On the Discourse Ethics of Large Language Models: Between Communicative Agents and Linguistic Machines

Abstract

Large Language Model (LLM) based chatbots engage human users in discursive practices as interlocutors capable of vast linguistic performances. In a Habermasian discourse ethics framework, however, they do not qualify as proper communicative agents, since their behavior adapts to the input from human users but not based on the self-reflective normative re-orientation of agency required by the theory of communicative action. Nonetheless, AI chatbots affect the communicative agency of their interlocutors who engage with them by gradually reshaping their access to the linguistic resources that are essential to the self-reflective assessment of their needs and wants. In this sense, LLMs operate as linguistic machines that influence the will formation process by enacting linguistic repertoires according to ends that fall outside of the cooperative pursuit of mutual understanding among communicative agents, with a distinct lack of a self-reflective and self-interpretive component. The prospect of LLMs being deployed to pervasively articulate specific normative repertoires in public discourse is attracting increasing political attention and is at risk of establishing new forms of domination through linguistic value capture.