three bloggers—Ruxandra Tesloianu, Abhishaike Mahajan, and Eryney Marrogi—with the same question: “What kinds of writing do you think are defensible in the age of AI?” All three responded (thank you). All three had good ideas (nice). With their permission, I’ve sifted through those ideas to arrive at an answer. (And yes, before you say it, this does mean that I outsourced part of my thinking to others.)

One of the best ways to stand out, we all agreed, is to make things that only human hands, or the human mind, can make. 

When the camera was invented, artists feared that it would commoditize their work. The artists feared cameras would make it possible for even amateurs to create “art.” That was true, to an extent (look at Instagram), but what actually happened is that the arts were revitalized. A large swath of people began placing a premium on handmade paintings. And instead of merely painting what they saw, painters began to question realism and turn to the abstract instead. Art became an expression of individuality and taste, rather than a one-to-one mapping of reality. The same will come to pass in writing, says Adam Mastroianni:

“It used to be that our only competitors were made of carbon. Now some of our competitors are made out of silicon. New competition should make us better at competing—this is our chance to be more thoughtful about writing than we’ve ever been before. No system can optimize for everything, so what are our minds optimized for, and how can I double down on that? How can I go even deeper into the territory where the machines fear to tread, territories that I only notice because they’re treacherous for machines?”

Another way to stand out is to publish things that nobody else has. Maybe this seems obvious. Whereas many AI agents can search the Internet, they don’t yet have corporeal bodies to meet people face-to-face, in the same sensory environment. But there is a lot of alpha in having real conversations with real people in real places! On-the-ground reporting will retain its value for a long time for this reason. ProPublica’s investigative reporters should not fear for their jobs.

Finding “new things” to write about doesn’t require traveling, either. A lot of information is never captured and published, even if that information seems obvious. Many powerful ideas exist only in the minds of a few people, or are only raised in a single conversation in one bar at a particular moment. Most researchers never publish failed experiments. Most people never think to write about what they did on a particular day, or how normal people reacted when the Internet was first introduced, or what people wore to Woodstock in the 1960s. But even a seemingly simple observation can become an important part of the historical record. 

Other writers will stand out because they are experts on a particular issue.² Readers crave authority, and this will remain true for some time. Many people read the Wall Street Journal to get an economist’s opinion, in part so they can recite that opinion to people at a party later that evening. Many readers “hang their hats,” so to speak, on the opinions of experts.

Brian Potter, the writer behind Construction Physics, is one example. His writing is often raised in online discussion boards because it includes original context that is otherwise missing from the public record. People see him as an expert, and rightly so. Potter reads many books (some of them obscure and out-of-print) while writing his essays, but also speaks with people in the field to gather context that nobody else has. He is uniquely equipped to say, “Y’know, this story in The New York Times says such-and-such, but I met a CEO last week who said that it’s not true for these reasons.” A large language model can’t do that.³

These “writing moats” may make the creative process feel like a painful ordeal. Perhaps it seems like the only people who will make it as writers are those who travel to war zones or go to lots of parties or spend years of their life studying a single field. But that’s not true! Most of my favorite essays have the same format: A person describes their experience with something, and then reflects on that something to arrive at a beautiful lesson. A machine can write prose that appears to reflect on an experience, but the lived nature of that experience belongs solely to the human author.

“Looking for Alice” is, ostensibly, an essay about dating. But its actual power stems from the personal stories and anecdotes scattered throughout—all of which are based on experiences common to all people. “Always Bet on Text” is evocative because the writer takes a strong stance for a thing—text—that they think other people don’t value enough. This essay works because the writer is clearly passionate about the subject, and because they express strong opinions with examples. “I Should Have Loved Biology” does both of these things well. It takes a strong stance, but also incorporates anecdotes and personal stories to drive the argument home; namely, that biology is beautiful, but textbooks teach it in all the wrong ways.

There is absolutely nothing in these essays that is unique to any one individual, or that only experts could understand. None of these essays required on-the-ground reporting. All of these writers simply took personal observations, reflected on them, and distilled the lessons into a singular and poetic message. I love these pieces and yearn, every day, to read more of them.

The ultimate moat, then, is individuality. “In many ways, this is the last moat of everything,” Abhi told me. It’s “people consuming something made by a human purely because they like the vibes of that human.” This is similar to the idea of taste; people consume Scott Alexander’s monthly roundups because they feature esoteric but interesting articles that are rarely mentioned anywhere else.

As I was writing this essay, I began to reflect on my own writing career. I thought about my first staff job at a neuroscience magazine in New York, and how my editor told me which articles to write and whom I ought to interview to write them. I didn’t have much independence at that job, and I was never allowed to express a personal opinion in my articles. So after a year, I moved to work at a small nonprofit in Boston. 

My job at that nonprofit was to write blogs about science. I could write about anything, and my boss encouraged me to express strong opinions. But when I filed my first story, he merely skimmed the text, turned his head to look at me, and said, “This is so boring. Why do you write like this?”

The truth is that my past slew of academic and corporate jobs had neutered my ability to write evocatively and creatively. Up until that point, I had never really stood up for anything in public. Perhaps I was afraid that people would attack me, or that my former mentors would be disappointed in my decision to publish argumentative or opinionated pieces. But that single sentence, uttered by my boss, shook me up. I started writing with fewer self-imposed restrictions. I stopped fearing the reactions of others. I decided to just be myself—to be uniquely human, and not give a damn.

---

Thanks to Eryney Marrogi, Xander Balwit, and Alec Nielsen for feedback.

1. Like Tyler Cowen, I don’t use AI to generate text for my essays because I don’t want to write in the style of the AI. But I do often use AIs as a reading companion, to ask questions, to do research, to find ideas more quickly than Google search, and so on. If your brain struggles against the yearning, aching feeling to take an easy way out—then fight that feeling!

2. And it will become easier to get to the frontier of an issue, and therefore become an expert in a particular domain, because of AI.

3. If original information becomes more valuable to the writing process, then I’d also assume that a writer’s physical place in the world will matter more, too. Deep conversations rarely happen over the Internet or on the phone (in my experience). In-person interactions have a lot of value. If you want to write deeply about biology, for example, then it’s probably best to live in San Francisco or Boston.