Reinforcement learning at the interface of artificial intelligence and cognitive science.

Area of Science:

• Computational neuroscience
• Artificial intelligence
• Cognitive psychology

Background:

• Reinforcement learning (RL) is a computational framework for sequential decision-making.
• RL is increasingly applied in healthcare for personalized treatments and clinical optimization.
• RL's learning properties make it a powerful tool for modeling human cognition.

Purpose of the Study:

• Introduce RL to neuroscientists, clinicians, and psychologists.
• Bridge artificial intelligence and brain science with accessible terminology and clinical analogies.
• Provide a roadmap for interdisciplinary research integrating computation, neuroscience, and clinical practice.

Main Methods:

• Outline foundational RL concepts and algorithms (e.g., temporal-difference learning, Q-learning).
• Connect RL mechanisms to neurobiological processes (e.g., reward prediction errors, frontostriatal loops).
• Examine RL's integration into cognitive architectures (e.g., ACT-R, SOAR).

Main Results:

• RL mechanisms align with neurobiological processes supporting learning, planning, and habit formation.
• RL effectively models cognitive functions like attention, memory, and decision-making.
• Deep RL shows promise in simulating complex cognitive tasks.

Conclusions:

• RL serves as a unifying framework for understanding cognition and guiding future research.
• Highlights limitations in current methods and suggests new directions, including hybrid models and multi-agent RL.
• Emphasizes adaptive healthcare applications and interdisciplinary integration.