How can writers use (or productively misuse) the collaborative potential of AI text generators? We are approaching this question as creative writers and via sharing the process of our project, The Shadow of Nothingness (SoN). SoN is a work of fiction written with the assistance of OpenAI’s GPT-2. The work explores the existential void looming at the heart of consumer culture. By subverting the utilitarian text form of the consumer review and turning it into a literary form, we created a narrative universe of interconnected reviews that examines the gap between what can be desired and what can be bought.
Influenced by our understanding of 14th-century German Mystic Meister Eckhart’s notion of Umbra Nihili; our reading of Sartre’s Being And Nothingness; and our love of literary Yelp reviews, we began by writing a stock of ca. 50 reviews from the perspectives of an array of fictional characters. We then ran the manuscript through GPT-2. This process generated a new batch of reviews, some offering astonishing remixes that suggested unusual expansions of our initial idea. For example, the AI proposed the presence of a mysterious child; we accepted and consequently deepened a central plotline to accommodate the new character.
The AI’s suggestions also nudged us to wrestle with the thematic essence of our creation. Because we read the AI’s offerings with the faith one might have in an oracle, we noticed how subconscious aspects of our work surfaced: “I am dominated by a vague fear of the absurd” (courtesy of GPT-2) is a spot-on distillation of our authorial queasiness. In that way, the AI performed inadvertently like a text-analysis tool.
When we began selecting sections of the AI’s output, we shifted from a generative mode into a curatorial one. Our main task became integrating the AI’s contributions, paying particular attention to what communication theorist Walter Fisher termed the “coherence” (the internal logic) and “fidelity” (ties to recognizable values and experiences of readers) of the overall work. Meaning: we had to shape the content to meet the fundamental logic of the world we had conceived. The AI’s verbiage also needed careful line-editing to address any linguistic insensitivities while retaining instances of unlikely poetry.
The collaboration with GPT-2 remained exciting because it played on our compulsion to make sense of (and weave into our context) what the AI had offered and because we approached its output with the aim of maintaining a stable narrative arc. Precisely by offering outlandish and “faulty” suggestions, the AI generator pushed us beyond the preconceived limits of our idea and off the well-trodden path along Gustav Freytag’s story pyramid. We see the text generators as tools for finding inspiring suggestions, not creative solutions.
“Whatever you now find weird, ugly, uncomfortable and nasty about a new medium will surely become its signature. CD distortion, the jitteriness of digital video, the crap sound of 8-bit - all of these will be cherished and emulated as soon as they can be avoided.― Brian Eno, A Year With Swollen Appendices