Can Large Language Models Think and Experience Emotions and Sensations?

Abstract

The rapid development of large language models (LLMs) raises fundamental questions about the extent to which their “thinking” resembles human cognition, emotions, and subjective experiences. Is embodiment – including a body sensory apparatus, and the ability to act in the world - necessary for such thinking to occur? This article addresses the issue from two complementary perspectives. First, we examine practical advances in generative AI, with emphasis on multimodal systems and robotic platforms. Second, we offer a theoretical analysis of the differences between these systems and the cognition and phenomenal consciousness of living organisms, grounded in a new framework: the Motivated Emotional Mind (MEM) model. We identify fundamental differences between artificial and biological systems in terms of knowledge acquisition and structuring; methods of categorization and generalization of sensory inputs; formation of their representations; and associative processes that enable deductive, inductive, and abductive reasoning. We argue that the pursuit of human-like thinking and feeling is only meaningful in embodied robotic systems. Finally, we advance the hypothesis that the emergence of subjective sensations and feelings requires satisfaction of the criteria specified by the MEM model, which we propose as a necessary condition for phenomenal consciousness. Thus, the MEM model may guide AI design, inform philosophy of mind, or shape ethical debates on machine consciousness.

Keywords