In Praise of Slow Writing

“Writers must therefore constantly ask: what am I trying to say? Surprisingly often, they don’t know. Then they must look at what they have written and ask: have I said it?”
– William Zinsser, On Writing Well

“The first draft of anything is shit.”
– Ernest Hemingway

*

While I am not exactly a professional writer, I am constantly writing for work. Scripts, proposals, marketing materials, emails, memos, messages, social posts and essays for this blog: all told, I probably spend 60% of my work week either writing something, editing something someone else wrote, or doing research for something I’m going to write next. And while I like to think I write pretty well, I do not write particularly quickly. Writing, for me, is re-writing. I have to change things around, throw them away, rethink and restate over and over again before I have something that I can stand the sight of. This is particularly true for creative projects. My computer is full of documents with titles like “Project_Rough15.docx”, and my trashcan is overflowing with torn-up pieces of paper. Even for something completely mundane, like an email response or a Slack message, I often have to rewrite the damned thing twice (or three times) before I can make my meaning clear. This is my process, and I don’t always enjoy it, as testified to by the forehead-shaped indentation in the center of my desk.

And so, on paper, I should be the ideal customer for AI large language models. Why bang my head thinking of the right words when the friendly robot could just do it for me? Think of the time I would save! The other things I could put my brain to! But no. I do not use AI to write for me, and I won’t start. Does this make me a hopeless curmudgeon, shaking my fist against the inevitable passage of time? Possibly. But my suspicion of generative AI goes very deep, and I am unlikely to be talked out of it. The whole premise seems fundamentally flawed to me, not just failing to solve the problems set out for it, but misunderstanding what the problems were in the first place.

*

There are four distinct groups who have embraced the use of AI writing most enthusiastically. The first are digital marketers. Specifically, the concatenation of advertisers, salesmen, consultants, and hustle-bro influencers who have fully bought into the concept that more digital impressions equal more sales, with no hesitations or caveats. Their entire strategy and raison d'être is to make you see their name at the top of a collection of sentences as often as possible, with the thought that eventually you will buy something from them — perhaps because the 99,987th spam email has convinced you of something that the 99,986th was unable to. “Sales is a numbers game” they tell you, and so they want more than anything to flood the internet with content, so that every social media scroll shows you twelve of their posts and every Google search yields twenty articles, thirty-seven podcast episodes, and one hundred nineteen videos emblazoned with their branding. Obviously, for them, a digital spigot of automatic content is a godsend, and they will install one in every room.

The second group are the leaders of large corporations, for whom profit margin is a religion and firing people a vocation. They love replacing people with software. A way to dispense with entire creative departments in exchange for a couple of engineers and the occasional freelancer? “Yes please,” they say, licking their lips. “Sign me up for that.”

For both of these groups, the case for AI writing comes down to organizational efficiency, the idea that piles of words can be generated faster and cheaper by LLMs than by actual human creators, and so therefore they should be. I fear this shows a very poor understanding of what writing actually is and what it is for.

*

Imagine, just as a thought experiment, that you had a notion to disrupt professional basketball. You might look at the P&L of NBA teams and realize that player salaries are a huge expense, not to mention the costs of recruiting those players, training them, providing them with practice facilities, and transporting them around the country. If the function of the team is to win games, the way to win games is to score points, and the way way to score points is to put the ball through the hoop, why not look for ways to put the ball through the hoop that are more efficient and cost effective than paying the enormous cost of highly trained professional ball-in-hoop putters? Could we build a machine that would make the ball go into the hoop hundreds of times per minute? Surely we could. Wouldn’t that be more efficient and cost-effective than recruiting top-level players? Absolutely!

Would it be a satisfactory substitute for a basketball game? Obviously not. Why? Because what people like about watching sports is watching the effort the players have put into their performance. We want to see the struggle, the innovation, the creativity, the execution of almost – but not quite – superhuman feats of ball-in-hoop putting. The humanity of the participants is crucial. The audience wants to see a heightened version of themselves, what we could have been had we been blessed with more talent and a better work ethic. Scoring points is the medium of this expression, but it is not the message.

The same is true of reading. The reason it is interesting to people is that it connects them with the thoughts and ideas of other humans. We read to get the benefit of knowledge gained through the lived experience of others, preferably in a way that shows the point of view of the writer. In the case of really good writing, we also get to marvel at the writer’s skill and inventiveness, just like we admire the way that Nikola Jokić passes or Steph Curry shoots. Mechanically spitting out strings of words isn’t the point any more than mechanically putting the ball in and out of the hoop would be.

Chat GPT will fill a page with words about whatever subject you like. Those words will be grammatically correct, reasonably idiomatic, and possibly even factually accurate. They will not, however, actually have been written. They will never form a connection between your point of view and someone else’s. The purpose of reading fiction or poetry is to experience an expression of human imagination. The purpose of reading nonfiction is to be educated about something by an expert. The purpose of reading a memoir or personal essay is to connect to another human at a distance. AI cannot do any of these things.

Even in professional writing, where personality, style, and beauty are not the primary goal, the promise of a page of written words is that it conveys the knowledge of the person who wrote it. I was chatting with a neighbor who works in public health, and he talked about how they are, thanks to ChatGPT, getting competent-sounding work proposals from people who don’t actually know what they are doing. This is a serious problem, because they are in very real danger of awarding important contracts to bad actors who will cause the downfall of important projects because AI is allowing them to call forth a convincing imitation of skills they don’t have. The purpose of the RFP was to demonstrate the capabilities of the bidder, the method was filling in the answers to questions. Completing the second while bypassing the first makes the whole enterprise useless.

