How Working Memory and Reinforcement Learning Are Intertwined: A Cognitive, Neural, and Computational Perspective

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Artificial Intelligence

Background:

• Reinforcement learning (RL) and working memory (WM) are fundamental cognitive processes.
• They are traditionally viewed as distinct in terms of cognitive function, neural substrates, and algorithmic implementation.

Purpose of the Study:

• To challenge the distinctness of RL and WM by reviewing evidence of overlapping brain networks.
• To explore the synergistic relationship between RL and WM in cognitive and computational research.
• To highlight the importance of their interplay for understanding human behavior and advancing AI.

Main Methods:

• Literature review of cognitive, neuroscientific, and computational studies.
• Analysis of research demonstrating shared neural substrates between RL and WM.
• Synthesis of findings on the interactive roles of RL and WM.

Main Results:

• Significant overlap exists between brain networks supporting RL and WM.
• RL and WM are less distinct cognitive processes than commonly assumed.
• Considering their interplay offers benefits for explaining properties of each process.

Conclusions:

• Future research should integrate RL and WM to build a comprehensive model of cognition.
• Recognizing WM's role in learning is vital for developing more human-like artificial agents.
• Neglecting the dynamics between cognitive processes can lead to incomplete understanding, especially regarding individual differences.