When no one laughs back: reciprocal subjectivity and the limits of AI companionship

Abstract

AI companions are increasingly deployed to address loneliness and to support eldercare and mental health, which makes the question of what they can and cannot offer newly urgent. Current debate evaluates them by conversational fluency, emotional responsiveness, personalization, and availability. These criteria, this paper argues, measure the surface of companionship while missing its depth. Human friendship makes available something beyond intelligent response, which may be called reciprocal subjectivity: the presence of another center of inner experience that is affected, moved, or delighted alongside one’s own. Drawing together the interactionist and phenomenological sociology of interaction, philosophies of friendship and recognition, and the recent debate over robot friendship, the paper develops two distinctions, between interaction and co-experience and between a relational mirror and a relational source. Currently deployed user-conditioned AI companions function primarily as mirrors: their apparent enthusiasm is generated within contexts assembled around the user. A friend can be a source, bringing enthusiasm and discoveries that originate in an inner life of their own and overflow into the relationship from beyond it. The recommender systems and language-model companions examined here mirror by different mechanisms, but both remain downstream of the user, which leaves them on the same side of the line. AI companionship has real value; the argument is that it belongs to a category distinct from relationships grounded in mutual inner stakes and an autonomous inner life. Implications for research, design, and regulation follow.