The distinct functions of working memory and intelligence in model-based and model-free reinforcement learning

Area of Science:

• Cognitive Psychology
• Neuroscience
• Computational Neuroscience

Background:

• Human behavior involves goal-directed planning and automatic choices.
• Reinforcement learning (RL) models describe two strategies: flexible but demanding model-based, and efficient but rigid model-free.
• Understanding how working memory (WM) affects these strategies is crucial.

Purpose of the Study:

• Investigate strategy adjustment in RL under varying WM loads.
• Examine the impact of learning strategies and mental abilities (WM capacity, intelligence) on learning performance.

Main Methods:

• Participants engaged in RL tasks under different WM load conditions.
• Linear regression models were used to analyze the relationship between strategies, mental abilities, and performance.

Main Results:

• Model-based strategy use decreased as WM load increased.
• Model-free strategy use increased with higher WM load.
• Both model-based strategy utilization and intelligence positively predicted learning performance.
• Model-based learning mediated the effect of WM load on performance.

Conclusions:

• Working memory capacity critically influences strategic selection in RL.
• Cognitive abilities like WM and intelligence are key determinants of learning performance in RL tasks.