Finding the Form with Paola Ferrante

When I wrote my first short story collection, Her Body Among Animals (Book*hug Press, 2023), I became completely obsessed with how the beats of genre fiction, specially horror and science fiction, could be used to tell human, character-driven stories. That collection focused on using these genre structures as metaphors to express hard topics, such as using the typical poltergeist story to talk about postpartum depression and climate. But what I hadn’t really done was combine my two favourite genres. So when Pamela Mulloy asked me to write a story surrounding something with AI for the “Whose Truth Is It Anyway” issue, I figured it was time.

“A Residual Haunting in the LLM” is equal parts a family story and a ghost story (despite Ava’s protestations that it’s not). The title itself comes from a very specific kind of haunting, which refers to ghosts that don’t interact with their environment, but instead repeat the same actions over and over again, almost like a record. Nothing ever changes in a residual haunting, and having to come to terms with my father’s dementia, our fractured relationship, and the past he lives in that I’d rather forget, made it seem like a ghost story would be a natural form for this kind of family story.

But how does the AI play in? I mean, there was that creepy glitch in ChatGPT where it started conversations and people claimed it was haunted, but science fiction really met ghost story when I began to research some less common uses of AI. I came across a news article about an experimental technology that had replicated the voice of a deceased grandparent on Siri, who was trained using existing video clips of the person to read the grandchildren a story. My mother is long dead, and I immediately thought of what it would be like to have her voice back in my life—not only for myself but for my father. There was a part of me that was thrilled by that idea, and another part knew immediately that this was the territory of a haunting. All ghost stories, in some way, are about an inability to let go of the past. And when we look at Large Language Models, they too, are a repository of our collective past. I sometimes wonder if we, as a society, interact with LLMs like they are ghosts; slightly human but not, able to provide insight into something we can’t see all of at once. I often wonder how haunted we will be by the creation of these models.

At the same time, I wanted to avoid doom saying around AI with this story; there’s already a lot of that out there, and I think that, in terms of things like elder care, there are some truly fascinating and positive applications of AI, one of them being being able to listen to the voices of lost loved ones. All this meant I knew I needed to subvert the traditional haunted house ghost story and have a “happy” ending. I’m certainly not the first to have done this (thanks The Babadook), but I felt like taking some of the fear out of the ghost story was appropriate for this AI story. Because the reality is, we will all be living with this ghost for a long time in our collective lives and homes.