Diálogos escritos com programas de inteligência artificial: entre comunicação e interação

Abstract

Motivated by the questioning made by the mathematician and computer scientist Alan Turing - "can machines think?" - in 1950, this research aims to investigate if, in dialogical communication between humans and chatterbots of Artificial Intelligence interaction may occur. The research was developed in the research group of Theories and Practices Discursive and Textual, specifically the line of research of Text, discourse and teaching: reading and writing processes of written and spoken text, Master's Program in Linguistics University Cruzeiro do Sul. The research has as corpus dialogues produced by the software of Artificial Intelligence Eugene Gootsman, winner of the event that promoted a competition of the Turing Test, organized by the University of Reading, at the Royal Society in London, in 2014; an interview conducted by the American news site TIME, and ELIZA, considered the first chatterbot, developed in 1966, by US computer scientist Joseph Weinzenbaum, in order to simulate the behavior of a rogerian psychotherapist through verbal language. The criterion adopted for development of the analysis of the texts that compose the corpus follows the methodological assumptions of the theoretical axis of pragmatic studies, being based on concepts of the Theory of Speech Acts proposed by Austin (1990), in studies referring to the principles that govern (1997), Wolfgang Iser (1999) and Jacques Moeschler (1986), and finally concepts related to intentionality in language, according to Searle (1995). The analyzes pointed out that chatterbots communicate adequately with humans, answer their questions, but these don’t follow the principles of verbal interaction that indicate that interlocutor has reconstructed implicitly expressed intentions. It’s possible to say that there is communication, however, there is no full interaction.