“Neural and computational signatures of metacontrol”
Hosted by the WashU Neuroimaging Community (WUNIC)
Abstract: Human decision making is sometimes guided by habit, and at other times by goal-directed planning. In the reinforcement learning literature, this distinction can be formalized as a competition between a computationally cheap but inaccurate “model-free” system, and a computationally expensive but more accurate “model-based” system. The last decade has seen a notable increase in our understanding of the neural and cognitive mechanisms that underlie these decision-making systems. However, it remains unclear how the brain allocates control between them. Here, we present behavioral, computational, and neuroimaging work, which together suggest that this arbitration is guided by a comparison of each system’s benefits, discounted by an intrinsic cost for model-based planning. First, I will describe behavioral results that underline the idea that people trade off costs and benefits during metacontrol. Next, I will report a computational model that is able to describe to perform such arbitration without incurring meta-level effort costs. Finally, I will present results from two neuroimaging experiments. The first was aimed to uncover neural correlates of two key signals in the metacontrol model, and the second to investigate whether the implementation of model-based planning is a form of mental simulation.
The WUNIC seminar series takes place virtually on the second Friday of each month starting promptly at 1.30 pm.
For inquiries contact Janine Bijsterbosch.