AI is Closer to Replacing Editors Than Writers
Ted Chiang, Barbra Kingsolver, and Me Espresso
Two weeks ago, The New Yorker published an article by acclaimed science fiction writer Ted Chiang, who made a persuasive case for why AI, even beyond its technical limitations, can’t and won’t ever make art. Essentially the conclusion he reaches is that AI can’t create art because it has no agency or meaning behind its output:
“It is very easy to get ChatGPT to emit a series of words such as ‘I am happy to see you.’ There are many things we don’t understand about how large language models work, but one thing we can be sure of is that ChatGPT is not happy to see you. A dog can communicate that it is happy to see you, and so can a prelinguistic child, even though both lack the capability to use words. ChatGPT feels nothing and desires nothing, and this lack of intention is why ChatGPT is not actually using language. What makes the words ‘I’m happy to see you’ a linguistic utterance is not that the sequence of text tokens that it is made up of are well formed; what makes it a linguistic utterance is the intention to communicate something.”
But something earlier on in Chiang’s article caught my attention, he calls art “something that results from making a lot of choices.” He goes on to say, “This might be easiest to explain if we use fiction writing as an example. When you are writing fiction, you are—consciously or unconsciously—making a choice about almost every word you type; to oversimplify, we can imagine that a ten-thousand-word short story requires something on the order of ten thousand choices.”
Chiang’s way of describing writing fiction articulates a truth I’ve thought about often since becoming a book editor. Part of the endless fascination of editing books and the awe of reading them is the realization of the mind-numbing number of choices that go into a novel or full-length book. Chiang, by his own omission, is indeed oversimplifying when he says that 10,000 words represent 10,000 choices. This 1-for-1 ratio, doesn’t really capture the enormity of how many choices a writer makes when writing. That’s because a word is one choice, but once words are combined they start to form relationships and thus one choice becomes multiple choices, then a few words act as a frame for next words, which will lead to a perspective on the next sentence, and on and on and on and on. The choices are exponential and fractured, not linear. Writing is like making a film, in which every new sentence constitutes a new camera angle and shot—to get continuity and clarity out of this is nothing short of a miracle.
Let’s take a random sentence from a book I’m currently reading, the Pulitzer-Prize-winner Demon Copperhead by Barbara Kingsolver. Here’s what opens chapter 3:
“From the day Murrell Stone walked up our steps with his Davidson boot chain jingling, Mom was like, He’s a good man. He likes you, and you like him. I had my instructions.”
These three simples(ish) sentences constitute an unfathomable number of choices. “From the day” is a common type of opening to a sentence or chapter, it’s a marker of time, a frame for the reader to say “we are here in time and space.” And Kingsolver choosing her marker of time as “from the day” is one of thousands if not millions of options she picked from. Kingsolver could have chosen to open with the day, month, and year that Murrell Stone walked up their steps, or just the day, or just the month, or just the year. Each of these four choices would carry its own associations and framing for what follows, saying all three would emphasize specificity, while saying just the year might emphasize the historical nature of the scene (pick any specific year and it carries different weight with it, “2001” a different kind than “2016” than “1977”).
Consider that Kingsolver could have chosen to tell us what season it was at the start of the chapter. Summer feels like an appropriate choice for some reason. So, for fun, let’s add that: “From the summer day Murrell Stone walked up our steps…” We’ve added one word to our frame and it has entirely changed the words that follow—now it is presumably hot and we’ve added a whole dimension to the scene in which we first meet Murrell that may or may not be desirable to how we perceive Murrell (is Murrell dressed appropriately for the weather?), who is walking up the steps. Add an adjective to this opening — “From the awful summer day Murrell Stone walked up our steps” — and we’re talking about a whole other set of chain reactions. Some writers and editors could (and do) spend hours and years driving themselves mad with these choices. And let’s not even talk about the fact that you can analyze any of these choices in terms of how someone else will interpret and associate your choice (is a “jingling” boot chain menacing or silly?). A single word is thus more than just one choice, it’s a choice from a million options with an infinite number of consequences.
There are so many choices and writing a novel is such a difficult thing to do that I doubt most writers go through the analytical process above all the time while they are writing a sentence— it would simply be too crushing and, ultimately, counterproductive. It’s why no great idea or writing guide — or yes, AI — will ever be sufficient enough to be an easy shortcut to writing a good novel. But what is perhaps the most unexplainable and difficult part to replicate for machines or for beginners is that some of the most interesting choices in art tend to appear spontaneous and illogical. There’s a fun article about Sabrina Carpenter’s mega pop hit “Me Espresso”, which unpacks the incorrect grammar and effectiveness of the song’s catchy, “nonsense” chorus: “Say you can’t sleep, baby, I know / That’s that me espresso.” Great artists who are very skilled at their craft often learn the rules so well that they are able to break them in a way that is pleasing and unexpected. And, more often than not, they are unable explain why they made the choice they did and are unable to replicate it the next time around. Music is a prime example of this and probably one of art’s most mercurial mediums. Such that even if a computer exhaustively learns the basics, it will struggle to find “that’s that me espresso”— those odd wrinkles that defy the rules and yet somehow work.
But Can AI Replace Editors?
If writing a novel is about making millions of conscious and intuitive choices, editing is about unpacking and analyzing those choices. It’s about doing that annoying process with Barbra Kingsolver’s opening sentence, and doing it on both the small and large levels Chiang talks about in his essay (i.e. how that small choice of opening affects and meshes with the big choices Kingsolver makes for the book) over and over again. A novel is not created out of logic or pure rules and that’s why it isn’t something a computer does very well, but a novel or piece of fiction does start to develop its own logic over time. An editor’s job is to figure out the idiosyncratic, unique logic a writer has created. An editor is like the producer who had to put music to Carpenter’s “Me Espresso,” — it’s not the producer’s job to judge or fix Carpenter’s lyrics but to figure out what the intention is behind them and how to make them work best as a song.
Editing and writing are distant cousins. It is often said that writing is rewriting, aka editing, and there are many times when an editor finds themselves writing or changing words in a manuscript. This often fools writers into thinking that they don’t need editors and editors into the idea that they are writers. But it should be obvious that in each instance there are distinct, non-transferable skills in both writing and editing. For writing it is that intuition and creation of new, weird logic that separates it from editing— it’s ultimately why it’s not possible to edit your way to a masterpiece or use Chat GPT to write a bad book and simply edit it into something good. On the flip side, writers often can’t explain their process or see the consequences of the weirdest choices they’re making. Editing is about helping writers make sense of their strange rules they create that the writer likely wouldn’t often break in the first place if they started with the analytical approach of an editor.
If one of these two skills is more vulnerable to automation, it is editing not writing. It’s easier to see how a machine is suited to decoding logic like an editor, after all, than it is to creating art. But because writing and editing are cousins, I suspect that an AI editor will likewise struggle. If good writers are out there creating new, idiosyncratic logic, a computer with its ordered thinking is going to have a hard time (without a human taking a lot of special attention to program it specifically) in doing what a good editor does, exercising that subjective muscle and saying “that’s that me espresso” not only makes its own beautiful, perfect sense, but it’s wrongness will work better than the safe choices that have come before it.