Abstract

This chapter revisits the notion of mindshaping in light of recent developments in artificial intelligence, with a particular focus on large language models (LLMs). Following Zawidzki, mindshaping is understood as a set of social and developmental practices that align agents with shared models of agency, rendering their behaviour intelligible within a folk-psychological framework. We extend this account by introducing artificial mindshaping: the suite of training procedures, interactional dynamics, and design constraints that shape the behavioural profiles of LLMs so that their actions can be interpreted in terms of familiar mentalistic categories, such as belief, desire, intention, and reason. Drawing on Shanahan, McDonell, and Reynolds (2023), we organise these mechanisms along a distinction between the simulator (the underlying language model and its trained dispositions) and the simulacrum or virtual personality (the persona the simulator instantiates in a given interaction). We characterise the property these mechanisms produce as interpretive alignment—the extent to which an artificial system’s behaviour can be subsumed under folk-psychological kinds, independently of whether its internal mechanisms mirror those categories. While interpretive alignment connects naturally with contemporary work in explainable AI, it also bears on issues of mentality. Specifically, interpretive alignment establishes the conditions for mindedness as per Dennett's intentional stance. In this sense, artificial mindshaping is a form of mindmaking—not merely a way of making minds explainable but also a means by which artificial forms of mindedness are brought into being. We close by noting that mindshaping is a relational notion that runs in both directions, and that the question of how human minds are in turn shaped by these artificial systems is becoming one of the most pressing raised by their growing integration into human cognitive life.