Synergizing habits and goals with variational Bayes.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Biological and artificial agents require efficient and flexible behavior.
• Behavior is typically categorized as fast/inflexible (habitual) or slow/flexible (goal-directed).
• Emerging evidence suggests a complex interaction between habitual and goal-directed systems, challenging distinct system models.

Purpose of the Study:

• To introduce a theoretical framework unifying habitual and goal-directed behaviors.
• To model the interplay between these behaviors using variational Bayesian theory and a Bayesian intention variable.
• To explain experimental observations of behavior interaction through computational simulations.

Main Methods:

• Development of a theoretical framework based on variational Bayesian theory.
• Incorporation of a Bayesian intention variable to represent behavioral states.
• Simulations of vision-based sensorimotor tasks to test the framework's predictions.

Main Results:

• Habitual behavior is modeled by the prior intention distribution, derived from sensory context.
• Goal-directed behavior is modeled by the posterior intention distribution, obtained via variational free energy minimization.
• The framework successfully explains key interaction properties between habitual and goal-directed behaviors observed in experiments.

Conclusions:

• The proposed framework offers a novel perspective on the neural mechanisms underlying habits and goals.
• It suggests a unified approach to understanding decision-making processes.
• This work provides a foundation for future research into the integration of different behavioral control systems.