As exemplified by the differential use of analogies by humans and Large Language Models (LLMs), there are fundamental differences between human cognition and LLM processing of language. Analogies are essential to human cognition. LLMs are also capable of generating and understanding analogies; however, their use of analogies differs fundamentally from that of humans. This article highlights the differences between human cognition and the functioning of LLMs. It first provides a thorough definition of analogies. It then describes their role in human cognition and explores the differences between analogies composed by humans and those by LLMs. It next discusses the broader differences between human cognition and LLM processing. Finally, it shares insights that can be gleaned from such a comparison and makes recommendations based on these differences, cautioning against the tendency to anthropomorphize agentic Artificial Intelligence (AI) and other LLM-based tools. Understanding how LLMs are similar—and dissimilar—to human cognition can be helpful in maintaining an appropriate relationship with these tools.
Nanni et al. (Wed,) studied this question.