How Subjective Idea Valuation Energizes and Guides Creative Idea Generation

Abstract

What drives us to search for creative ideas and why does it feel good to find one? While previous studies demonstrated the positive influence of intrinsic motivation on creative abilities, how reward and subjective values play a role in creative mechanisms remains unknown. The existing framework for creativity investigation distinguishes generation and evaluation phases, and mostly aligns evaluation to cognitive control processes, without clarifying the mechanisms involved. This study proposes a new framework for creativity by 1) characterizing the role of individual preferences (how people value ideas) in creative ideation and 2) providing a computational model that implements three types of operations required for creative idea generation: knowledge exploration, candidate ideas valuation (attributing subjective values), and response selection. The findings first provide behavioral evidence demonstrating the involvement of valuation processes during idea generation: preferred ideas are provided faster. Second, valuation depends on the adequacy and originality of ideas and determines which ideas are selected. Finally, the proposed computational model correctly predicts the speed and quality of human creative responses, as well as interindividual differences in creative abilities. Altogether, this unprecedented model introduces the mechanistic role of valuation in creativity. It paves the way for a neurocomputational account of creativity mechanisms.

Toward greater computational modeling in neurocognitive creativity research.

“…Likewise, the diversity of creative tasks implies that NCR may require a corresponding diversity of models, at least initially (Poile & Safayeni, 2016; Wilson & Collins, 2019). While a single model capable of explaining performance across multiple tasks or domains would be a significant advance for the field, models focusing on individual creative tasks have proved highly useful to our understanding of how creative outcomes can arise from cognitive processes (Lopez-Persem et al, 2022; Olteţeanu & Falomir, 2016; Schatz et al, 2018).…”

“…For example, paradigms such as free association and the RAT have just single words as input and output, removing the need to model sentence generation. Indeed, several recent models of creative cognition have investigated the generation of single words using semantic networks, as an effective means to study creativity quantitatively (Lopez-Persem et al, 2022; Olteţeanu & Falomir, 2015; Schatz et al, 2018). Semantic networks are formed of nodes representing concepts, and edges representing associative links, and can simulate how activation spreads from a cue to a response (Beaty & Kenett, 2023).…”

“…Another approach is to construct networks from individual participant data, and explore their structural properties in relation to measures of creative cognition (Benedek et al, 2017; He et al, 2020). A further option for semantic network models is to build a single, standard network using free association data (Nelson et al, 2004) or distributional semantics methods (Rotaru et al, 2018), and then fit the model to individual participants by modifying the simulated processes that operate on this network (e.g., Lopez-Persem et al, 2022; see also Benedek & Neubauer, 2013; Volle, 2018). This involves defining a set of processes that determine how activation spreads through memory, such as associative and controlled processes, and then adjusting the parameters governing these processes to fit an individual participant’s data and mimic their creative behavior.…”

“…In addition, since any task can be modeled in various ways, it is important to develop multiple models of each task and then compare their performance to human data (Poile & Safayeni, 2016; Wilson & Collins, 2019). For example, semantic memory retrieval can be modeled as a random walk (Kenett & Austerweil, 2016; Lopez-Persem et al, 2022) or as an exploratory process of optimal foraging (Hills et al, 2012). By comparing the goodness of fit of different models to human data, we can determine which model and its underlying hypotheses are more supported, leading to the development of further models and empirical research questions.…”

“…One method to increase integration between the two fields would be to improve the value of models to empirical researchers. For example, with some exceptions (e.g., Lopez-Persem et al, 2022; see also Augello et al, 2016), the models discussed have not explicitly aimed to embody a particular cognitive theory from NCR in a way that would enable researchers to examine the theory’s predictions or to test new hypotheses. Indeed, several clear steps could be taken to improve future models of creativity, to increase their ability to simulate human cognition and maximize their explanatory value to NCR.…”