A Model of Learning the Optimal Balance Between Cognitive Stability and Flexibility

Abstract

Theories of cognitive control have long recognized the existence of a limitation on the intensity of cognitive control that can be allocated to not only multiple tasks, but also to a single task. Approaching this facet of cognitive control from the perspective of the stability-flexibility dilemma yields a rational account for why a limitation may exist even on cognitive control allocation to a single task. Stated simply, higher demands for flexibility within a task set put a limitation on the intensity of cognitive control that can be allocate to any given task in the sequence so as to facilitate the ability of an agent to switch more freely between tasks on demand. In this study, we construct, implement, and test a computational model of learning an optimal balance between cognitive stability and flexibility in a two-task environment. We first analyze the basic behavior of the model and compare its outputs to previously developed models of the stability-flexibility tradeoff, and use these findings to test our model’s capability to learn an optimal control signal given a set of flexibility demands. Critically, we find that our computational model is indeed capable of learning to optimize control signal allocation based on demands for flexibility. Upon feeding our model with simulations with high flexibility demands, our model outputs low values of optimal control signal intensity compare to optimal control signal intensities found when given simulations with low flexibility demands. These results are congruous with similar computational models within the literature, and cast this study as a promising learning model for cognitive demands that can be built upon within future studies.