Establishing the behavioral feasibility of a structured story-generation paradigm for neurophysiological research

Historically, many standard measures evaluate creativity at a single time point and are not suitable for capturing the continuous neural activity underlying naturalistic, creative cognition. The aims of our study were to: (a) establish the feasibility of a behavioral paradigm suitable for investigating the continuous, temporal dynamics of creative cognition using neurophysiological methods, and (b) evaluate the feasibility of using large language models (LLMs) for scoring. Sixty undergraduate students (M age = 21.65 years; M/F = 50%) wrote a brief story integrating three unrelated images into a single narrative. Each story was scored by two large LLMs. The ICC(2,1) for total story scores was 0.67 (95% CI [0.51, 0.79]), indicating moderate agreement between models and were averaged into a composite score. The story task produced a broad distribution of scores (M = 9.2, SD = 2.1; range = 4.5-12.5) supporting its ability to generate variable creative output. We compared the total story creativity score to the Inventory of Creative Activities and Achievements (ICAA) and the Divergent Association Task (DAT). There were no significant correlations with ICAA or DAT scores. A random subset (n = 30) of stories rated by two human raters, resulted in moderate agreement with LLM scores (ICC(3,1) = 0.50, 95% CI [0.18, 0.73]). These findings provide preliminary evidence for the behavioral feasibility and characterization of a structured, temporally extended naturalistic, narrative paradigm that can be integrated with neurophysiological methods. Future studies will examine the associated neural signatures and further explore LLM scoring with human raters.