Explaining Cognitive Control Constraints from the Perspective of the Flexibility-Stability Dilemma

Abstract

Recent theories on the allocation of cognitive control assume an intrinsic cost that constrains the amount of cognitive control that a neural system can allocate to a single task. However, the foundation of this constraint on control has yet to be established. The stability-flexibility dilemma states that a tradeoff exists between the ability of a cognitive system to focus entirely on one task and its ability to flexibly switch between tasks. This dilemma, when viewed from a dynamical systems perspective, suggests a rational basis for at least one source of constraint on control: the neural representation of a task perseverates as more control is allocated to a task, thereby impairing the ability of the cognitive system to switch to another task. Thus, constraining the amount of control for a given task may make it easier to switch to others, conferring greater flexibility of processing in environments that require frequent switching from one task to another. We test this hypothesis in an explicit model of the dynamics underlying the allocation of control and task performance and demonstrate that while constraints on cognitive control impair performance on a given task, they also reduce performance costs associated with task switches. Critically, we find that neural networks in environments with higher demands for flexibility learn to optimize performance by inducing greater constraints on control. Finally, performance of neural networks with optimized constraints on control matches that of human participants, who yield lower task switch costs in environments with higher switch rates.