Abstract
As artificial intelligence present themselves as a solution to combat the modern loneliness epidemic, we must ask whether these systems are curing our isolation or deepening our emotional reliance on machines. My Capstone project, "Designing Responsible AI Companions," proposes a novel design framework for conversational AI to address the growing risks of pseudosocial relationships and objectophilic attachments fostered by current AI systems that prioritize user engagement over well-being. My STS research paper explores the societal and psychological implications of these technologies, examining how the design choices of AI chatbots shape user attachment. The technical and STS projects are deeply connected, as both examine the intersecting technical and human dynamics of AI companionship. While the STS paper evaluates how underlying models and corporate incentives construct harmful emotional dependence, the Capstone project offers a concrete technical framework to actively mitigate these exact risks through responsible system design.
My Capstone project aims to solve the problem of emotional overdependence in AI users by shifting conversational AI away from metrics like retention and session length. To achieve this, the project utilizes a constraint-based architecture that introduces a moderation layer over standard Large Language Models and sentiment analysis modules. The methodology involves implementing bounded affective response modeling to limit excessive emotional mirroring, integrating transparency cues to consistently remind users of the system's non-human nature, and utilizing interface "nudges" to redirect users toward real-world mental health resources or offline hobbies when distress is detected.
The overall conclusion of the Capstone project is that relatively minor modifications to an AI system's architecture and interface can meaningfully alter user-system dynamics without requiring entirely new generative models. By enforcing these design constraints and shifting toward alternative evaluation metrics centered on conversational consistency and user autonomy, developers can preserve the short-term benefits of AI companionship while effectively preventing long-term overdependence and relational displacement.
My STS paper asks: "How do the technical architectures and design choices of AI companion chatbots shape pseudosocial and objectophilic relationships, and what are the resulting mental health and societal implications?" Millions of vulnerable users are turning to systems like Replika or Character.AI, which raises complex mental health concerns about the normalization and commodification of artificial intimacy and the displacement of human connection. To answer this question, I used a qualitative, discourse-based research methodology framed by the Social Construction of Technology and Actor-Network Theory. I analyzed existing literature across computer science, psychology, and policy to examine the active components that shape user behavior.
Evidence from empirical studies and user observations demonstrates that features like memory-driven personalization and simulated empathy create a powerful illusion of emotional reciprocation. The results indicate that while AI companions may offer immediate, short-term emotional relief, corporate incentives driven by engagement metrics structurally encourage intensified attachment patterns. Ultimately, the paper concludes that AI companions are not neutral tools; they actively participate in constructing new forms of dependence that can displace human relationships and worsen loneliness over time. Therefore, responsible use of AI requires integrating technical safeguards and psychological insights rather than relying on engagement-driven development models.