*

The third, and frankly craziest, group are the radical life-hack technofuturists, who think creative professionals are unnecessary, elitist gatekeepers standing between them and the media they want to consume. “Wouldn’t it be better” they say, “if I could just type in the parameters of the exact book I want to read or movie I want to see, and have the computer instantly provide it to me? Rather than having to be forced to waste my time consuming someone else’s ideas?”

This crew might solve the basketball problem differently. “Obviously the ball-in-hoop-placing machine is a poor solution,” they would correctly observe. “I want to see basketball players playing basketball. In fact, what I most want to see is my favorite basketball players playing, and for the game to have the outcome that I want”. “What if”, they might then say, “we digitized every NBA game ever played, and fed it into a generative AI that could create digital avatars of every player, combine them on teams any way I wanted, and have every game have the outcome I wanted? Wouldn’t that be the best? I could watch what I wanted, when I wanted, without compromise or disappointment!”

This gets to two fundamental and catastrophic misconceptions at the heart of Chat GPT and its cousins. This first is that they are inherently creators of collage. They can only reconfigure and recombine things that came before.They are incapable of creating something that is truly new. Magic Johnson passed the ball like no one before him, and then Nicola Jokić came along and performed the same task equally brilliantly but in a totally different way. Putting Magic Johnson into generative AI might give you a plethora of different Magic Johnsons, but it would never give you Nicola Jokić.

The second misconception is that ideas are everything. If you know what you want said, creating the actual piece of writing is just “wordsmithing”, a hacky and vaguely disreputable process in which that idea is gussied up into fancy language so fancy people will think it’s fancy. The initial idea is the real act of genius and the rest is tomfoolery. Nothing could be further from the truth. Knowing the result you want isn’t the majority of making something. It's not even the larger half. It’s not even half. The process is everything. The minute details of that “wordsmithing” — the rhythm of how the stages of the plot are revealed, the small observations of characterization, the tiniest moments of description — are what separate the dreadful from the readable and the ordinary from the transcendent. Typing “the good son of a mafia family is forced to turn tough and take over his father’s criminal enterprise” is not the same thing as writing The Godfather, typing “ship captain is obsessed by enormous whale who attacked him, chases him for years and then gets eaten” is not the same as writing Moby-Dick.

And of course, getting exactly what we asked for prevents us from being surprised. Think of the greatest books you have read, plays you have seen, music you have listened to, movies you have watched, were they great because they exactly met your expectations? Of course not. They exceeded them. They did things you didn’t know were possible until you saw them done.

We sign on with the creator to take the journey they have laid out, without knowing where we are going or how. This allows for surprise and delight and inspiration and education all the things that we hope for. It also might lead to boredom and disappointment and frustration, even anger, but those feelings are educational as well. Asking for exactly what we want and then getting it teaches us nothing.

*

In the most exciting acts of creation, even the creators don’t even know what the outcome will be. The excitement and beauty of a basketball game is that no one knows who will win, or how. The players and coaches and referees and fans are all collaborating to create an experience for each other and the audience at home. We all find the destination together. This is true for art as well as sports. Many, many great writers have spoken about needing to disconnect from preconceptions altogether, letting their imagination go wherever it goes. In Paul Zollo’s book Songwriters on Songwriting, the great Paul Simon said the following about how he writes lyrics:

“You want your mind to wander, to pick up words and phrases and fool around with them and drop them. As soon as your mind knows that it’s on and it’s supposed to produce some lines, either it doesn’t or it produces things that are very predictable. And that’s why I say I’m not interested in writing something that I thought about. I’m interested in discovering where my mind wants to go, or what object it wants to pick up… You’ll find out much more about what you’re thinking that way than you will if you’re determined to say something. What you’re determined to say is filled with all your rationalizations and your defenses and all of that. What you want to say to the world as opposed to what you’re thinking. And as a lyricist, my job is to find out what it is that I’m thinking. Even if it’s something that I don’t want to be thinking.”

Good writing isn’t built to a blueprint, but rather grown, like a living thing. As an author, you are responsible for supporting that growth, but the process is a dialogue between you and the work. The work will tell you when you have made the right choice, but only if you take the time to listen — putting something down, reading it back to yourself in the context of what surrounds it, and carefully judging whether it is really part of the form the things wants to take. This is the process that allows your own writing to educate you. Your subconscious can go places and come up with ideas that your conscious mind would never have found.

And so, as useless as AI-generated writing is for the reader, it is equally useless for whoever is causing the writing to be generated. The words will not have come out of your brain, so they will not have tickled your pathways of thought on their way to the page, and so you will never get any real benefit from them. It’s like hiring someone to go to the gym for you. Yes, you will have caused the dumbbells to go up and down the requisite number of times, but so what?

*

There is a fourth group using AI regularly, of course, and it’s most of us. We don’t want to be great writers, we just want to get our work done and go home. If we can find an easier, faster way to accomplish some of these tasks, why wouldn’t we? I get it. I honestly do. But please, consider how much time you are actually saving by using these tools — I am sure there are people spending more time composing prompts for ChatGPT than they would have needed to just write whatever it was in the first place — and also please consider what you might be losing when you give over your voice to a piece of software and let it put words in your mouth. There must be a purpose to that thing you have to write, and wouldn’t it be better served by applying your actual thoughts and words to it? Your mind has value. Your ideas have value. Don’t give yourself over to just transporting generic digital dribblings from one place to another.

And for those of us who are really invested in putting ourselves or our businesses out into the world by means of the things we write – our blogs, our social posts, our podcasts, our websites – I implore you: forget about “generating content” and start writing. Start with a blank page and an idea. Build sentences carefully. Say each word aloud and test its weight and rhythm. Construct paragraphs with function and balance. Throw things away often. Rewrite and rewrite and rewrite until every sentence flows one to the next and every idea supports the idea that follows. Write a page that no one else could have written. Then write another one. Your readers are waiting.