Executive control difficulties are common in individuals with autism and are associated with challenges completing tasks and managing time. The study, published in Biological Psychiatry: Cognitive Neuroscience and Neuroimaging, sought to tease out whether these difficulties represent a disruption in proactive executive control (engaged and maintained before a cognitively demanding event) or in reactive executive control (engaged as the event occurs). Using functional magnetic resonance imaging (fMRI), the researchers took brain scans of 141 adolescents and young adults ages 12-22 (64 with autism, 77 neurotypical controls) enrolled in the Cognitive Control in Autism Study. During the scan, the participants completed a task that required them to adapt their behavior. They were shown a green or red cue, followed by a white arrow (probe) pointing left or right.

How do people decide how long to continue in a task, when to switch, and to which other task? Understanding the mechanisms that underpin task interleaving is a long-standing goal in the cognitive sciences. Prior work suggests greedy heuristics and a policy maximizing the marginal rate of return. However, it is unclear how such a strategy would allow for adaptation to everyday environments that offer multiple tasks with complex switch costs and delayed rewards. Here we develop a hierarchical model of supervisory control driven by reinforcement learning (RL). The supervisory level learns to switch using task-specific approximate utility estimates, which are computed on the lower level. A hierarchically optimal value function decomposition can be learned from experience, even in conditions with multiple tasks and arbitrary and uncertain reward and cost structures. The model reproduces known empirical effects of task interleaving. It yields better predictions of individual-level data than a myopic baseline in a six-task problem (N=211). The results support hierarchical RL as a plausible model of task interleaving.

One major limitation of current cognitive architectures is that models are typically constructed in an "empty" architecture, and that the knowledge specifications (typically production rules) are specific to the particular task. This means that general executive control strategies have to be implemented for each specific model, which means a lack of consistency and constraint. Alternatively, they are implemented as part of the architecture itself, which is often implausible, because strategies are learned and differ among individuals. The alternative is to assume executive control consists of strategies that can transfer from one task to another. The PRIMs theory (Taatgen 2013) provides a modeling framework for this transfer. The approach is discussed using the example of working memory control.