Hello all! This past fall I spent a great deal of time working on a project for an English class called Medieval Proof in which I set out to build a text generating recurrent neural network with the goal of generating “literature.” It so happened that as I was doing this ChatGPT exploded in popularity and has begun to raise all sorts of ethical questions. As a physics/math/CS nerd just the challenge alone was motivation enough, but as someone for whom literature is a mainstay of life, this particular problem–that is, can a machine ever write literature–has a special importance because I believe it allows us to investigate the nature of literature. Moreover, it is crucial to explore the implications of the technology that I along with others doing similar work am helping to bring into the world.
I have attached the records of my two readerly experiments which document the development of my model Thinking Parrot. The first one includes a description–aimed to be complete but not too technically overbearing–of how the code actually works as well as an analysis of some of the original output. The second, which is significantly longer, explores more specifically the question of the relationship of the machine to literature and wonders if we can ultimately define literature on a character (as in ‘a’, ‘b’, etc) level, and thus learn a statistical pattern of their arrangements from which we reconstruct art; the second also attempts to wrestle with these abounding ethical questions.
I am currently working on a selection of poems that is a conversation between me and Thinking Parrot that I intend to submit to the literary magazine Rattle among others, which I will post when complete.