Hedging your bets by learning reward correlations in the human brain.

Area of Science:

• Neuroscience
• Cognitive Science
• Decision Making

Background:

• Humans effectively track reward statistics like mean and variance using prediction errors.
• Ecological environments often feature multiple reward sources with complex interdependencies.

Purpose of the Study:

• To investigate if humans can learn higher-order correlations between distinct environmental outcomes.
• To understand the neural mechanisms underlying the representation of these correlations.

Main Methods:

• Subjects performed a task manipulating the correlation between distinct outcomes.
• Behavioral choices were analyzed to assess learning of outcome correlations.
• Neuroimaging (fMRI) was used to identify neural correlates of correlation learning.

Main Results:

• Subjects successfully learned and utilized outcome correlations to minimize variance in their choices.
• Explicit model-based strategies were employed to learn these correlations.
• Neural representations of outcome correlations were found in the right midinsula.
• A learning prediction error signal was identified in the rostral anterior cingulate cortex.

Conclusions:

• The human brain explicitly represents higher-order correlation structures between rewards.
• This ability is crucial for adaptive decision-making and optimized environmental sampling.
• Neural mechanisms involve distinct brain regions for correlation representation and error signaling.