Abstract
In this paper, we examine the potential effects of relationships with Large Language Model (LLM)-based digital doppelgängers (DDs) on users' values, concerns, and interests, that is, on their practical identity. DDs are artificially intelligent conversational agents trained on individuals' data to replicate their speech patterns, mannerisms, and personality traits. We start by showing that practical identity is largely defined by the relationships we find ourselves in or cultivate. Next, we discuss how DDs work and distinguish between task-specific and relational DDs, arguing that the latter are specifically created to take on the role of relational partners by imitating other people we care about. We then consider how relationships with DDs might shape individuals' practical identities, before concluding with some general thoughts regarding what we should be paying attention to before developing and deploying DDs.