In a Quasi-Social Relationship With ChatGPT. An Autoethnography on Engaging With Prompt-Engineered LLM Personas

Abstract

As conversational AI like ChatGPT becomes more sophisticated, understanding emerging quasi-social relationships with it is crucial. Through analytical autoethnography, we explore the nuances of these relationships with two autobiographically designed ChatGPT personas to augment the social needs of the first author: the Endless Enthusiast (always responding positively and encouragingly) and the Socratic Tutor (asking questions to stimulate critical thinking). After six weeks of interaction, we find that for a successful relationship, the non-human counterpart must be authentic about its machine nature and limitations. Using deception to appear more human-like makes the relationship fail. We thus suggest designing machine relationships as complementary to human-human relationships. For authentic interactions, humans should be in control, with the machine authentically assuming a role that "naturally" fits machines. Here, the unique qualities of machines in social interactions offer promising starting points for designing such roles.