Investigating Cognitive Control Adaptations to Different Reward Environments

Abstract

Constraints on control allocation are defining characteristics of control-dependent processes (Shiffrin & Schneider, 1977). In situations where multitasking is performed through flexible switching, the constraints on control intensity reflect the fundamental tradeoff between how much control is allocated to a single task (stability) and the time needed to switch between tasks (flexibility). Recent computational work uses a recurrent neural network model to provide a normative account of control intensity constraints from the perspective of the stability-flexibility dilemma (Musslick et al., 2018). In this work, we seek to provide empirical support for the predictions made by the model. To this end, we developed an open-source Web browser-based cued task switching paradigm that can be deployed online for rapid, large-scale data collection. We designed and deployed a series of six experiments, two of which manipulated task switching rate and four of which manipulated reward rate contingencies between experimental groups. In the switch rate experiments, we were able to fully account for the model predictions as well as to reproduce and extend the results of previous studies that varied task switching rate (Mayr, 2006; Monsell & Mizon, 2006; Seong, 2018). In the reward rate experiments, we do not have strong evidence to support that human participants instructed to optimize for reward rate seem to adapt their constraints on control in different task environments in accordance with the model predictions. However, this work advances our understanding of reward rate maximization in humans, taking the first steps to show how different learning strategies may produce different behavioral results, and how this may relate to defining model assumptions.