Representations of cognitive maps underlying model-based planning

Poster Session D - Monday, April 15, 2024, 8:00 – 10:00 am EDT, Sheraton Hall ABC

Ata Karagoz¹ (a.b.karagoz@wustl.edu), Wouter Kool¹, Zachariagh Reagh¹; ¹Washington University in St. Louis

When faced with a decision, we sometimes rely on habit, which serves us well in predictable situations. However, in more complex scenarios, we often need to plan towards a goal. Over the last decade, goal-directed planning has been studied using sequential decision-making tasks coupled with model-based reinforcement learning algorithms. However, model-based systems can only successfully plan if they have access to an accurate model of the environment. There is a surprising lack of knowledge on how these internal representations of task structure, or cognitive maps, are constructed. This is partly because many prior studies ensured that participants had full knowledge of the structure of the task before having to make decisions. How do humans construct cognitive maps, and how is this instantiated in the brain? How do their cognitive maps relate to their ability to use model-based planning? To address these questions, we modified an established two-step decision-making task and conducted behavioral and neural representational similarity analyses (RSA) over planning relevant versus irrelevant associations between choice options in the task. In a sample of participants undergoing fMRI scanning, behavioral RSA reveals that participants come to represent abstract, planning-related relationships among items in the task. Alongside this, neural patterns in parahippocampal cortex, and orbitofrontal cortex come to represent these relationships. Further, a whole brain searchlight reveals a broader set of brain regions involved in representing task structure, including mediodorsal thalamus. Our results suggest that cognitive maps represent multiple facets of task structure, which are in turn encoded by distinct neural regions.

Topic Area: THINKING: Decision making