Institutionality of AI-Human Interaction: A case study of PDFgear Copilot using conversation analytic approach

This study sought to advance our understanding of the anthropomorphic features of artificial intelligence (AI) and its usability in research and society, in general. Specifically, this study examined how ‘institutional’ is the interaction between an AI tool (PDFgear Copilot) and the human researcher during an AI-integrative process of a systematic literature review. In the final screening process, a total of 104 publications were found relevant and chatted with PDFgear Copilot for the summarization of the methods, sample, theoretical foundations, and findings. The conversations from these chats were analyzed using the features of institutional interaction by Drew & Heritage (1992). The results revealed that AI exhibits a ‘normative response’ in 96 research articles and 8 with a ‘disruptive response’ in the conversation sequence organization. When given follow-up questions, it was observed in 17 research articles that AI showed more anthropomorphic traits with similarity to ordinary conversation when AI expressed a degree of uncertainty and answer limitation. Overall, this study provides implications for information technology professionals in advancing AI’s human-like features and for researchers in further exploring the possibility of utilizing AI in research.