The pragmatics of trusting (artificial) others

Trust is a precondition for cooperation, for social learning, and for the everyday willingness to rely on what others tell us and do (Mayer, Davis, and Schoorman 1995;Sperber et al. 2010).It is also a pragmatic achievement.We extend or withhold trust on the basis of what interlocutors communicate and how they communicate it, inferring competence, benevolence, and integrity from cues embedded in situated interaction, and coordinating our understanding as a joint activity rather than receiving it ready-made (Grice 1975;Clark 1996).Increasingly, however, the others we are asked to trust are artificial: chatbots, voice assistants, recommender systems, autonomous vehicles, and social robots.A system becomes available as an interactional "other" when it is presented or treated as a source, respondent, adviser, partner, or accountable participant, whether through language, voice, embodiment, sequential responsiveness, or institutional positioning, and hence as an interactional party.At the same time, the communicative agency of these systems does not map straightforwardly onto human forms of knowledge, intention, experience, and accountability.These two features make artificial others a distinctive case for pragmatics.First, they recruit the social rules and machinery of human social cognition (Nass and Moon 2000) and engage interactants in a layered pretense, inviting interactants to understand these systems as depictions of social actors staged over a mechanical base (Clark and Fischer 2023).Second, because these systems' competence and their perceived trustworthiness can diverge sharply, the central problem is not to maximize trust but to calibrate it: to align perceived with actual trustworthiness so that users neither over-rely on a fallible system nor dismiss a capable one (Lee and See 2004;Schlicker et al. 2025).Artificial others therefore pose, with new urgency, an old set of questions: What makes an interlocutor trustworthy, and how is trustworthiness signalled and read?How does trust discourse differ across people, technologies, and domains?And how can trust be calibrated so that it matches what the respective other can actually do?