Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Serendipity by Design: Evaluating the Impact of Cross-domain Mappings on Human and LLM Creativity

Qiawen Ella Liu, Marina Dubova, Henry Conklin, Takumi Harada, Thomas L. Griffiths

Abstract

Are large language models (LLMs) creative in the same way humans are, and can the same interventions increase creativity in both? We evaluate a promising but largely untested intervention for creativity: forcing creators to draw an analogy from a random, remote source domain (''cross-domain mapping''). Human participants and LLMs generated novel features for ten daily products (e.g., backpack, TV) under two prompts: (i) cross-domain mapping, which required translating a property from a randomly assigned source (e.g., octopus, cactus, GPS), and (ii) user-need, which required proposing innovations targeting unmet user needs. We show that humans reliably benefit from randomly assigned cross-domain mappings, while LLMs, on average, generate more original ideas than humans and do not show a statistically significant effect of cross-domain mappings. However, in both systems, the impact of cross-domain mapping increases when the inspiration source becomes more semantically distant from the target. Our results highlight both the role of remote association in creative ideation and systematic differences in how humans and LLMs respond to the same intervention for creativity.

Serendipity by Design: Evaluating the Impact of Cross-domain Mappings on Human and LLM Creativity

Abstract

Are large language models (LLMs) creative in the same way humans are, and can the same interventions increase creativity in both? We evaluate a promising but largely untested intervention for creativity: forcing creators to draw an analogy from a random, remote source domain (''cross-domain mapping''). Human participants and LLMs generated novel features for ten daily products (e.g., backpack, TV) under two prompts: (i) cross-domain mapping, which required translating a property from a randomly assigned source (e.g., octopus, cactus, GPS), and (ii) user-need, which required proposing innovations targeting unmet user needs. We show that humans reliably benefit from randomly assigned cross-domain mappings, while LLMs, on average, generate more original ideas than humans and do not show a statistically significant effect of cross-domain mappings. However, in both systems, the impact of cross-domain mapping increases when the inspiration source becomes more semantically distant from the target. Our results highlight both the role of remote association in creative ideation and systematic differences in how humans and LLMs respond to the same intervention for creativity.
Paper Structure (18 sections, 1 equation, 3 figures)

This paper contains 18 sections, 1 equation, 3 figures.

Table of Contents

  1. Introduction
  2. Background
  3. Creativity as the ability to bridge remote associations
  4. Eliciting Novel Ideas from Humans and LLMs
  5. Human idea generation
  6. Participants.
  7. LLM idea generation
  8. Rephrasing and Format Normalization
  9. Evaluating Ideas
  10. Results
  11. Originality, Feasibility, Usefulness, and Investment Worthiness
  12. Meaningful differences in originality between humans and LLMs, and among LLMs
  13. Cross-domain mappings increase originality for humans, but not for LLMs
  14. What combinations tend to yield more original ideas?
  15. Discussion
  16. ...and 3 more sections

Figures (3)

  • Figure 1: Cross-domain mapping increases originality for humans but yields weaker benefits for LLMs. Ideas (points) are presented along originality (vertical axis) and feasibility (horizontal axis), with density contours showing distributions for human (green) and LLM (orange) outputs. Cross-domain prompting (left) produces more highly original human ideas relative to the user-need prompt (right).
  • Figure 2: Originality ratings by model and condition. Distributions show originality ratings for ideas generated by different models under each prompting conditions. Diamonds indicate estimated marginal means, with error bars showing 95% confidence intervals. Box plots show medians and quartiles, with individual data points overlaid.
  • Figure 3: Originality varies by cross-domain combinations and semantic distance: (a) originality ratings for human vs LLM ideas given the same target-source combinations are significantly correlated. (b) Mean originality ratings plotted against Wikipedia-based semantic distance between source and target concepts by source (human vs LLM). (c–d) Means of originality for each target concept (rows) and inspiration/source domain (columns), displayed separately for LLMs (c) and humans (d); numbers denote average originality and colors encode magnitude.