Digital Cultures Research node: Discourse Machines – Power and knowledge in the age of large language models

Simon Lindgren, Professor of sociology and director of DIGSUM, Umeå University, 

## Discourse Machines: Power and knowledge in the age of large language models

Large language models tend to be discussed in terms of intelligence, accuracy, or bias. In this talk, which draws on my book Discourse Machines (Polity), I argue that they are better understood as discourse machines, that is, as systems that intervene in what Foucault called the discursive field. Trained to predict what is probable, and further shaped by alignment work and corporate design choices, these models tend to reproduce dominant discourses while presenting them with the fluency of settled knowledge. The central issue with ‘bias’, is therefore discursive power, rather than technical error. I end by outlining some ways of speaking back.