Bidirectional and Asymmetric Relationship-Level Configuration in Generative AI: A User-Defined Framework for Customizable Human–AI Relationships
Year: 2026
Type: preprint
Abstract: Generative artificial intelligence (AI) systems are increasingly used in multiple relational contexts, including friendship, companionship, professional collaboration, mentorship, personal assistance, and emotional support. Although existing AI companion applications allow users to select predefined relationship categories, and previous research has investigated human–AI attachment, intimacy, relational development, and social interaction, a systematic mechanism for independently configuring relationship levels in both directions remains insufficiently developed. In particular, existing approaches do not adequately distinguish between the relationship classification assigned by a user to an AI system and the relationship configuration that governs how the AI system interacts with that user. This paper proposes the Bidirectional and Asymmetric Relationship-Level Configuration Framework (BARLC), a conceptual and technical framework for customizable relationship management between users and generative AI systems. The framework introduces two independently configurable relational directions: user-to-AI relationship configuration and AI-to-user interaction relationship configuration. These configurations may be symmetrical or asymmetrical, independently defined, selectively enabled, or restricted to a single direction. The proposed framework further supports customizable relationship categories, multidimensional relationship attributes, hierarchical and non-hierarchical classification, dynamic relationship transitions, context-sensitive interaction adaptation, relationship-specific permissions, and persistent relationship configuration across conversations. Importantly, the framework distinguishes a system's operational relationship configuration from claims that an AI possesses subjective emotions, genuine attachment, or human-like relational intentions. By treating relationship classification as an explicit, configurable interaction mechanism rather than merely a predefined chatbot persona, BARLC provides a foundation for more transparent, controllable, and context-aware human–AI interaction. The paper establishes the conceptual architecture, formal representation, operational principles, potential applications, evaluation methodology, and ethical requirements of the proposed approach.
Author: Zhang, Jincheng
Language: English
Subfield: Social Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7233837502
