Techart Conference

Steven Shaviro (Wayne State University, USA) 

■ Abstract

Large language models (LLMs) such as ChatGPT, Bing, and Bard, have made a strong impression recently due to their fluency with language and the large range of topics that they seem to be able to discuss. However, they are frequently plagued by the problem of hallucinations: they tend to invent supposed "facts" out of whole cloth, and more generally to confabulate, or to produce extended discourses that are more or less continuous and consistent, but that make no actual sense. These AIs pass the Turing Test, but they nonetheless seem devoid of true interiority or understanding. Though big tech companies like Microsoft and Google are trying to eliminate the problem of AI hallucination, I argue that it is this capacity that most strongly suggests that LLMs already display a capacity to signal across the void that, according to the speculative realist philosopher Graham Harman, separates entities from one another.

■ Bio

Steven Shaviro has taught at the University of Washington in Seattle, and at Wayne State University in Detroit. He is the author of books about science fiction, movies and music videos, and the philosophy of Alfred North Whitehead.