On the edge of understanding: ChatGpt and the limits of artificial sense-making

Abstract

The future challenge of generative AI such as ChatGpt will be to be as similar as possible to a human speaker. Certain notions such as linguistic sentiment and linguistic value in Saussure or decipherer in Jakobson will show how, although the technological evolution of AI is undisputable, the future of AI is more complex than it seems. The human speaker “lives” language – or probably lives within language – while AIs “train themselves” to recognise it mechanically. A fundamental aspect which connotes the human speaker is the adaptability to the communicative situation. During a dialogue, two or more speakers stipulate “a dynamic pact”, mutually adjusting and cooperating to render communication effective and to generate sense. The adaptability to the communicative situation implies the recognition of certain extralinguistic factors (distraction, ambiguity between two different signifiers, the request for confirmation of understanding, etc.). ments which constitute Hymes’ taxonomy, especially the environmental context (spatio-temporal definition of the communicative situation) and the scene (cultural definition of the communicative situation). According to this idea, AIs reproduce a communicative exchange – by generating predetermined messages, through a collective mind, following Lotman – whereas human speakers produce it through their individual minds. From Lotman’s perspective, sense-generating structures rely on a dynamic interplay between symmetry and asymmetry, continually entering into relationship with both synchronic and diachronic structures; in the full article, I will analyze selected dialogues with ChatGPT to explore the potential coexistence of artificial language and natural language, considered by Lotman as a crucial component in the process of sense-making. In these terms, if generative AIs are considered capable of generating sense, then probably the concept of sense – and mind (?) – itself can be semiotically redefined.