LLMorphism: When humans come to see themselves as language models (arxiv.org)

A new concept dubbed "LLMorphism" highlights a growing belief that human cognition functions like that of large language models (LLMs). This phenomenon suggests that as conversational LLMs increasingly generate human-like language, people may mistakenly infer that if LLMs can express themselves like humans, then human thought processes must also mirror LLM operations. Such a bias could have profound implications for how society perceives human cognition, as LLMorphism can spread through analogical transfer—projecting characteristics of LLMs onto humans—and metaphorical availability, where LLM terminology becomes a dominant framework for discussing human thought. The implications of LLMorphism are extensive, influencing areas such as work, education, healthcare, and communication. It raises critical questions about the responsibilities we attribute to both machines and humans, potentially diminishing our recognition of human cognitive uniqueness. This trend poses a risk of dehumanization, as people may increasingly underestimate the complexity of human thought in favor of simpler, model-like descriptions. The conversation around AI's societal impact must not only address whether we attribute too much intelligence to machines but also consider the risk of attributing too little to ourselves.