Similarities and differences in spatial and non-spatial cognitive maps.

Area of Science:

• Cognitive science
• Neuroscience
• Computational modeling

Background:

• Spatial learning and generalization are often linked to a
• cognitive map
• neural representations of spatial relationships.
• Recent findings suggest similar neural mechanisms are involved in abstract, conceptual reasoning.

Purpose of the Study:

• To investigate the extent to which spatial and conceptual reasoning share common computational principles.
• To examine the behavioral implications of these shared and distinct mechanisms.
• To model human generalization and exploration strategies in both domains.

Main Methods:

• A within-subject design was employed, comparing spatial and conceptual multi-armed bandit tasks.
• Participant behavior was analyzed using a Bayesian model incorporating distance-dependent generalization and uncertainty-guided exploration (Gaussian Process regression).
• Confidence judgments were collected to assess sensitivity to uncertainty.

Main Results:

• Behavior in both spatial and conceptual tasks was well-captured by the Gaussian Process model, showing equivalent generalization.
• Participants exhibited domain-specific differences: reduced uncertainty-directed exploration and increased random exploration in the conceptual domain.
• A one-directional transfer effect was observed: spatial task experience enhanced conceptual task performance, but not vice versa.

Conclusions:

• Value-guided learning and generalization recruit cognitive-map dependent mechanisms in both spatial and conceptual domains.
• Despite shared mechanisms, domain-specific differences exist in how representations map to actions, particularly in exploration strategies.
• Participants' ability to leverage uncertainty estimates for exploration differs between spatial and conceptual tasks.