Study Finds Humans Benefit from Cross-Domain Mapping, LLMs Show No Effect

A recent study published on arXiv (2603.19087) explores the impact of cross-domain mapping—using analogies from unrelated source domains—on creativity in humans and large language models (LLMs). Participants, both human and LLMs, were tasked with creating innovative designs for ten everyday items (such as backpacks and TVs) under two scenarios: one involving cross-domain mapping, which drew inspiration from randomly assigned sources (like octopuses, cacti, or GPS), and another centered on unmet user needs. Findings indicate that humans consistently gain from these random mappings, yielding more original concepts. Conversely, while LLMs typically produce more original ideas than humans, they did not show a significant response to cross-domain mapping. This research enhances our understanding of creativity in AI and humans, impacting design and innovation. The paper can be accessed on arXiv with the identifier 2603.19087.