Metaphor in population-based AI

Abstract

Abstract English serves as the primary source language for Artificial Intelligence terminology, playing a pivotal role in shaping the metaphorical expressions that underpin specialized lexicons and facilitate knowledge transfer. This study investigates the patterns of metaphorical term formation within English-language AI discourse, focusing specifically on the domain of population-based algorithms. Through a qualitative, corpus-informed analysis of 1,540 terminological units drawn from academic literature and specialized glossaries, we identify and classify 85 metaphorical terms (5.5% of the dataset). The analysis reveals that metaphorical transfer in this field is systematically anchored in four dominant source domains: evolution, living nature, inanimate nature , and society . Employing Conceptual Metaphor Theory (Lakoff & Johnson), Frame Theory (Minsky), and Sociocognitive approach to terminology (Temmerman), the study maps these cross-domain mappings drawn from societal organization and biological systems. The findings demonstrate that these metaphors function as essential cognitive tools that structure technical understanding, foster interdisciplinary dialogue, and enhance the intelligibility of complex systems. Besides, the analysis critically examines potential communicative effects, such as unintended anthropomorphism or the naturalization of competitive framings, which carry implications for ethical discourse and human-centered AI. The given study provides a framework for analyzing term formation, supports transparent naming practices, and highlights the influence of metaphorical choices on both the conceptual architecture and the societal perception of Artificial Intelligence.