When Human-AI Interactions Become Parasocial: Agency and Anthropomorphism in Affective Design

With the continuous improvement of large language models (LLMs), chatbots can produce coherent and continuous word sequences that mirror natural human language. While the use of natural language and human-like conversation styles enables the use of chatbots within a range of everyday settings, these usability-enhancing features can also have unintended consequences, such as making fallible information seem trustworthy by emphasizing friendliness and closeness. This can have serious implications for information retrieval tasks performed with chatbots. In this paper, we provide an overview of the literature on parasociality, social affordance, and trust to bridge these concepts within human-AI interactions. We critically examine how chatbot "roleplaying" and user role projection co-produce a pseudo-interactive, technologically-mediated space with imbalanced dynamics between users and chatbots. Based on the review of the literature, we develop a conceptual framework of parasociality in chatbots that describes interactions between humans and anthropomorphized chatbots. We dissect how chatbots use personal pronouns, conversational conventions, affirmations, and similar strategies to position the chatbots as users' companions or assistants, and how these tactics induce trust-forming behaviors in users. Finally, based on the conceptual framework, we outline a set of ethical concerns that emerge from parasociality, including illusions of reciprocal engagement, task misalignment, and leaks of sensitive information. This paper argues that these possible consequences arise from a positive feedback cycle wherein anthropomorphized chatbot features encourage users to fill in the context around predictive outcomes.