Experiential Literature? Comparing the Work of AI and Human Authors

Abstract

Using artificial intelligence (AI)-authored texts as a baseline for reading literary originals can help us discern what is new about today's literature, rather than relying on the AI itself to embody that newness. GPT-3 is a language model that uses deep learning to produce human-like text. Its writing is (in)credible at first sight, but, like dreams, quickly becomes boring, nonsensical, or both. Engineers suggest this shortcoming indicates a complexity issue, but it also reveals an aspect of literary innovation: how stylistic tendencies are extended to disrupt normative reading habits in ways that are analogous to the disruptive experience our present and emergent reality. There is a dark irony to GPT-3's inability to write coherently into the future: large language models are exploitative and wasteful technologies accessible only to multi-million-pound corporations. The commercial ambitions of the tool are evident in a curiously banal kind of writing, entirely symptomatic of the corporate-engineered sense of normalcy that obscures successive, irreversible crises as we sleep walk through the glitch era. Contrary to this, experimental literary practices can provoke critical-sensory engagement with the difficulties of our time. I propose that GPT-3 can be a measure of what effective literary difficulty is. I test this using two recent works,The Employees, a novel by Olga Ravn, and the 'Septology' series of novels by Jon Fosse. I contrast their 'experiential literature' with blankly convincing machine-authored versions of their work.