Towards Automaticity in Reinforcement Learning: A Model-Based Functional Magnetic Resonance Imaging Study.

Area of Science:

• Neuroscience
• Cognitive Science
• Decision Making

Background:

• Previous research indicates medial prefrontal cortex (mPFC) neurons adapt to high-value options during learning.
• Later learning stages show a shift to automatic processing governed by the basal ganglia.

Purpose of the Study:

• To test the hypothesis that early learning involves a goal-directed system in the mPFC.
• To investigate if late learning relies on a habitual system in the dorsal striatum.

Main Methods:

• Functional magnetic resonance imaging (fMRI) was used to collect blood oxygen level dependent (BOLD) signal data.
• Participants (N=12) performed a reinforcement learning task with monetary rewards and penalties.
• Region of interest (ROI) analysis focused on basal ganglia and medial frontal regions.

Main Results:

• During early learning (reward condition), significant activity was observed in the medial frontal cortex.
• During later learning (reward condition), activity shifted to the putamen.
• In the loss condition, the globus pallidus internal segment showed activity during later learning trials.

Conclusions:

• Brain activation shifts from medial frontal regions to dorsal striatum during reinforcement learning.
• Findings support the existence of separable early goal-directed and late habitual learning systems.