Outlines of a Fuzzy Approach for Characterizing Purposefulness in Human and LLM Generated Text

Following the recent growth in popularity of large language models (LLM), conflicting intutions have arisen as to whether such models are simply “parrots” of human language, or if they represent a unique cognitive profile. In this paper, we identify the dimensions of purposefulness and distinctiveness underlying semantic uniqueness of text-based data, and we propose a fuzzy approach to help characterize the purposefulness in the text generated by humans and LLM models. We do this with the aim of understanding areas where the outputs of LLM tools are qualitatively different from human outputs, as well as with the goal of evaluating changes in these differences through the course of time. Based on our preliminary evaluations, we draw the conclusion that within different parts of speech categories, current flagship LLMs may be more or less purposeful in their generation of text than humans are, exhibiting comparatively different degrees of variation in the specific words that they use.