Cognitive Science of Creativity
Despite its critical importance to society, the process by which creative ideas emerge remains elusive. To assess the creative quality of ideas, traditional approaches have largely relied on human ratings. In my first publication (Orwig et al., 2021, Journal of Cognitive Neuroscience) my collaborators and I applied a novel semantic distance measure to automatically assess the novelty of ideas. Additionally, we analyzed functional Magnetic Resonance Imaging (fMRI) to capture dynamic interactions between brain regions, identifying patterns of resting-state network connectivity associated with creativity.
Our analysis revealed a robust correlation between computational scores and human ratings of creative ideas, validating semantic distance assessments as a viable measure of creative performance. We applied voxel-level, graph theory analysis in resting-state fMRI data to describe individual differences in brain connectivity. Results show reduced connectivity of visual cortex associated with creativity, assessed via computational semantic distance and human ratings.
To replicate and extend these findings, we performed the same analysis in a cohort of creative experts. In this follow up study (Orwig et al., 2023, Network Neuroscience) we found that creative experts show reduced connectivity between primary visual cortex and the rest of the brain, compared to controls. Here, we further examined associations between functional brain connectivity and distal simulation, finding a negative association between distal simulation vividness and connectivity to the lateral visual cortex in creative experts.
Creative Writing in Humans & AI
Recent advances in artificial intelligence have raised important questions about whether large language models (LLMs) can replicate, and potentially surpass, aspects of human creativity. To investigate this, we conducted a series of behavioral studies to characterize salient features of creative writing and compare stories written by humans with those generated by AI (Orwig et al., 2024, Journal of Creative Behavior). Participants and LLMs were each prompted with three-word cues and asked to compose short stories elaborating on those cues. Across both human- and AI-generated narratives, we found that greater semantic diversity and the inclusion of more perceptual details predicted higher ratings of creativity. Notably, creativity ratings produced by AI were highly consistent with those made by human raters, and overall creative quality did not differ between human and AI stories. These results suggest that current LLMs can reproduce key dimensions of human creative expression, offering new insight into the cognitive and computational basis of storytelling and raising the possibility of future human-AI collaboration in creative domains. This finding was highlighted in a newsletter by the Harvard Brain Science Initiative.
Forecasting Impact of Ideas
Ongoing investigations at the intersection of cognitive psychology, innovation and entrepreneurship examine how evaluators forecast the potential impact of new ideas. Using a combination of observational data and lab-based experiments, my collaborators at Harvard Business School and I describe find that pitches written in more concrete language were consistently judged to have greater potential impact. Across two large-scale observational studies and one preregistered experiment, this effect was explained in part by evaluators’ ability to more vividly imagine the future outcomes of a venture. These findings suggest that concrete language supports idea evaluation by scaffolding episodic simulation, thereby helping evaluators envision how a proposed solution might unfold in the real world. More broadly, this work extends cognitive theories of constructive memory into an organizational context, offering practical insights into how innovators can effectively communicate the promise of new ideas.
