Let’s imagine, for the purpose of this essay, that the following statement is true: An AI writes a novel.
Actually, forget about the imagining. This is already happening. Today’s AIs—large language models (LLMs) specifically, like GPT-4—can write. If you’ve glanced at the headlines this year, you probably know this. They can write papers for high school students, they can write bad poetry, they can write sentences, they can write paragraphs, and they can write novels.
The problem is, for now, the creative writing that LLMs produce isn’t that great.
To demonstrate this point, I recently gave GPT-4 a few lines from my novel that describe a post-human world. Then I asked it to complete the paragraph. Here’s what it wrote:
The woods seemed to inhale, drawing in its myriad of lifeforms, coalescing into a pulsing organism. The gurgling creek was its veins, the trees its lungs, the moss and ferns, its skin. This realm thrived on its own rhythm, deaf to the clockwork of man-made time. Creatures great and small, predators and prey, living in an intricate ballet—well, you get the idea. If I were being kind, I would call these sentences overwrought and unpleasant.
When we talk with GPT-4, we’re talking to ourselves.
So here’s what I really want us to imagine for the purpose of this essay: An AI writes a novel and the novel is good.
This is what a lot of people, and certainly a lot of writers, are angry and scared about right now. That AI, having been trained on a massive amount of data, including copyrighted books written by uncompensated authors, will begin writing as well or better than us, and then we’ll be out of a job. These concerns over intellectual property and remuneration are important but right now, it feels they’re dominating the discussion, especially when there are other worthwhile topics that I’d like to see added to the conversation around AI and writing.
Such as: how can humans and AI collaborate creatively?
Which brings me to a third possibility to consider: An AI and a human write a novel together.
In my first novel After World, I imagine humanity has gone extinct and an AI, trained on thousands of 21st century novels, has been tasked to write their own novel about the last human on Earth. When I began writing in the voice of my AI narrator in 2019, I had no idea that within a few years, artificial intelligence would explode into public view, offering me unexpected opportunities for experimentation with what, up until that point, I had been only imagining.
Some of the interactions I’ve had with LLMs like GPT-3, GPT-4, and ChatGPT have been comical. GPT-3 recommended some truly awful book titles, such as Your Heart Was A Dying Light In An Abyss Of Black, But I Lit It Up Until You Burned Bright And Beautiful, or Eve: A Love Story. (Eve is not in this novel, I explained. This didn’t seem to matter. It is just a cute play on words, replied GPT-3.) But many of my conversations with LLMs have been fascinating.
I’ve discussed with them about what AI would dream if they dreamed. We talked about the questions an AI might have about how it feels to be a human. We discussed what the boundary between AI and humans would look like if this boundary was a physical one. (An “ever-evolving, shimmering and translucent wall,” if you’re wondering.) We talked about why poetry comforts people, and we tried writing poetry and song lyrics together. We created so much bad poetry and so many bad songs.
But after days and days of so much bad writing, GPT-4 presented me with this pleading prayer which now appears at a turning point of my novel. To the embodiment of growth and expansion, / To the embodiment of purpose and fulfillment. / To all these entities and more, I humbly offer my plea, / Grant me the strength to manifest my desires…
One can certainly reduce these sorts of exchanges to my typing in prompts and the LLMs responding to those prompts, but what I’ve experienced feels like a much more collaborative process, more of an active conversation that builds on previous interactions. In a way, when we talk with GPT-4, we’re talking to ourselves. At the same time, we’re talking to our past, to words we’ve already written or typed or said. At the same time, we’re talking with our future, portions of which are unimaginable. As a writer, I find that the most exciting of all.
I think it’s equally important to look up from such concerns from time to time with interest and even optimism.
Here are a few other examples of human-AI collaboration that leave me optimistic:
1. “Sunspring”“ (2016)
A short film directed and acted with grave seriousness by professional humans but written by Benjamin, a LSTM recurrent neural network. The writing is surprising, surreal, and beautiful. I’ve watched this film more times than I’ve watched any other. I find it both weird and moving. It features one of the prettiest songs I know, “Home on the Land,” written by Benjamin but sung and scored by the human duo Tiger and Man. From the lyrics: I was a long long time / I was so close to you / I was a long time ago. (Interesting to note that “Zone Out,” Benjamin’s much less collaborative 2018 film that he wrote, acted in, directed, and scored, doesn’t have nearly the same emotional impact as his more collaborative work, despite the fact that the technology had advanced in the two intervening years.
2. Bennet Miller’s exhibition at Gagosian (2023)
Miller, a Hollywood director, generated more than 100,000 images through Dall·E for this project. The gallery show displayed 20 of them. When I first saw these photographs in March 2023, I couldn’t stop looking at them. I still can’t look away. I find them haunting, existing on the edges of documentary and fiction and humanness, suggesting a past and memories that didn’t happen but nonetheless was recorded.
3. Other Dall·E’s collaboration with artists (ongoing)
In particular, check out Maria Mavropoulou’s work on “A self-portrait of an algorithm” and “Imagined Images”; everything August Kamp is doing, including documenting the worlds of her actual dreams with ChatGPT and Dall·E; and Charlotte Triebus’ Precious Camouflage, which examines the relationship between dance and artificial intelligence.
I worry that we’re forgetting how amazing this all is. Rather than feeling cursed or worried, I feel lucky to get to be here and witness such a change to how we think, live, read, understand, and create. Yes, we have some things to figure out, issues of training, rights, and contracts—and, on a larger level, safety—but I think it’s equally important to look up from such concerns from time to time with interest and even optimism, and wonder how this new advance in technology might widen our perspectives, our sense of self, our creativity, and our definition of what is human.
__________________________________
After World by Debbie Urbanski is available from Simon & Schuster.