Now that I have a clear and exciting trajectory for my creative component, I’m beginning to worry about defining my critical approach. This will of course be responsive to the particular questions the less predictable parts of my research—namely, my outputs—raise, but defining a structure, style, and audience for the paper will be important as I decide where to focus my research and documentation of process, even if the exact critical concerns shift somewhat during the research process.
I am certainly not lacking for options. The more I explain my project in various contexts, the more I find it could generate relevant, timely, and novel discussions across a number of disciplines. I don’t want to limit my considerations to any one of these, but I do still need to produce a work that is legible to an audience beyond myself—it isn’t reasonable to expect anyone else to share exactly my interests, background, and vocabulary across literature, arts, computer science, and philosophy. And in 7,000 words, I can’t take the time to explain each of these contributing components thoroughly enough to allow for audiences who may be familiar with only some or none of them.
In this post, I’ll lay out some of the interesting angles on and implications of my project. Ideally my paper can integrate multiple of these, but it’s doubtful that I could even refer to them all, let alone explore them each with any sort of rigor, so I’m hoping to find an emergent theme(s) that will allow me to connect as many of them as possible in a transdisciplinary yet cohesive way, with a clear audience in mind.
First, the philosophical concerns. What is identity, and how does data reflect or fail to reflect that? Can language models be useful for self-knowledge, as might be supported by philosophies of habit and the notion of model training as encoding/emulating habits of speech? Alternately (or perhaps concurrently?), can the gap between the model and our personal sense of identity reveal something about the model itself as an Other, supported by new materialisms and more-than-human ontologies?
This tiptoes toward data ethics and perhaps even the coloniality of data as I could also ask questions about the reductive nature of digital identities and movements like the Quantified Self.
It also opens up possibilities within human-computer interaction and digital media studies. How can/should generative models be presented? As voices of their own, or a kind of remix or chorale of their inputs? What is at play in the rhetorical situation where you are the reader/user and the speaker, in some sense?
Which leads into the already hotly debated questions of authorship and the role of AI, as seen through a literary studies lens. Can we understand the (un)reliability of generative writing as a kind of fictionalization, perhaps using vocabulary like Margot Livesy’s description of antifiction in discussing the pseudo-autobiographical writing of Marcel Proust?
In terms of creative writing studies and practice, it may also be worth considering how writers already do or don’t treat source texts as data or voice as something curated. If this method seems useful as a brainstorming, co-creativity, or reflective tool, a strongly practice-driven paper directed toward writers could make sense.
Lastly, this work aligns with current interests within natural language processing concerning small data sets, few-shot learning, and model personalization. A more technical paper explaining the method, its limitations and successes, and where it could fit into the future of LMs would also be valuable research in a field of interest to me.
I know I don’t want this to be a purely philosophical project; I need to place some emphasis on the real-world impact of either my model or what it reveals about LLMs/data/digital writing. But I also can’t leave the theoretical and humanistic concerns behind in favor of a purely technical paper. So some other focusing question that allows me to move between at least two or three of these disciplines would be ideal.
