To be fair—to myself—I had published “Ghosts” partly as a provocation. I wanted to bring attention to a promise I imagined Al companies might make in the future—that they could help us tell our own stories—and then demonstrate that promise getting uncannily close to coming true while ultimately being broken. At the same time, I wanted to demonstrate my own complicity in taking the bait in the first place. But I could also see an argument that, whatever my intent, the exercise of “Ghosts” had been fundamentally corrupt—that my desire to make a point mattered little, set against my collusion with the harmful practices required for “Ghosts” to exist. A small part of me even hoped critics—at least one critic—would censure me for it.

Instead, the opposite happened. One writer cited my piece in a hot take with the headline “Rather Than Fear AI, Writers Should Learn to Collaborate with It,” using it as evidence that people and AI “will learn to coexist, with humans using their deep learning-based counterparts to enhance and improve their prose.” Teachers assigned “Ghosts” in writing classes, requiring students to produce their own Al collaborations. A beloved indie filmmaker emailed me looking for advice on using Al in a screenplay. A venture capitalist invited me to help judge an Al writing contest for established authors and journalists. Knowing that the proliferation of Al language models would depend on convincing people that the benefits were worth the costs, I was starting to feel that I’d contributed to making that case.