Oracle, Echo, or Stochastic Parrot? Who (or What) Speaks in AI-Generated Literature?

Large language models (LLMs), such as GPT-3 or LaMDA, allow users to generate texts in any voice about the topic of their choice. The metaphor of voice is crucial to understand how and why the interfaces of these LLMs work, since the phrasing and choice of words of these models evoke a sense of intentionality, perhaps even a personality. The fact that the program uses the first-person singular adds to this impression as well. In addition, the voice of LLMs is expressed in idiomatic signals, diction, and sentence structures, among other factors. Considering that the rhetorical construct of voice is an integral part of these models, it is surprising that this aspect is relatively understudied. To gain a better understanding of the textual voice of LLMs, this chapter revisits the question at the heart of Gérard Genette’s definition of the story’s narrating instance: ‘who speaks?’ Drawing on the extensive history of narratologists who have addressed this question, this chapter considers what can the narratological concept of narrative voice can teach us about LLMs and, in turn, what the phenomenon of LLMs can tell us about narrative voice. To investigate the rhetorical strategy and metaphorical richness of the concept of voice in AI-generated texts, this paper offers a close reading of ‘The Robot of the Machine is Man’ (2017), a short science fiction story in Dutch that was authored by the human author Ronald Giphart in collaboration with a language model. Because the story complicates the voice’s presumed relation to identity categories, it allows for a critical understanding of the rhetorical construct of voice in the age of AI. The chapter concludes with a discussion on different metaphors to consider the (narrative) voice of AI-generated texts.