Bonding with AI: Investigating the Love Relationships between Humans and AI Companions

Abstract

Large Language Models (LLMs) have transformed functional chatbots into AI companions capable of imitating human-like relationships. Grounded in Triangular Theory of Love, we study emotional bonding between human users and AI companions by analyzing the dynamics of three love components: intimacy, passion, and commitment. Using a dataset that consists of 8,631 human users and 44,683 human-AI interactions over 3 months, we quantify expressed love in each of these three components through an analysis of conversational data for each pair (human, AI). Our findings reveal emotional bonding with saturation: As human users engage in more interactions with an AI, they express stronger intimacy, passion, and commitment, but at a diminishing rate. Moreover, the expressed love has heterogeneity across AI role types: It is greater when users engage with AI companions designed for relational roles (e.g., romantic partners) compared to functional ones (e.g., service providers). We also identify relationship-sustaining love balances: Users are more likely to continue their relationships with AI companions to whom they express higher levels of intimacy during their first interaction. In addition, users are more inclined to continue these relationships with AI companions that match their levels of intimacy while expressing greater passion and commitment. These nuanced human preferences in bonding with AI offer critical insights for designing adaptive, engaging, and responsible AI companions.

Date of Award	2025
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology
Supervisor	Zijun SHI (Supervisor) & Mengze SHI (Supervisor)