Some Thoughts on Algorithmic Writing

Do not go gentle into that Google Sheet

I’ve been thinking lately about what it’s like to write under a very precise, almost mathematical set of rules—in a word: algorithms.

A good example of this is the villanelle: a nineteen-line poetic form with exact repeating lines following an A1 b A2 / a b A1 / a b A2 / a b A1 / a b A2 / a b A1 A2 sequence. In fact, it occurred to me the other day that the villanelle structure is so formulaic you can literally construct one with the aid of Excel formulas.

I created a simple tool to do this here:

Of course, while some forms of writing really are bound by villanelle-like precision (Knock knock! Who’s there?), most are far less prescriptive. Still, there are usually some constraints operating in the background (essays have a thesis, novels have characters in conflict, etc.) What seems unique in the current moment is that computers are learning to recognize these more abstract kinds of writing constraints. With the right prompting and training, AIs will soon be able to generate plot twists and punchlines as easily as my template fills in cell B13. Indeed, AIs already can pull this off quite convincingly.

We might worry that algorithmic writing will rob creative work of its joy and meaning—but the simple example of the “auto villanelle” suggests otherwise to me; there are a few reasons for this.

First, the algorithms here are doing the least creative part of the process. If I were in the habit of writing lots and lots of villanelles, these formulas would significantly improve my output—both by saving me time and reducing errors/inconsistency. Even for looser genres, one of the major benefits of algorithmic writing will surely be to reduce the more tedious, mechanistic parts of the process.

Second, there are always choices to make in writing—the algorithm just restricts which choices. (If lines 6, 12, and 18 must match line 1, well, someone still has to write line 1!) Often, these constrained choices are more striking or delightful than what the writer might have produced with total freedom.¹ In the words of Joseph Brodsky, “The rhyme is smarter than the poet.”

Third, there’s no reason writers must blindly accept the dictates of an algorithm; the algorithm is merely a means to an artistic end. With practice, we can learn to thoughtfully override whatever it is the “rules” suggest a rhyme or punchline or narrative beat ought to be. Indeed, the more established a form is, the more opportunity there is to write against it to positive effect; when a great piece of writing deviates from the pattern it had seemed to invoke, we feel the thrill of simultaneous recognition and surprise. Sometimes the effect of this is amusing, like this limerick of unknown origin:

There was a young man from Japan
Whose limericks never would scan.
          And when they asked why,
          He said “I do try!
But when I get to the last line I try to fit in as many words as I can.”

Sometimes it is energizing, like the slow, sad song that resolves unexpectedly to a major chord.

And sometimes it feels wistful, like the fleeting glimpse of a path not taken.

1. For readers, the pleasure frequently lies in knowing what constraints the writer is working under; a great rhyme is great in part because it plays within the specified rules of the game. That said, there are certainly also artists who work under invisible constraints, like those of the Oulipo group; and following a set procedure can surely be generative for a creator whether or not the audience knows about it.