LARGE LANGUAGE MODELS AND BULLSHIT

Sarah A. Fisher, Ethics and Information Technology

Newly powerful large language models have burst onto the scene, with applications across a wide range of functions. We should now expect to encounter their outputs at rapidly increasing volumes and frequencies. Some commentators claim that large language models are essentially bullshitting, generating convincing output without regard for the truth. If correct, that would make them distinctively dangerous discourse participants. Bullshitters do not only undermine the norm of truthfulness (by deliberately saying false things) but the normative status of truth itself (by treating it as entirely irrelevant). So, do large language models really bullshit? I argue that they can but need not, given appropriate guardrails. As with human speakers, the propensity for a large language model to bullshit depends on its own particular make-up. My analysis sheds light on the behaviour of large language models. It also recommends a new definition of bullshitting: I argue that it should be understood as the production of verbal output without assessment for truth preservation; and I show how this definition stands up against familiar criticisms. 

Written by Sarah A. Fisher