A normative account of the trade-off between cognitive stability and flexibility in task switching

Optimal decision making often requires a compromise between opposing functional demands. This is illustrated by the extensive literature that focuses on the optimal trade-off between response speed versus accuracy. However, a formally rigorous, normative account of the trade-off between cognitive stability versus flexibility is still lacking. In this article, we provide such an account in the context of task switching and consider how the stability-flexibility and speed-accuracy trade-offs can be jointly optimized. We present a mechanistic process model of task switching, in which attractor dynamics of control induce a trade-off between stable processing and flexible reconfiguration. Two behavioral experiments show that human participants broadly conform to normative model predictions, prioritizing flexible reconfiguration (i.e., shallow attractor dynamics) when switching frequently between tasks. These changes in participants’ control dynamics are accompanied by changes in neurophysiological markers of cognitive flexibility (pupil size and posterior baseline alpha power). Crucially, we find that optimization of control is not always reflected in traditional behavioral measures of stability and flexibility. We suggest that relying exclusively on these behavioral measures can explain recent proposals that stability and flexibility can vary independently. More broadly, we argue that hypotheses on the nature of control should be developed and tested using formally rigorous, mechanistically precise models that can make quantitative predictions, capture individual differences, and account for the complex interactions that can occur among multiple, often subtle factors that influence decision-making processes.