Dittos: Mimetic, Reciprocal Agents in AI Mediated Communication
Organized by ACM
Recent advances in generative AI have enabled agents that can represent specific people in social interactions when they are unavailable. These agents—often described as digital twins—can look and sound like an individual and participate in conversations on their behalf. In this position paper, we focus on a particular class of such systems that we refer to as Dittos: mimetic, reciprocal AI agents that not only interact with others as a proxy for a person but also report back what occurred so that human relationships can continue to develop over time. Drawing on our experiences designing, deploying, and studying Dittos, we argue that this class of systems surfaces a distinct set of challenges to core assumptions in AI mediated communication research. These challenges concern how presence is evoked through representation, how trust is established when AI speaks in someone’s voice, how people remain meaningfully informed about Ditto mediated interactions, and how ethical risks can be anticipated and mitigated. We position these challenges as central to understanding a new phase of AI mediated communication, where systems increasingly act on behalf of people rather than merely between them.