Human versus artificial creativity: A case study in poetry

Is human creativity distinct from the artificial variety exhibited by Large Language Models (LLMs)? I examine this question in the context of a case study ( N = 1 poet and 1 poem) in which the writing of an original poem by a human is compared to the generation of a poem by an LLM given the same prompt that inspired the human poet. The writing process was extended over months for the human poet, drawing upon broad motivations and background knowledge coupled with emotional reactions to seemingly unrelated experiences and memories. In a few seconds, the LLM generated a reasonably coherent, thematically appropriate poem in systematic meter and rhyme. Two expert poets compared and contrasted the two versions (without being informed as to which was LLM-generated). Each expert reported that the human-authored poem uses formal complexity to convey a nuanced perspective on its theme, whereas the LLM-generated poem uses a simple conventional structure to express bland platitudes. Human creativity can be distinguished from the artificial variety by the nature of the underlying creative processes.