Artificial Intelligences’ Self-Perceptions and their Social Representations in the Light of the Bakhtinian Dialogic Theory

Abstract

Objective: To investigate, from the perspective of the Theory of Social Representations and Bakhtinian dialogism, the self-perceptions of LLMs and the social representations that emerge from their statements. Method: Qualitative, descriptive and exploratory research. Similarity Analysis, Word Cloud and Factorial Plan in responses generated by generative Artificial Intelligences based on Large Language Models freely available on the Internet, through IRAMUTEQ and Trideux-Mots. Results: When generative artificial intelligences were instigated to express their self-perception, they revealed a discourse strongly aligned with a technicist, functional and positive perspective. Discussion: From a theoretical-methodological point of view, these associations cannot be understood as mere relations between words, but as ideologically oriented statements that are part of a broader discursive field. Conclusion: The representations evoked by AIs about themselves are, in fact, human social representations internalized in the systems.