May 2025

The White Matter of Aha! Moments

Individuals can solve creative problems via step-by-step (analytical) thinking or through insight ("Aha!" moments). Studies have identified widespread functional brain activity involved in insight problem-solving, yet, the white matter substrate of insight remains unexplored. In this study, we employed DTI to investigate how white matter microstructure relates to insight versus step-by-step analytical reasoning. After controlling for age and gender, insight was found to be linked to lower FA in the left posterior Arcuate Fasciculus and bilateral Superior Longitudinal Fasciculi III. Conversely, step-by-step reasoning was linked to higher FA in the left Vertical Occipital Fasciculus and higher FA in the anterior corpus callosum. This work exposes distinct structural connectivity patterns associated with different modes of creative problem-solving, contributing to our understanding of the neural architecture supporting creative cognition.
Read more

April 2025

Enhancing Creativity with Covert Neurofeedback: Causal Evidence for Default-Executive Network Coupling in Creative Thinking

Correlational evidence has linked creativity to coupling between the Default Mode Network (DMN) and Executive Control Network (ECN). In this study, we leveraged covert neurofeedback to endogenously modify functional connectivity between DMN and ECN without the participants' knowledge. We compared this to a control neurofeedback condition, entraining coupling between medial prefrontal cortex and supplementary motor area. Approximately 24 hours after neurofeedback, DMN-ECN neurofeedback led to increased coupling between these networks during a creative thinking task (generating creative object uses), extending to broader DMN regions. Behaviorally, we observed a double dissociation: the DMN-ECN neurofeedback increased idea originality, while the control neurofeedback improved go/no-go reaction times. We thus provide the first evidence that DMN-ECN coupling causally enhances creative performance.
Read more

April 2023

Accelerating Creativity: Effects of Transcranial Direct Current Stimulation on the Temporal Dynamics of Divergent Thinking

tDCS over the DLPFC enhances creative thinking, but the temporal dynamics and underlying cognitive mechanisms remain unclear. We investigated how this tDCS affects convergent and divergent thinking, focusing on the serial order effect (ideas becoming more original over time) and the role of cognitive inhibition. In a within-subjects design, participants received three types of cross-hemispheric tDCS (left cathodal/right anodal, L-R+; left anodal/right cathodal, L+R-; and sham) over the DLPFC. They completed flanker tasks measuring cognitive inhibition, plus the AUT measuring divergent thinking and RAT measuring convergent thinking. Results showed that L+R- stimulation significantly enhanced originality in the AUT compared to sham, with no effect on the RAT. The L+R- condition also showed a diminished serial order effect, with accelerated production of original ideas, and was accompanied by better flanker task performance. These findings suggest that L+R- tDCS over the DLPFC accelerates idea originality, with cognitive inhibition potentially mediating the enhancement in divergent thinking.
Read more

August 2022

Automated creativity prediction using natural language processing and resting-state functional connectivity: an fNIRS study

Creative thinking relies on large-scale brain connectivity. Specifically, functional coupling of the DMN and ECN has been linked to creative performance. In this study, we show for the first time that brain signals captured with fNIRS can be used to predict creative performance measured via an automated computational method. We also find that dynamic resting-state functional connectivity is twice as effective at predicting creative performance, as opposed to static resting-state functional connectivity. We extend past work in the network neuroscience of creativity by showing that stable predictions of objective creative performance can be derived from dynamic fNIRS signals.
Read more

May 2024

Mapping the Memory Structure of High-Knowledge Students: A Longitudinal Semantic Network Analysis

In this paper we explored the relationship between semantic network and learning across two studies in undergraduate students enrolled in an introductory psychology course. We administered a cumulative multiple-choice test of psychology knowledge and estimated semantic networks across two domains: domain-specific (psychology) and domain-general (animals). Based on performance on the psychology test, we categorized students into a high-knowledge or low-knowledge group and compared their semantic networks. We found that the high-knowledge group had semantic networks that were more clustered, with shorter distances between concepts compared to the low-knowledge group. We also found the semantic networks of high-knowledge students became more interconnected over time. Successful learners show a distinct semantic network—characterized by high connectivity and short path distances between concepts—both domain-general and domain-specific, highlighting the utility of cognitive network science for studying variation in student learning.
Read more

April 2024

The role of semantic memory networks in crystallized intelligence and creative thinking ability

Crystallized intelligence has been found to support creativity. Yet, the mechanisms that link crystallized intelligence to creativity remain poorly understood. Across two studies, we demonstrate that crystallized intelligence is linked to creative ability as well as semantic memory network structure. We found similar semantic memory network structures that support creativity also underlie high crystallized intelligence. Our findings suggest that flexible access to semantic memory supports both verbal intelligence and creativity.
Read more

June 2023

Convergent thinking and insight problem solving relate to semantic memory network structure

The associative theory of creativity holds that creative thinking involves connecting remote concepts in semantic memory, yet, research has overlooked its applicability to convergent thinking and insight. Convergent thinking problems can be solved with insight (the sudden "aha" experience) or analysis (deliberately and incrementally working towards the solution). In this work, we used network science to compare semantic network structure across two grouping variables: 1) convergent thinking ability (i.e., problem accuracy), and 2) the tendency to solve problems with insight or analysis. Our findings show that convergent thinking ability is linked to flexible and interconnected semantic networks—with short paths and many connections between concepts. Moreover, participants who primarily solved problems with insight showed shorter average path distances between concepts, even after controlling for accuracy. Our results extend the literature on semantic memory and creativity by linking the organization of semantic memory to convergent thinking and insight.
Read more

April 2023

Free Association Ability Distinguishes Highly Creative Artists From Scientists: Findings From the Big-C Project

People's ability to connect remote associations allows them to form creative ideas. Past creativity research has almost exclusively focused on the general population, with far less work examining eminently-creative individuals across the arts and sciences. We analyzed data from the Big-C Project—a sample of world-renowned visual artists and scientists, and an intelligence-matched comparison group—and tested whether the ability to generate remote word associations differs as a function of creative expertise. We found an interaction between domain expertise and association condition: while artists generated more distant associations overall, this was particularly true for free associations (generating the first word that comes to mind). Visual artists spontaneously produced more remote associations but their creative expertise was less relevant for producing associations involving goal-directed cognitive search.
Read more

May 2025

The Roles of Idea Generation and Elaboration in Human-AI Collaborative Creativity

In this work, we explore how humans and AI can collaborate effectively in creative tasks. We investigate the distinct roles of idea generation and elaboration in human-AI collaborative creativity, examining how these processes interact and contribute to creative outcomes. Our findings reveal that while AI excels at generating diverse ideas, human creativity shines in the elaboration and refinement of these ideas. We demonstrate that the most successful creative outcomes emerge when humans and AI work together, leveraging their complementary strengths. This research provides insights into optimizing human-AI collaboration for creative tasks and suggests new directions for developing AI systems that better support human creativity.
Read more

May 2025

Creative Preference Optimization

While LLMs have shown impressive performance on many natural language tasks they still struggle with creativity. Past attempts at enhancing LLM creativity have narrowly focused on improving their outputs' diversity on very specific tasks. In our work, we introduce Creative Preference Optimization, an alignment method that injects signals from multiple creativity dimensions (novelty, quality, diversity, and surprise) to guide the LLMs output. We compiled a large dataset of 200k+ human responses to creativity tasks, rated on their creativity by trained human judges. We used this large dataset to train a model, which was found to outperform several baselines including GPT-4o on creativity tests. Our findings show that creative preference optimization is a promising direction for boosting LLM creativity without compromising output quality.
Read more

March 2025

Automated assessment of creativity in multilingual narratives

LLMs have shown remarkable success on creativity tasks, yet they have not been applied to scoring multilingual creativity data or narratives. In this work, our goal was to develop an LLM that could score the originality of narratives across 11 different languages. We fine-tuned RoBERTa-base on multilingual stories translated into English, and found it strongly predicted human originality ratings (r ≥ .73). We also fine-tuned XLM-RoBERTa on the same stories, in their original language, and found that it also reliably predicted human originality scores (r ≥ .72). We thus demonstrated that LLMs can successfully score narrative creativity in 11 different languages, surpassing the performance of the best previous automated scoring techniques (e.g., semantic distance). This work represents the first effective, accessible, and reliable solution for the automated scoring of creativity in multilingual narratives.
Read more

March 2025

Automated scoring of creative problem solving with large language models: A comparison of originality and quality ratings

Creative problem solving is a naturalistic form of creative thinking involving the generation of solutions that are not only original but also of high quality (i.e., plausible and effective). We examine whether both originality and quality can be automatically scored for a naturalistic creativity task, by two open-source LLMs. We gathered data from 10 studies, amounting to 3,243 participants who completed different items of the creative problem-solving task. We fine-tuned two LLMs, RoBERTa and GPT-2, and few-shot prompted two separate LLMs, Claude and Llama, to predict human ratings of originality and quality on the CPST. We found that RoBERTa and GPT-2 models predict human ratings of solution quality (RoBERTa, r = .83; GPT-2, r = .83) better than solution originality (RoBERTa, r = .79; GPT-2, r = .80). Few-shot prompting was less effective than fine-tuning at predicting both originality (r = .66–.11) and quality (r = .62–.26). We show for the first time that naturalistic creativity tasks can be automatically scored for both originality and quality.
Read more

February 2024

Bridging the measurement gap: A large language model method of assessing open-ended question complexity

Question-asking, an essential yet often understudied activity, holds significant implications for fields such as learning, creativity, and cognitive development. Question complexity has been found to be a crucial factor affecting question quality level. However, assessing the complexity of questions, and especially open-ended questions, remains a methodological challenge. In this study, we develop a computational model that automatically scores the complexity of open-ended questions. Our results reveal that our LLM-generated Bloom scores correlated strongly with human ratings of complexity (r = .73), whilst also greatly exceeding baseline methods tested. The research emphasizes the significance of LLM in automating the assessment of open-ended question complexity, fostering cost-effective, automatic, and reliable measurements in this domain.
Read more