Speed/accuracy trade-off between the habitual and the goal-directed processes

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Instrumental responses are driven by two systems: habitual (fast, inflexible) and goal-directed (flexible, slow).
• These systems offer distinct advantages at different learning stages.

Purpose of the Study:

• To propose a normative model for arbitration between habitual and goal-directed learning systems.
• To balance decision-making speed and accuracy using reinforcement learning principles.

Main Methods:

• Developed a computational model based on reinforcement learning theory.
• Assumed distinct characteristics for goal-directed (flexible, slow) and habitual (fast, inflexible) systems.
• Incorporated midbrain dopamine neuron activity (phasic and tonic) into the model.

Main Results:

• The model explains outcome sensitivity changes during learning and sustained sensitivity with equal incentives.
• It accounts for variations in choice reaction time with learning progression and increased choices.
• Predicts distinct roles for phasic dopamine (reinforcement) and tonic dopamine (system competition).

Conclusions:

• The model provides a normative framework for understanding the arbitration between habitual and goal-directed learning.
• It integrates behavioral and neurobiological findings on decision-making and dopamine function.
• Highlights how dopamine modulates competition between learning systems, impacting reaction times.