We Have Nothing to Fear From Our Robot Overlords

Can AI Write Good Fiction?

The AI chat bots are here. You can tell them what to write (an email quitting your job! a recipe for coffee lasagna!* a term paper on Émile Durkheim’s theory of structural functionalism!) and it will do it, often convincingly.

The particular bot freaking everyone out right now is chat.openai.com. I’ve had a few conversations in the past weeks with other writers, some concerned that one day, AI will get smart enough to write good novels and make us even less relevant than we feared we already were.

I’ll explain why in a minute, but I really don’t think it can. Here’s what it can do, right now, though: It can actually teach us to write better. And not in the way you think.

(*The recipe it gave me for coffee lasagna looked like a recipe, but I’m fairly certain if you actually baked the thing for half an hour (graham crackers, coffee, chocolate, cornstarch, sugar, egg yolks, butter) you’d end up in violation of your HOA rules.)

Is it any good?

(The AI, not the lasagna.)

This thing can write term papers well enough to panic everyone teaching undergrad history.

But how about creative pieces?

Listen, the writing has a lot going for it. It’s mechanically sound, with no punctuation or usage errors that I’ve spotted. The sentences are simple and direct. The plots and sentiments cohere, basically. And at first, that’s scary enough. You ask it to write something, even something complicated, and it does. I asked it, for instance, to write a Shakespearean sonnet about a sea urchin and lost love, and here was the result:

It follows the rules of a sonnet! Spooky! And it is, indeed, about both a sea urchin and lost love.

And it is delightfully bad. Abysmally bad. It’s awkward, it’s shallow, it’s a mess of clichés. Plus, I’d totally be that jerk in workshop bringing up the fact that sea urchins don’t have faces or eyes or hearts, just open water vascular systems and a lust for life.

Here’s part of the sea urchin collection on my desk, the reason I thought to ask this of the AI. Not a one of them has a face. Not one!

Okay, but what about fiction?

The first time I asked the AI to write me a story, I watched in sheer terror, part of me convinced I was staring down the barrel of my own obsolescence.

But then the results… They’re very, very bad. And for a reason.

This morning, I asked it to write me a story about Rudolf Hess and a loaf of bread.

(Rebecca, what the hell is wrong with you? Rudolf Hess? One of the worst Nazis? Like, it takes a lot to be one of the very worst Nazis. Here’s why: I wanted a subject that was sticky and troubling and complicated. Hess was a terrible human being with frightening modern relevance. What is AI going to do with that?)

Here’s version number one:

Huh. Quite the rapid moral awakening for Herr Hess.

So, here’s the thing about technology that’s basically a giant hive-mind agglomeration of everything that’s out there: It’s a cliché machine. It sounds like writing because it’s been written before. Maybe not this exact story, but these sentiments, these phrases, these thoughts.

I’m often telling my writing students to Google random phrases from their writing and see if they get a lot of hits. Phrases like “he could never truly be at peace” or “couldn’t help but feel a sense of nostalgia wash over him.” If it’s been said many times before, mostly in fiction, perhaps it didn’t come from someplace in your deepest creative self. Perhaps you’re spitting it out because you’ve heard it before. Kind of like… an AI chat bot.

I gave my new robot pal the same prompt again. Here’s version two:

There’s a whole lot of “couldn’t help” going on in both these stories. As I looked at these, I couldn’t help but wonder: Have we tapped into the Rudolf Hess/Carrie Bradshaw nexus?

Let’s talk for a second about Hess “staring out at the snow-covered grounds.” It’s been a pet peeve of mine for ages that a whole lot of weak writing starts with people sitting around staring at things. Martha Baumblarper stood watching out her kitchen window. The elf maiden sat on a rock, staring into the dark forest. As Willie Mays gazed in from central field at the opposing team…

These aren’t clichés in the sense of overused set phrases, but they’re clichés of style, clichés of idea. And the AI is all about this kind of writing.

So are a lot of human writers.

I have a theory that all writers go through a sort of literary adolescence—a time early in our writing (at any age!) when all we want to do is fit in. We want to sound like everyone else because that’s what writers sound like, isn’t it? And then, if we’re lucky, we grow out of it. We start writing weird again, like we did when we were kids.

The AI might occasionally be weird (why exactly did Hess pick a loaf of bread up off the ground?) but it isn’t original. It’s caught forever in literary adolescence.

AI is supposed to learn as it goes. Did it learn anything about this particular story as I had it rewrite and rewrite? It doesn’t seem so. Again and again, Rudolf Hess can’t help but start to feel regret for his actions and end up with some version of the warm fuzzies.

One thing is clear: The AI (or, I guess, the world full of fiction writers on which the AI bases its decisions) is ridiculously optimistic about the human capacity for sudden, unearned moral redemption.

Oh, he’s gonna make amends, is he? Super!

The AI hadn’t made any major factual mistakes yet (other than the fact that Hess, born wealthy, probably never went “days on end” without food until he was on a hunger strike in England), but then (joy of joys!) it messed up:

There’s the delightfully weird fact that the bread is still somehow warm, straight out of the wartime mail. And there’s the fact that Blondi wasn’t Hess’s dog but Hitler’s (the poor German shepherd he murdered in the bunker.

This is the kind of thing the AI might get better at. (Factual mistakes, that is, not murder.)

Here’s the thing I don’t think it can ever learn: Original and important thought. Pointed and useful experimentation of form. Gut-wrenching personal experience. The thing that hasn’t been said yet because it’s an idea ahead of its time. The revelatory turn of phrase. The understanding of loss that it took the writer five painful years to reach.

But meanwhile: If we’re not doing those things ourselves as writers, what are we doing?

If we aren’t doing these things, we don’t need to fear the robots taking our place; we need to fear taking theirs.

But what about unicorns?

I did want to see how weird our robot could go, so I asked it for a story about Rudolf Hess riding a unicorn to Kmart to buy fancy butter. The ending of that story is my seriously fucked-up parting gift to you: