About the project:
CreHack is a project of cognitive neuroscience, partially funded by the Paris Region Fellowship Program, Horizon 2020 Marie Skłodowska- Curie n° 945298.
The main part of this project has received funding from the European Union’s Horizon 2020 research and innovation program under the Marie Sklodowska-Curie grant agreement No 101026191.
The Fondation des Treilles supported this work
An official ethics committee approved the project (CPP Ouest II – Angers, RCB: 2019-A02511-56).
Creativity is an impressive cognitive ability that makes humans an extraordinary species. To face the timely challenges of our society, creativity is a critical and necessary skill that plays a game-changer role in innovation. Yet, the cognitive and neural mechanisms of creativity are still poorly understood. Formally, creativity is defined as the ability to produce an object/idea that is both original and efficient. Then, creativity should involve an evaluative process of efficiency and originality, interacting with a generative process that generates candidate ideas. Today, a unitary modelling approach is still lacking in the field of creative cognition, and the evaluative process has been poorly explored.
Through computational modelling, the CreHack project aims at investigating the specific role of the evaluative component in the creativity mechanisms. This approach allows untangling the different steps of the creative process. The project uses a novel experimental design to develop this model and investigates its neural validity through neuroimaging (fMRI).
A video explaining the results of the first study of CreHack. In french, with english subtitles:
During the fellowship, we conducted two behavioral studies and one fMRI studies.
The first behavioral study has been published (Lopez-Persem et al, 2023) and introduces a new computational model to explain the production of creative ideas. Critically, this model includes a "valuator" component, that assigns subjective values to candidate ideas considered by an agent before producing their final response. We found that agents select their final idea based on the values of ideas. Subjective values can be understood as the quantification of how much we like/enjoy things. It is directly linked to subjective preferences. In this study, participants had to provide creative word associations in response to a cue word. Interestingly, we found that the more participants like their ideas, the faster they want to provide them, and the faster they type them on a keyboard. This study demonstrates for the first time the mechanistic role of preferences in semantic creativity.
The second behavioral study (Battistello et al, PsyArXiv 2024) replicated and extended the results to two additional domains of creativity: drawings and alternative uses of objects. It demonstrates that preferences have a role in any domain of creativity.
The third study was conducted inside an MRI scanner (with the same design as the first behavioral study) and we discovered that the Brain Valuation System, an equivalent to the reward system, is proportionally activated with how much participants like their responses when generating them. This study confirms the crucial role of preferences in creativity, through a biological explanation. It demonstrates that the Brain Valuation System assigns values to any kind of items, including self-generated ideas, and that creativity partly relies on how we like our ideas. The manuscript published here: Moreno-Rodriguez et al, 2025.
Those three studies have been presented by the first authors (Battistello and Moreno-Rodriguez) during international conferences on posters and by A. Lopez-Persem through invited talks in national and international seminars and international conferences.
General public outreach was also achieved during the project, with A. Lopez-Persem regularly participating in national and international interviews for diverse media (BBC, Le Monde, France Culture, Phosphore, etc) and general science events (Brain week).
You will find below an online tool to estimate the valuation parameters (preference profile) from a dataset containing ratings of likeability, adequacy, and originality.
You can download an example of data here.
Note that the tool allows the estimation of parameters for a single individual, it does not deal yet with multiple participants.
Data should be between 0 and 100 for a proper display of the figure. However, the parameter estimation will normally work with any range of values, as soon as all ratings have the same range (parameter values might change a bit but overall stay in the same range).
When uploading data in the tool, no copy of your data is made. We don't save the file you submit.
If you use this tool and publish results with it, please cite Lopez-Persem et al, 2024, Moreno-Rodriguez et al 2025 and Battistello et al, 2024.
NB: The method to estimate the parameters in this online tool differs from the one used in our papers, that utilises the VBA toolbox.
Method
Here, we employed a nonlinear least-squares optimization approach to estimate the parameters α and δ. The goal of the fitting procedure was to minimize the sum of squared errors (SSE) between the observed and predicted values of Likeability.
Parameter Estimation
The optimization was performed using a numerical minimization algorithm, specifically the unconstrained nonlinear minimization function from the numeric.js library. The parameters α and δ are constrained between 0 and 1, and -1 and 2, respectively. The initial parameter estimates were chosen to correspond approximately to α≈0.5 and δ≈1.
The optimization was conducted iteratively by evaluating an objective function, which computed the SSE for a given set of parameter values. The numerical optimization was constrained by a termination criterion based on the tolerance level (ϵ=10−6) and a maximum iteration limit of 1000.
Model Validation
To assess the goodness-of-fit, the coefficient of determination (R2) and root mean squared error (RMSE) were computed.
Help from AI: To develop this tool and its explanations, I used Claude.ai and ChatGPT-4o. However, I take full responsibility for the functionality of the script and the accuracy of the explanations.