Funded Grants

Foundations of goal-directed behavior: Neural and cognitive processes supporting cognitive control

A fundamental problem in cognition concerns how we control our behavior based on goals, plans, and social rules. Psychologists addressing this cognitive control problem have long proposed that we must have a capacity for hierarchical control, or the ability to simultaneously control immediate actions while also holding more abstract goals in mind. Consider an everyday task like making coffee. Behaviorists explained this task as an action-trigger chain, wherein grinding beans triggers putting the grounds in the coffee machine, and so forth. However, people sometimes forget or reverse individual steps of a task, but still proceed with the next steps (e.g., starting the coffee machine having forgotten the grounds). Chaining cannot account for these "broken links". Rather, such slips suggested to cognitive psychologists that we hold an abstract, overarching task context in mind, which runs in the background and influences more specific choices.

Similar hierarchical abstractions also influence rule-governed behavior. For example, as children, we learn rules like speak softly when indoors ("indoor voice"), but outside, it is okay to shout. Here, an element of the context (i.e., indoor) constrains one's manner of speaking. However, children may eventually learn that the "indoor voice" rule is among the broader class of rules that only apply when a caregiver is present. Hence, "Mom" provides an overarching contextual element that determines a class of other context-action relationships. Processes supporting such hierarchical abstractions may underlie learning, planning, decision making, and reasoning. Yet, despite its importance for cognition, the cognitive and neural mechanisms that support hierarchical control remain unknown.

In my lab, we have gained new understanding of this longstanding problem using a computational cognitive neuroscience approach. We have built on recent empirical and theoretical innovations in working memory and reinforcement learning to generate novel hypotheses about the mechanisms of hierarchical control, including by building explicit computational models. This theoretical work has highlighted a capacity termed contingent updating supporting hierarchical control, which is the ability to hold one overarching element of context in working memory in order to influence what else is maintained. We test our hypotheses using a range of cognitive neuroscience methods, including behavioral experimentation, functional and structural brain imaging, and testing neurological patients. These experiments have provided evidence that nested circuits between frontal cortex and striatum may support contingent updating during hierarchical control. In light of the importance of hierarchical control, this relatively simple circuit, when elaborated in the human brain, may be a core component in our capacity for goal-directed thought and action.

My lab is extending this research in several new directions. (1) We plan to use transcranial magnetic stimulation to test how hierarchical control is affected when we disrupt the corticostriatal network. (2) We are exploring sources of individual differences in learning, generalization, and transfer of knowledge to new situations as a function of hierarchical control, including considering effects of early environment and genotype. (3) We are characterizing the coding of contextual representations at different levels of abstraction, through the use of machine learning approaches to fMRI data analysis.