Model-Free RL or Action Sequences?

Area of Science:

• Cognitive Neuroscience
• Computational Psychiatry
• Behavioral Economics

Background:

• Model-free reinforcement learning (MF RL) is a dominant computational framework for decision-making and habit formation.
• Recent challenges propose model-based action sequencing as an alternative explanation for similar behavioral and neural patterns.
• Dissociating these mechanisms is crucial for understanding habitual control.

Purpose of the Study:

• To empirically differentiate between model-free reinforcement learning (MF RL) and model-based action sequencing.
• To provide unconfounded evidence supporting the role of MF RL in human decision-making.
• To investigate the simultaneous use of both MF RL and model-based strategies.

Main Methods:

• Two experiments were designed to dissociate MF RL from model-based selection of action sequences.
• Behavioral and neural data were collected to analyze decision-making patterns.
• Analysis focused on identifying distinct signatures of MF RL and model-based control.

Main Results:

• The study presents empirical evidence that dissociates MF RL from model-based action sequencing.
• Results demonstrate that humans utilize MF RL for habitual control.
• Evidence also shows simultaneous application of model-based selection of action sequences.

Conclusions:

• MF RL plays a significant and distinct role in human decision-making and habit formation.
• Humans employ dual mechanisms for habitual control: MF RL and model-based action sequencing.
• These findings solidify the central position of MF RL in computational models of behavior.