Portrait of the AI as a Young Metaphorist.

This work builds on Zottola and Conoscenti (forthcoming), in which we explore the similarities and differences in the representation and discursive construction of AI. Our analysis compares two corpora—one authored by academics and the other generated by AI itself. Given that AI now regularly interacts with humans, this study focuses on conversations around the concept of metaphor, derived from interactions between myself and various AI systems. All the queried systems operate through human-trained Large Language Models (LLMs) to produce natural language output. As such, these interactions can be understood within the framework of dialogic communication (Wodak and Meyer, 2009) and, with regard to the linguistic-psychological dynamics of the exchange, as conversational joint actions between human and machine (Clark, 1996). In this specific study, the AI does not conceal its artificial nature, in contrast to the original premise of the Turing Test. Indeed, Jones and Bergen (forthcoming) have shown that contemporary LLMs are capable of passing the Turing Test. Here, interactions take the form of ‘interviews’ on the nature of metaphor—both as defined by the AI and in terms of the metaphors AI uses to describe itself. The purpose of these questions is to prompt the system to reveal its perceived characteristics and self-concept, thereby enabling the construction of a taxonomy of metaphors for AI derived from the system’s own perspective. It will be demonstrated that the AI-generated corpus exhibits advancements in logical-abstract reasoning and increasingly natural interaction patterns. The platforms offer critical and balanced self-descriptions, which are demonstrably influenced by the tone and phrasing of their human interlocutors.